ax-grep 0.1.0 → 0.1.1

package/docs/progress.md

@@ -0,0 +1,1306 @@
+# Current Progress
+
+Status: about 99.2% fit for the goal of making `ax-grep --agent` a useful
+first-pass page/search checker before `agent-browser`.
+
+Last updated: 2026-06-16.
+
+This is a research track, so the target can expand when comparison work finds a
+new class of browser accessibility-tree signal that static HTML can expose
+safely. When that happens, add it to this file instead of treating the earlier
+percentage as a fixed contract.
+
+## Progress Model
+
+| Area | Current estimate | Evidence |
+| --- | ---: | --- |
+| README/docs hygiene | 90% | Root README is short; detailed docs live under `docs/`. |
+| Process safety | 90% | `AGENTS.md`, `pnpm check:processes`, non-browser gates, explicit risky npm script inventory, the single-target real-page smoke gate, four-target `agent-browser` smoke gate, browser comparison lock, and finally-based browser session close checks are in place. |
+| Search result handoff | 92% | Result choices, top choice snippet/host/provenance/official-source shortcuts, normalized recommended/source-search/first-official date shortcuts, recommended, first-official, and source-search source type/hint shortcuts, brief-mode search provenance, source-search selected/alternate/failure path shortcuts, source-search engine-attempt recovery shortcuts, failure retry commands, alternate-choice arrays, source hints, verification, search-decision command shortcuts, decision counts/reasons, first-official result shortcuts, command args, command strings, source scores, relevance, and likely-official flags are exposed. |
+| Page check handoff | 97% | Forms, action targets, FAQ answers, code block text, resource titles and follow-up commands, media/section text, and structured metadata labels/selectors in brief mode, provenance and author follow-up commands, timeline ISO/epoch shortcuts, offer numeric price shortcuts, dataset coverage/creator shortcuts, table navigation shortcuts, brief-mode table path/selector targeting, semantic table/grid read-current fallback, hidden signal group counts/top shortcuts/selectors, author/source metadata, static-readiness reason codes/reasons, fetched-HTML usability in brief mode, control-plane shortcuts, generic top-choice execution metadata, page/search decision command shortcuts, runbook/execution-plan/answer-plan/executor/handoff command shortcuts, barrier-specific browser-capture reason codes, primary execution shortcuts, barriers, read targets, and full focused-content length scoring are exposed in full and brief modes. |
+| Semantic accessibility signals | 96% | Landmarks, headings, links, interactive/focusable controls, keyboard shortcut selectors in brief mode, buttons, fields, values, relations, choices, selected-choice shortcuts with controlled targets, states, modal/live state shortcuts, list item refs, first list item shortcuts, list samples from compacted link items, interactive/focusable/actionable selectors in brief mode, table/list names and container selectors in brief mode, table header/cell navigation shortcuts, selected table/grid cell shortcuts, sample cell spans, table ownership refs, owned and selected-owned sample-cell shortcuts, semantic counts, outline selectors, and brief-mode semantic selectors/states are exposed. |
+| Browser-tree parity research | 99% | Fifteen local browser fixtures compare browser-side semantic output with `--agent-brief`; `readiness:real-page-smoke` proves one fetched page can use static handoff; `readiness:agent-browser-smoke` now covers four real `agent-browser` targets, including catalog, link-heavy listing, and government index/search pages. `readiness:agent-browser-text-heavy-smoke` separately tracks Korean Wikipedia with strict text recall split from structural content recall. |
+
+Overall estimate: 99.2%. This is still conservative because the final goal is
+comparative usefulness across diverse real pages and `agent-browser`, not just
+passing the current smoke tests.
+
+Forecast from the current evidence:
+
+- 85% means the tool is already useful for common static search/page handoff
+  loops.
+- 88-96% reflects landed non-browser shortcuts, fallback-policy work, and the
+  current fifteen-fixture browser parity evidence; the local browser parity
+  subtrack is higher, but broader real-page comparison still gates completion.
+- 98%+ still requires a broader but still sequential `agent-browser` comparison
+  slice with each observed gap classified as `implement`, `browser-only`, or
+  `defer`.
+- The estimate may drop if comparison finds a high-value browser-tree signal
+  that static output does not yet expose.
+
+Current forecast board:
+
+| Phase | Status | What is being measured | Expected estimate movement | Blocker or expansion trigger |
+| --- | --- | --- | --- | --- |
+| F1: Non-browser brief parity cleanup | Complete for current scalar shortcuts | Whether `--agent-brief` still drops high-value shortcuts already present in full compact output. A static guard now checks every declared `semanticTop*` shortcut stays present in both compact and brief agent projections. | Compact-vs-brief scalar losses are closed; the remaining 13 differences are intentionally large nested payloads (`actions`, `signals`, `semanticSummary`, and similar). | Stop adding fields unless a fixture shows that brief output forces unnecessary full output or browser capture. |
+| F2: Browser fallback clarity | Watch | Whether static output explains `use fetched HTML`, `capture browser HTML`, or `interact in browser` without prose parsing. | Page handoff is now 97% after semantic table/grid, static-readiness, browser-reason, execution-stage fallback, and error-runbook shortcuts reduced unnecessary browser handoff. | Any new page whose current reason code does not explain why browser capture or live interaction is needed reopens this packet. |
+| F3: Semantic table/list/control comparison | Started | Whether static table, list, and control shortcuts cover the browser accessibility-tree signals agents actually use. | Semantic accessibility is now 96% after selected, owned, merged-cell, and text-parity table/grid shortcuts reduced sample-ref parsing. | Browser-tree comparison may reveal a new P1 signal, which can hold or lower the estimate until tracked. |
+| F4: Browser parity evidence set | Started, gated | A small sequential fixture set plus single-target real-page smoke and four-target `agent-browser` smoke gates. | Browser parity research remains 99%, and overall stays 99.2% after the local fixture set added direct `aria-owns` virtual rowgroup parity and `ko.wikipedia.org/wiki/대한민국` showed actionable/navigation recall above `0.82` but low strict text recall from reference StaticText/listmarker volume; score now uses structural content recall while still reporting text recall. | Do not run broad browser checks; each run needs pre/post `pnpm check:processes`. |
+| F5: Re-estimation and closure | Planned | Whether all P1 gaps have either landed code, a browser-only note, or a defer reason. | Overall can rise, hold, or fall based on evidence. | Research goals can expand when a useful accessibility-tree signal appears mid-work. |
+
+Near-term queue with prediction:
+
+| Order | Work item | Current coverage | Next concrete step | Done when | Predicted result |
+| ---: | --- | ---: | --- | --- | --- |
+| 1 | Keep progress auditable without bloating README. | 90% | Maintain this progress file as the detailed tracker and keep root README short. | README tests and diff checks pass after docs changes. | Documentation remains stable; no overall increase expected. |
+| 2 | Audit remaining full-vs-brief shortcut losses. | 100% | Scalar losses are closed; every declared `semanticTop*` shortcut is now guarded against disappearing from compact or brief output. | Reopen only when a fixture proves a large omitted payload needs a shallow shortcut. | No further increase expected without new evidence. |
+| 3 | Keep the browser parity fixture set small and local. | 100% | `pnpm compare:browser:fixture` uses local HTML fixtures, one Puppeteer browser, and no external network or `agent-browser` process. | Pre/post process checks are clean and each fixture result is recorded in the gap ledger. | Enabled overall 95%. |
+| 4 | Classify browser-tree gaps. | 100% | The fixture set classified ARIA state normalization, compacted list samples, modal/live shallow shortcuts, selected-choice shortcuts, selected tab panel targets, selected gridcell shortcuts, range/value state parity, busy live status parity, invalid field error-message parity, sorted header state parity, multiselect listbox state parity, drag/drop state parity, disabled/readonly field state parity, mixed checkbox state parity, and field details relation text parity. | Every gap has priority, evidence, decision, validation command, and estimate impact. | Raised browser parity research to 99%. |
+| 5 | Recalculate the headline estimate. | 100% | This file records why the estimate moved from 89% to 90%, 91%, 92%, 93%, 94%, then 95%. | The estimate explains why it moved, held, or dropped. | Overall moves only after evidence, not optimism. |
+
+Reading guide:
+
+- 90% or higher means the area is useful in normal agent handoff flows and only
+  needs maintenance unless new research expands the target.
+- 75-89% means the area works for known fixtures but still has known comparison
+  or fallback gaps.
+- 60-74% means the area has infrastructure and early evidence, but the research
+  set is still too small to call stable.
+- Below 60% means the area should not be used as a completion signal.
+
+The estimate can move down as well as up. If research finds a browser
+accessibility-tree signal that is important for agents and not yet represented
+by static output, record it here, add a candidate task, and adjust the estimate
+instead of hiding the new scope.
+
+## Active Work Tracker
+
+| Track | Progress | Current work | Remaining work | Next checkpoint |
+| --- | ---: | --- | --- | --- |
+| Documentation control | 90% | Keep README short and move long operational detail into `docs/`. | Add only status, safety, and research notes that help future sessions resume quickly. | README length/mojibake tests pass after each docs change. |
+| Process containment | 90% | Keep validation sequential and check for leftover browser/test/comparison processes; `readiness:real-page-smoke` adds a no-browser remote-page gate, `readiness:agent-browser-smoke` adds a four-target browser-backed gate, and readiness audit now rejects unreviewed risky npm scripts. | Add more explicit notes when a task would require browser-backed validation, including why it is necessary. | `pnpm check:processes` before and after risky work shows no leftovers. |
+| Search handoff | 92% | Expose enough ranked-result/source context for an agent to choose, open, skip, or recover from failed opened results in full and brief modes. | Identify whether deeper snippet dedupe, provenance, official-source ambiguity, or failed-open categories remain beyond top choice shortcuts, alternate-choice arrays, source-search quality shortcuts, engine-attempt shortcuts, and failure kind/retryable/path shortcuts. | A static search/source fixture lets an agent choose or recover from a result without browser inspection first. |
+| Page handoff | 97% | Surface barriers, read targets, FAQ answers, code block text, resource titles, media/section text, structured metadata labels/selectors, action targets, author/source metadata, table navigation shortcuts, semantic table/grid read-current fallback, hidden signal group shortcuts, static-readiness reason codes/reasons, fetched-HTML usability, primary execution shortcuts, generic top-choice execution metadata, barrier-specific browser-capture/live-interaction reason shortcuts, and full focused-content length scoring in full and brief modes. | Tighten remaining client-rendered and interaction-required categories when fixture evidence shows ambiguity. | Fixtures show clear `use static output`, `need browser capture`, or `handle live browser interaction` reasons. |
+| Semantic accessibility | 96% | Continue adding high-value shortcuts from roles, states, relations, choices, selected choices and their controlled targets, selected table/grid cells, lists, tables, ownership, owned rowgroups, controls, semantic counts, outline selectors, and brief-mode handoff output. | Compare more table/grid/list/control output against browser-tree expectations and add only useful static equivalents. | Each accepted signal has a public type, CLI output, compact/brief output, fixture coverage, and static gate scoring. |
+| Browser parity research | 99% | Fifteen local browser fixture cases are in place and passed; one real-page static handoff smoke, one four-target `agent-browser` smoke, and one separate text-heavy document smoke are wired. | Expand the gap list as research finds new browser accessibility-tree signals; lower estimates if new important gaps appear. | Each gap is tagged `implement`, `browser-only`, or `defer` with priority and evidence. |
+
+## Remaining Work Breakdown
+
+The percentages below are not time estimates. They are confidence estimates for
+how much of each work packet is already covered by code, tests, and documented
+evidence.
+
+| Packet | Current coverage | Remaining decision | Expected next result | Scope can expand when |
+| --- | ---: | --- | --- | --- |
+| P1: README/docs containment | 90% | Keep root README short while this progress file carries detailed state. | Stable README tests and a readable progress ledger. | A new workflow needs durable instructions. |
+| P2: Search result routing | 92% | Watch whether snippet provenance, official-source ambiguity, or failed-open categories need anything beyond top result/source shortcuts, alternate-choice arrays, decision reasons, source-search quality shortcuts, source-search engine-attempt shortcuts, and failure path shortcuts. | Keep current top choice shortcuts unless fixtures show agents need deeper dedupe/provenance. | Search/source fixtures show agents still need to inspect nested choices before opening or recovering. |
+| P3: Failed source-search recovery | 95% | Watch for missing recovery hints after selected-result HTTP/fetch failures now that selected/alternate command strings, source, source score, relevance, likely-official flags, rate-limit-aware failure kind, retryable flags, and retry-after hints are shallow. | Keep current shortcuts unless a new failure class appears. | A fixture shows unclear recovery guidance even with `sourceSearchFailureKind`, `sourceSearchFailureRetryable`, and `sourceSearchFailureRetryAfter`. |
+| P4: Static readiness and browser fallback | 90% | Watch whether client-rendered, blocked, source-search failure, and interaction-required cases need more detail beyond static-readiness reason codes, semantic table/grid fallback, browser/live-interaction reason codes, execution-stage fallback shortcuts, and error-runbook shortcuts. | Keep current codes unless fixtures show ambiguous guidance. | A page needs browser capture or live browser interaction but current reason text/code does not explain why. |
+| P5: Semantic table/grid parity | 88% | Decide whether deeper virtualized row sampling is worth exposing beyond current header/ownership refs, owned row/cell sampling, first owned sample-cell shortcuts, selected/current cell state, selected-cell shortcuts, and direct local browser `aria-owns` rowgroup parity. | Watch whether agents need more than the first owned sample cell before adding larger payloads. | Browser-tree comparison exposes useful table structure not represented in static output. |
+| P6: General accessibility parity | 96% | Run only small sequential comparisons and record every new signal before implementation. | Add more fixture rows only after the current one-command gate remains clean. | Any browser accessibility-tree signal repeatedly helps agent routing. |
+| P7: Process safety | 90% | Keep every validation step single-process and record browser-backed exceptions before running them; readiness audit now tracks the approved test/comparison/browser script inventory. | Clean process check before final handoff. | A new test path would spawn browsers, servers, or long-running comparison jobs. |
+
+## Active Work Detail
+
+These are the items currently being worked or prepared. Use this table to
+estimate whether the overall percentage should move.
+
+| ID | Item | Progress | Current output | Still needed | Completion signal | Estimate impact |
+| --- | --- | ---: | --- | --- | --- | ---: |
+| A1 | Answer/evidence citation shortcuts | 100% | Top-level citation count and first citation id were implemented, validated, committed, and pushed. | Watch for regressions only. | Typecheck, focused CLI tests, static comparison gate, readiness audit, and process check passed. | Landed. |
+| A2 | Static-vs-browser gap ledger | 70% | Research scope ledger exists, observed gap rows track evidence/decision/status/validation/impact, and every local browser parity check now carries a `ledgerId` tied back to a documented G-row. | Add new rows whenever comparison finds a browser-tree or handoff signal not represented in static output; keep new browser checks linked to a ledger row. | Each new gap has source, priority, decision, status, next command, and any local browser check points to its G-row. | +2-4% to browser parity research. |
+| A3 | Table/grid ownership research | 95% | Header refs include path, role, row/column index, sort state, and selector; cell refs keep row/column/header context; table/grid `aria-owns` targets are exposed as ownership refs, included in row/cell/header sampling, first/selected owned sample cells are exposed as shallow and text shortcuts, and a local browser fixture now compares owned virtual rowgroup parity directly. | Decide whether deeper virtualized row sampling should become separate shortcuts after future fixture evidence. | Either implement missing static shortcuts or mark as `browser-only`/`defer` with evidence. | +3% P5 table/grid confidence; overall remains 99.2%. |
+| A4 | Browser fallback policy | 90% | Browser HTML reason/action/operation/args/capture shortcuts, static-readiness status/reasonCode/reason/readFrom shortcuts, barrier-specific reason codes, execution-stage fallback reason shortcuts, contract markers, scorer coverage, and error-runbook recovery shortcuts exist. | Add more observed client-rendered, blocked, failed-open, and interaction-required categories only if fixture evidence shows ambiguity. | Fixtures show clear `use static`, `need browser capture`, `recover failed open`, or `handle live browser interaction` reasons without manual interpretation. | Watch only; no increase without new fixture evidence. |
+| A5 | Process-safety guardrails | 90% | `AGENTS.md`, progress rules, `pnpm check:processes`, and readiness audit risky-script inventory are in place. | Keep every validation run sequential and document any browser-backed exception before running it. | No leftover browser/test/comparison processes before final handoff. | Prevents regression rather than raising feature %. |
+| A6 | Top choice snippet shortcuts | 100% | Top-level result/source snippet shortcuts were implemented for the first ranked choice, with public type and CLI coverage. | Watch for deeper snippet provenance needs only. | Focused CLI/public type tests, static gates, README test, diff check, and process check pass. | +1% search handoff. |
+| A7 | Top interactive control state shortcuts | 100% | Top-level interactive `pressed`, `expanded`, `haspopup`, and `controls` shortcuts were implemented so agents can branch on the first interactive target without parsing state strings. | Watch for additional control states only after fixture evidence. | Focused CLI/public type tests, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A8 | Top focusable control state shortcuts | 100% | Top-level focusable `disabled`, `pressed`, `expanded`, `haspopup`, and `controls` shortcuts were implemented so keyboard/accessibility navigation can branch without parsing state strings. | Watch for additional focusable states only after fixture evidence. | Focused CLI/public type tests, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A9 | Top list first-item shortcuts | 100% | Top-level first list-item text, role, position, set size, selected/current/expanded state, and selector shortcuts were implemented so agents can route through list/tree/menu evidence without parsing `semanticTopListItemRefs`. | Watch for nested list or tree-level shortcuts only after fixture evidence. | Focused CLI/public type tests, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A10 | Brief semantic selector/state preservation | 100% | `--agent-brief` now preserves top semantic field/description/value/relation/choice/state selectors and core state booleans that were already available in full compact output. | Watch for other full-vs-brief semantic losses only after fixture evidence. | Focused CLI test, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A11 | Brief search provenance preservation | 100% | `--agent-brief` now preserves top result source hints, first sitelink title/URL, and top source-choice reason that full compact output already exposed. | Watch for other full-vs-brief search choice losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% search handoff. |
+| A12 | Brief page metadata preservation | 100% | `--agent-brief` now preserves top author-link name/source and best hidden read-target reason, matching full compact output for source attribution and static hidden-data routing. | Watch for other full-vs-brief page metadata losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A13 | Brief data-table targeting preservation | 100% | `--agent-brief` now preserves top data-table path and selector so table read handoff can target the exact static table without requesting full output. | Watch for other full-vs-brief structured read-target losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A14 | Brief semantic container targeting preservation | 100% | `--agent-brief` now preserves top semantic table and list container selectors so agents can jump to the table/list itself, not only a child header, cell, or list item. | Watch for other full-vs-brief semantic container losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A15 | Brief actionable semantic selector preservation | 100% | `--agent-brief` now preserves top link, in-page link, button, and image selectors so executor loops can act on the first actionable semantic target without requesting full output. | Watch for other full-vs-brief actionable semantic losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A16 | Brief table/list accessible context preservation | 100% | `--agent-brief` now preserves top semantic table/list names plus first sample-cell row/column spans, keeping accessible labels and merged-cell structure available without full output. | Watch for other full-vs-brief table/list context losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A17 | Brief interactive/focusable selector preservation | 100% | `--agent-brief` now preserves top interactive and focusable selectors so keyboard and interaction handoff can jump to the first operable target without requesting full output. | Watch for other full-vs-brief interaction target losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A18 | Brief fetched-HTML usability preservation | 100% | `--agent-brief` now preserves `canUseFetchedHtml` so executor loops can decide to use static output directly instead of escalating to browser capture. | Watch for other full-vs-brief fallback-policy losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A19 | Brief FAQ answer preservation | 100% | `--agent-brief` now preserves the top FAQ answer with the top FAQ question so executor loops can answer from static FAQ evidence without requesting full output. | Watch for other full-vs-brief structured answer losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A20 | Brief code block text preservation | 100% | `--agent-brief` now preserves top code block text so executor loops can use command snippets and examples without requesting full output. | Watch for other full-vs-brief structured text losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A21 | Brief resource title preservation | 100% | `--agent-brief` now preserves the top resource title so executor loops can distinguish feeds, PDFs, and document resources without requesting full output. | Watch for other full-vs-brief resource metadata losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A22 | Brief keyboard shortcut selector preservation | 100% | `--agent-brief` now preserves the top keyboard shortcut selector so accesskey and aria-keyshortcuts targets can be reached without requesting full output. | Watch for other full-vs-brief keyboard accessibility losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| A23 | Brief structured page evidence preservation | 100% | `--agent-brief` now preserves top media text, section text, breadcrumb/pagination/TOC/embed/transcript context, and provenance/offer/dataset/identity/timeline/contact labels, sources, and selectors that full compact output already exposed. | Watch for other full-vs-brief page-check losses only after fixture evidence. | Focused CLI tests, typecheck, static gates, README test, diff check, and process check pass. | +1% page handoff. |
+| A24 | Local browser parity fixture gate | 100% | `pnpm compare:browser:fixture` compares local browser-side semantic trees with `--agent-brief` across heading/link/button/field/image, table header/cell context, form state/relation context, list samples, keyboard shortcuts, expanded popup controls, current links, modal dialogs, live regions, selected choices, selected grid cells, range/value state, busy live status, invalid field error-message state, sorted header state, multiselect listbox state, drag/drop state, disabled/readonly field state, mixed checkbox state, and field details relation text. | Expand only with one fixture at a time and pre/post process checks. | Typecheck, CLI/public type tests, browser fixture gate, static gates, README test, diff check, and process check pass. | +5% browser parity research. |
+| A25 | Browser-side ARIA state extraction bug fix | 100% | Browser extraction no longer crashes on `aria-controls`, `aria-live`, or `aria-valuetext` because `getState` receives the walk context used for normalization. | Watch for other browser-only state crashes during fixture expansion. | Focused extractor test and browser fixture gate pass. | +1% semantic accessibility. |
+| A26 | List sample fallback for compacted link items | 100% | Static agent summaries now use link/button descendants as list samples when list item wrappers compact away, matching the browser fixture's compact tree behavior. | Watch for noisy list samples in broader fixtures. | Focused CLI test and browser fixture gate pass. | +1% semantic accessibility. |
+| A273 | Current link text row parity | 100% | Text-mode agent output now mirrors current-link fields as individual prefix rows, including name, path, URL, path/query, target, relation, type, language, current state, download, and selector. | Keep this as text-only exposure of existing `semanticTopCurrentLink*` fields; add deeper navigation shortcuts only when a fixture shows first/current link ambiguity. | Focused CLI current-link text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves nav/current-page routing for text-only subagents. |
+| A27 | Stateful overlay browser parity fixture | 100% | Added a second local browser fixture covering `aria-expanded`, `aria-haspopup`, `aria-controls`, `aria-current`, `aria-modal`, and `aria-live` against `--agent-brief`. | Add another fixture only after current browser gate remains stable and process checks stay clean. | `pnpm compare:browser:fixture` passes with two local fixtures and clean pre/post process checks. | +5% browser parity research. |
+| A28 | Modal/live state top shortcuts | 100% | `--agent` and `--agent-brief` now expose `semanticTopModalState*` and `semanticTopLiveState*` so agents can see dialog/live-region state without parsing `semanticSummary.stateItems`. | Watch for other high-value non-first state signals before adding more top-level shortcuts. | Typecheck, focused CLI test, public type test, and browser fixture gate pass. | +1% semantic accessibility. |
+| A29 | Combobox active-descendant browser parity fixture | 100% | Added a third local browser fixture covering combobox expanded/popup state, `aria-activedescendant`, listbox options, selected/current option state, and position metadata against `--agent-brief`. | Superseded by later fixtures; add more only after the current five-fixture gate remains stable and process checks stay clean. | `pnpm compare:browser:fixture` passes with three local fixtures and clean pre/post process checks. | +5% browser parity research. |
+| A30 | Selected choice top shortcuts | 100% | `--agent` and `--agent-brief` now expose `semanticTopSelectedChoice*` so agents can see the selected/current option even when the first option is not selected. | Watch for other high-value non-first choice signals before adding more top-level shortcuts. | Typecheck, focused CLI/public type tests, and browser fixture gate pass. | +1% semantic accessibility. |
+| A31 | Tablist selected panel browser parity fixture | 100% | Added a fourth local browser fixture covering selected tabs, `aria-controls`, and resolved `tabpanel` targets against `--agent-brief`. | Superseded by the fifth fixture; add more only after the current five-fixture gate remains stable and process checks stay clean. | `pnpm compare:browser:fixture` passes with four local fixtures and clean pre/post process checks. | +5% browser parity research. |
+| A32 | Selected choice controlled-target shortcuts | 100% | `--agent` and `--agent-brief` now expose selected-choice control target role/name/selector so agents can jump from a selected tab or choice to its controlled panel without parsing relation arrays. | Watch for other high-value selected-choice relation signals before adding more top-level shortcuts. | Typecheck, focused CLI/public type tests, and browser fixture gate pass. | +1% semantic accessibility. |
+| A33 | Grid selected cell browser parity fixture | 100% | Added a fifth local browser fixture covering `grid`, column/row headers, selected `gridcell` text/state, and row/column position against `--agent-brief`. | Superseded by the sixth fixture; add more only after the current six-fixture gate remains stable and process checks stay clean. | `pnpm compare:browser:fixture` passes with five local fixtures and clean pre/post process checks. | +5% browser parity research. |
+| A34 | Selected table/grid cell shortcuts | 100% | `--agent` and `--agent-brief` now expose selected table/grid cell text, row/column index, selected/current state, and selector so agents can route to the active cell without parsing sample-cell arrays. | Watch for virtualized grid row/column gaps before adding more table shortcuts. | Typecheck, focused CLI/public type tests, and browser fixture gate pass. | +1% semantic accessibility. |
+| A35 | Semantic table/grid fallback without browser handoff | 100% | Low-prose pages with semantic table/grid headers or sample cells now use `read-content` from `agent.semanticSummary` instead of browser interaction when no barrier or source-link handoff is present. | Watch for client-rendered pages where semantic evidence is too weak and browser interaction should still win. | Typecheck, focused CLI test, browser fixture gate, and pre/post process checks pass. | +1% page handoff. |
+| A36 | Top source official hint shortcut | 100% | Page source links from known official organizations now carry `isLikelyOfficial`, and `topSourceChoiceLikelyOfficial` is exposed in full, text, compact, brief, and public types. | Watch for false positives before expanding the official-source heuristic beyond the existing source profile. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff; overall 95% -> 96%. |
+| A37 | Brief primary execution shortcut preservation | 100% | `--agent-brief` now preserves shallow primary command args, after-interaction command args, rank/open-result, browser-interaction flag, readability reasons, and best read-target reason that full compact output already exposed. | Leave nested-heavy objects such as full `actions`, `signals`, and `semanticSummary` out of brief unless a fixture shows agents need them shallow. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff; overall remains 96%. |
+| A38 | Brief decision and diagnostic shortcut preservation | 100% | `--agent-brief` now preserves shallow search/page decision reasons, response status/OK/content-type/final-URL-change fields, verification counts, and barrier source/diagnostic code from full compact output. | Leave full decision, signal, and quality-gate arrays out of brief unless a fixture shows shallow summaries are insufficient. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff; overall remains 96%. |
+| A39 | Brief semantic count and outline selector preservation | 100% | `--agent-brief` now preserves semantic description/value/relation/choice/state counts and the top semantic outline selector that full compact output already exposed. | Keep full `semanticSummary` out of brief; add only scalar shortcuts or direct target selectors that reduce browser/full-output handoff. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility; overall remains 96%. |
+| A40 | Brief control-plane scalar preservation | 100% | `--agent-brief` now preserves routing intent, continuation mode, runbook/execution reasons, alternative-action count/reason, diagnostic codes, and diagnostic info count from full compact output. | Keep large nested control-plane objects out of brief; add shallow shortcuts only from fixture evidence. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | Closes current non-browser brief parity cleanup; overall remains 96%. |
+| A41 | Range value-state browser parity fixture | 100% | Added a sixth local browser fixture covering `slider` name, orientation, min/max/current value, value text, value item, state selector, and field value shortcuts against `--agent-brief`. | Superseded by the seventh fixture; add more only after the current seven-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with six local fixtures; pre/post process checks are clean. | Browser parity research 85% -> 88%; overall remains 96%. |
+| A42 | Busy live status browser parity fixture | 100% | Added a seventh local browser fixture covering named `status`, `aria-busy=true`, `aria-live=polite`, state selector, and live-state shortcuts against `--agent-brief`. | Superseded by the eighth fixture; add more only after the current eight-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with seven local fixtures; pre/post process checks are clean. | Browser parity research 88% -> 90%; overall remains 96%. |
+| A43 | Invalid field error-message browser parity fixture | 100% | Added an eighth local browser fixture covering `aria-invalid=spelling`, `aria-errormessage`, resolved error-message text, state selector, and error-message relation against `--agent-brief`. | Superseded by the ninth fixture; add more only after the current nine-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with eight local fixtures; pre/post process checks are clean. | Browser parity research 90% -> 92%; overall remains 96%. |
+| A44 | Sorted header browser parity fixture | 100% | Added a ninth local browser fixture covering `aria-sort=descending`, sorted `columnheader` state, first-header sort shortcut, and state selector against `--agent-brief`. | Superseded by the tenth fixture; add more only after the current ten-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with nine local fixtures; pre/post process checks are clean. | Browser parity research 92% -> 94%; overall remains 96%. |
+| A45 | Multiselect listbox browser parity fixture | 100% | Added a tenth local browser fixture covering `aria-multiselectable=true`, selected option state, position metadata, and selected-choice shortcuts against `--agent-brief`. | Superseded by later fixtures; add more only after the current fourteen-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with ten local fixtures; pre/post process checks are clean. | Browser parity research 94% -> 95%; overall remains 96%. |
+| A46 | Drag/drop state browser parity fixture | 100% | Added an eleventh local browser fixture covering `aria-grabbed=true`, `aria-dropeffect=move`, button name, and state selector against `--agent-brief`. | Superseded by later fixtures; add more only after the current fourteen-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with eleven local fixtures; pre/post process checks are clean. | Browser parity research 95% -> 96%; overall remains 96%. |
+| A47 | Disabled/readonly field state browser parity fixture | 100% | Added a twelfth local browser fixture covering native `disabled` and `readonly` field state, textbox name, field value, state selector, and shallow field/state shortcuts against `--agent-brief`. | Superseded by later fixtures; add more only after the current fourteen-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with twelve local fixtures; pre/post process checks are clean. | Browser parity research 96% -> 97%; overall remains 96%. |
+| A48 | Mixed checkbox state browser parity fixture | 100% | Added a thirteenth local browser fixture covering `role=checkbox`, `aria-checked=mixed`, checkbox name, field checked shortcut, state checked shortcut, and state selector against `--agent-brief`. | Superseded by the fourteenth fixture; add more only after the current fourteen-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with thirteen local fixtures; pre/post process checks are clean. | Browser parity research 97% -> 98%; overall remains 96%. |
+| A49 | Field details relation browser parity fixture | 100% | Added a fourteenth local browser fixture covering `aria-details`, resolved details text, relation target selector, field selector, and shallow field/relation shortcuts against `--agent-brief`. | Add another fixture only after the current fourteen-fixture gate remains stable and process checks stay clean. | Typecheck passes; `pnpm compare:browser:fixture` passes with fourteen local fixtures; pre/post process checks are clean. | Browser parity research 98% -> 99%; overall remains 96%. |
+| A50 | Real-page static handoff smoke gate | 100% | Added `pnpm readiness:real-page-smoke`, which checks `https://example.com` with `--agent-brief`, `canUseFetchedHtml=true`, `needsBrowserHtml=false`, named semantic roles, and no browser continuation request. | Use this as the first real-page guard before any broader or `agent-browser` comparison slice. | Typecheck, real-page smoke, readiness audit, README test, diff check, and process checks pass. | Overall 96% -> 97%; broader `agent-browser` comparison remains. |
+| A51 | Four-target `agent-browser` smoke gate | 100% | `pnpm readiness:agent-browser-smoke` now runs `pnpm compare` for `https://example.com`, `https://books.toscrape.com/`, `https://news.ycombinator.com`, and `https://www.gov.uk/foreign-travel-advice`; it requires `agent-browser` snapshots, exact overlap for the simple page, catalog-page floors, link-heavy listing floors, and government index/search-page floors. | Use this as the browser-backed guard before any broader comparison slice. | Typecheck, process checker test, four-target agent-browser smoke with pre/post process checks, readiness audit, README test, diff check, and process checks pass. | Overall 97% -> 99.2%; broader `agent-browser` comparison remains. |
+| A52 | Source-search quality shortcut preservation | 100% | `--agent` and `--agent-brief` now expose selected/alternate source-search source, command string, source score, relevance, and likely-official flags next to rank/title/url/host/open-result/command args. | Watch for deeper source-search provenance only after a fixture shows these shallow fields are not enough. | Typecheck, focused source-search CLI tests, public type test, readiness audit, static fixture gate, README test, and process check pass. | Search handoff 89% -> 90%; overall remains 99.2%. |
+| A53 | Source-search alternate-choice regression coverage | 100% | Static fixture scoring and focused CLI/public type tests now verify `sourceSearchAlternateChoices` keeps source, command, source score, relevance, and likely-official fields for fallback candidates. | Watch for deeper source-search provenance only after a fixture shows these choice-level fields are not enough. | Typecheck, focused source-search CLI tests, public type test, static fixture gate, README test, diff check, and process check pass. | No estimate increase; strengthens the 90% search handoff claim. |
+| A54 | Source-choice command shortcut preservation | 100% | Top page source-choice output now exposes `topSourceChoiceCommand`, static scoring verifies source-choice command and likely-official shortcuts, and source-search alternate choices keep command strings even when compacted for brief output. | Watch for deeper source-link provenance only after a fixture shows command/source-quality fields are insufficient. | Typecheck, focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page source-link and failed-open recovery handoff. |
+| A55 | Generic top-choice command shortcut preservation | 100% | Generic `topChoiceCommand` and result-specific `topResultChoiceCommand` are now exposed and scored so common result/source executor loops do not need kind-specific nested choices for copy-ready commands. | Watch for deeper top-choice execution metadata only after a fixture shows the common fields are insufficient. | Typecheck, focused agent CLI test, public type test, static fixture gate, README test, diff check, and process check pass. | No estimate increase; reduces nested parsing for generic top-choice routing. |
+| A56 | Generic top-choice quality shortcut preservation | 100% | Generic `topChoiceSourceType`, `topChoiceSourceScore`, `topChoiceRelevance`, and `topChoiceLikelyOfficial` are now exposed and scored so common result/source executor loops can judge quality without kind-specific nested choices. | Watch for deeper top-choice quality metadata only after a fixture shows these common fields are insufficient. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens generic top-choice handoff. |
+| A57 | Generic top-choice context shortcut preservation | 100% | Generic `topChoiceHost` and `topChoiceSnippet` are now exposed and scored so common result/source executor loops can inspect host and preview context without kind-specific nested choices. | Watch for deeper top-choice preview metadata only after a fixture shows these common fields are insufficient. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens generic top-choice handoff. |
+| A58 | Generic top-choice selection shortcut preservation | 100% | Generic `topChoiceOpenResult`, `topChoiceRecommended`, and `topChoicePrimary` are now exposed and scored so common result/source executor loops can see open-result index and selection state without kind-specific nested choices. | Watch for deeper top-choice execution metadata only after a fixture shows these common fields are insufficient. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens generic top-choice execution handoff. |
+| A59 | Top source-choice detail shortcut preservation | 100% | `topSourceChoiceKind`, rank, text, selector, relevance, matched term, and find match are now exposed and scored so page source-link loops can judge and target the first source choice without parsing `sourceChoices[0]`. | Watch for deeper source-link provenance only after a fixture shows these shallow fields are insufficient. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page source-link handoff. |
+| A60 | Source-search preview shortcut preservation | 100% | Selected and first-alternate source-search shortcuts now expose snippet, date text, first matched term, first find match, and first sitelink metadata so failed-open recovery loops can compare candidates without parsing nested source-search result objects. | Watch for deeper source-search provenance only after a fixture shows these shallow fields are insufficient. | Typecheck, focused source-search CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens failed-open recovery handoff. |
+| A61 | Read-target kind shortcut preservation | 100% | `readTargets[*].kind` and `topReadTargetKind` now classify evidence, semantic, structured, hidden-data, interaction, barrier, navigation, code, media, resource, verification, and source-search read targets so executor loops do not parse path strings before choosing how to read current payload data. | Watch for additional read-target classes only after a fixture shows the current taxonomy is ambiguous. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page/source-search read routing. |
+| A62 | Best read-target kind shortcut preservation | 100% | `bestReadTargetKind` is now exposed and scored so best-target routing can classify the selected read source even when it differs from `topReadTarget`. | Watch for additional read-target classes only after a fixture shows the current taxonomy is ambiguous. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens best read-target routing. |
+| A63 | Action read-target kind shortcut preservation | 100% | `runbookReadTargetKind`, `nextReadTargetKind`, `executorReadTargetKind`, and `handoffReadTargetKind` now expose the selected read target category beside each stage-specific `readFrom` shortcut. | Watch for additional stage-specific read routing fields only after a fixture shows kind is not enough. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens stage-specific read-current routing without browser work. |
+| A64 | Action read-target quality shortcut preservation | 100% | Stage-specific read target shortcuts now also expose count, score, primary, and reason for runbook, next, executor, and handoff so agents can judge the selected read source without opening nested read-target objects. | Watch for stage-specific confidence gaps only after a fixture shows count/score/primary/reason are not enough. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens stage-specific read-current quality routing without browser work. |
+| A65 | Plan read-target quality shortcut preservation | 100% | `executionPlanReadTarget*` and `answerPlanReadTarget*` shortcuts now expose kind, count, score, primary, and reason beside plan-level `readFrom` values so executors can trust plan surfaces without nested read-target parsing. | Watch for plan-level routing gaps only after a fixture shows these read-target fields are not enough. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens execution/answer plan read-current routing without browser work. |
+| A66 | Decision read-target quality shortcut preservation | 100% | `pageDecisionReadTarget*` and `staticReadinessReadTarget*` shortcuts now expose kind, count, score, primary, and reason beside decision-level `readFrom` values so page/readiness decisions can be consumed without nested read-target parsing. | Watch for decision-level routing gaps only after a fixture shows these fields are not enough. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page/readiness decision routing without browser work. |
+| A67 | Top ARIA key shortcut preservation | 100% | `semanticTopAriaKeyShortcut*` now points to the first `aria-keyshortcuts` target even when an earlier accesskey item is the generic top keyboard shortcut. | Watch for multi-shortcut ranking only after a fixture shows first `aria-keyshortcuts` target is insufficient. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens keyboard accessibility handoff without browser work. |
+| A68 | Client-rendered fallback reason code | 100% | JavaScript app shells now produce `CLIENT_RENDERED`, `browserHtmlReasonCode: client-rendered`, and `staticReadinessReasonCode: client-rendered` instead of collapsing into generic no-inspectable/browser-required fallback. | Watch for false positives on server-rendered pages that legitimately use root/app containers. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; clarifies browser fallback routing without browser work. |
+| A69 | Source-search failure host shortcut preservation | 100% | `sourceSearchFailureHost` is now exposed in text, full/brief compact output, public types, static gate scoring, and readiness audit so failed-open recovery can compare failed and alternate domains without URL parsing. | Watch for deeper failure provenance only after a fixture shows host plus kind/retryability is insufficient. | Typecheck, focused source-search CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens failed-open recovery handoff. |
+| A70 | Source-search alternate host decision shortcut | 100% | `sourceSearchAlternateDifferentHost` is now exposed in text, full/brief compact output, public types, static gate scoring, and readiness audit so failed-open recovery can branch to a different domain without comparing host strings. | Watch for deeper recovery decisions only after a fixture shows first alternate host comparison is insufficient. | Typecheck, focused source-search CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens failed-open recovery handoff. |
+| A71 | Browser interaction need shortcut | 100% | `needsBrowserInteraction` is now an always-present boolean in text, full/brief compact output, public types, static gate scoring, and readiness audit so agents can distinguish browser HTML capture from live browser interaction without interpreting optional fields. | Watch for deeper browser fallback decisions only after a fixture shows interaction-vs-capture routing is still ambiguous. | Typecheck, focused browser fallback CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; clarifies browser fallback routing. |
+| A72 | Observed semantic table column count shortcut | 100% | `semanticTopTableColumnCount` now exposes the observed column count from header/sample-cell refs, including `columnSpan`, in text, full/brief compact output, public types, static gate scoring, and readiness audit. | Watch for deeper virtualized table sampling only after fixture evidence shows first-table observed column count is insufficient. | Typecheck, focused semantic CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens table/grid handoff. |
+| A73 | Selected table-cell header shortcut preservation | 100% | `semanticTopSelectedTableCellHeaders`, row headers, and column headers now expose the active table/grid cell's accessible header context in text, full/brief compact output, public types, static gate scoring, and readiness audit. | Watch for deeper selected-cell context only after fixture evidence shows header arrays are insufficient. | Typecheck, focused semantic CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens table/grid handoff. |
+| A74 | Selected table-cell span shortcut preservation | 100% | `semanticTopSelectedTableCellRowSpan` and `semanticTopSelectedTableCellColumnSpan` now expose active merged-cell geometry in text, full/brief compact output, public types, static gate scoring, and readiness audit. | Watch for deeper merged-cell context only after fixture evidence shows span fields are insufficient. | Typecheck, focused semantic CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens table/grid handoff. |
+| A75 | First table sample text span parity | 100% | The text `semanticTopTableFirstSampleCell` shortcut now includes `rowSpan` and `columnSpan`, matching the JSON shortcut fields so text handoff does not require parsing `sampleRefs` for merged-cell geometry. | Watch for other text-only parity gaps only after fixture evidence shows a JSON shortcut is missing from text output. | Typecheck, focused text semantic CLI test, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity gap. |
+| A76 | First owned table sample text parity | 100% | The text `semanticTopTableFirstOwnedSampleCell` shortcut now exposes the first `aria-owns`/virtualized sample cell and its `ownedTarget`, matching JSON shortcuts so text handoff does not require scanning `sampleRefs`. | Watch for deeper owned-row sampling only after fixture evidence shows the first owned cell is insufficient. | Typecheck, focused text semantic CLI test, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity gap. |
+| A77 | Selected owned table-cell target shortcut | 100% | `semanticTopSelectedTableCellOwnedTarget` now exposes when the active selected table/grid cell comes from an `aria-owns`/virtualized rowgroup in text, full/brief compact output, public types, static gate scoring, and readiness audit. | Watch for deeper virtualized selected-cell context only after fixture evidence shows `ownedTarget` is insufficient. | Typecheck, focused semantic CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens virtualized table/grid handoff. |
+| A78 | First sample owned-target text parity | 100% | The text `semanticTopTableFirstSampleCell` shortcut now includes `ownedTarget` when the first sample cell comes from an `aria-owns`/virtualized rowgroup, matching the existing JSON shortcut. | Watch for other text-only parity gaps only after fixture evidence shows a JSON shortcut is missing from text output. | Typecheck, focused text semantic CLI test, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity gap. |
+| A79 | Top interactive disabled text parity and scorer coverage | 100% | The text `semanticTopInteractive` shortcut now prints `disabled`, matching the existing JSON/brief scalar, and the static fixture scorer now checks first interactive/focusable `disabled`, `pressed`, `expanded`, `haspopup`, and `controls` shortcuts. | Watch for other text-only scalar gaps only after fixture evidence shows JSON/brief state requires parsing in text output. | Typecheck, focused CLI text test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity gap and strengthens regression detection. |
+| A80 | Top accessibility state text parity | 100% | The text `semanticTopState` shortcut now prints shallow `hidden`, `disabled`, `checked`, `selected`, `current`, `focused`, `required`, `haspopup`, `invalid`, and `readonly` scalars using undefined checks so false values remain visible. | Watch for other text-only scalar gaps only after fixture evidence shows JSON/brief state requires parsing in text output. | Typecheck, focused CLI text test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity gap for browser-tree state handoff. |
+| A81 | Semantic state text-parity readiness evidence | 100% | `readiness:audit` now has a dedicated `semantic-state-text-parity` evidence check requiring the text formatter, focused CLI test, and progress ledger to preserve shallow accessibility state scalars. | Keep this audit narrow; add new evidence rows only when a landed shortcut needs durable completion proof. | Typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens completion-audit evidence. |
+| A82 | False-valued state formatter parity | 100% | Text formatter conditions for top interactive/focusable/button/field `haspopup` and field `invalid` now use undefined checks, matching JSON/brief preservation of false-valued ARIA state scalars. | Watch for other formatter truthiness checks only when the underlying type can carry a meaningful false value. | Typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output parity edge case. |
+| A83 | Generic choice shortcut brief evidence | 100% | A focused `--agent-brief` search test now proves generic `topChoice*` and `recommended*` command shortcuts survive compact executor output, and readiness audit requires that coverage. | Watch for future brief trimming that drops scalar choice routing before executor loops can open the recommended result. | Typecheck, focused CLI brief test, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens search handoff completion evidence. |
+| A84 | Generic form choice brief evidence | 100% | The focused executable-form `--agent-brief` test now proves generic `topChoiceKind=form`, path, rank, method, and selector shortcuts survive compact output alongside nested `formChoices`. | Watch for future brief trimming that would force executor loops to parse nested form arrays before routing to a site-search form. | Focused CLI brief test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page/form handoff completion evidence. |
+| A85 | Generic action-target choice brief evidence | 100% | The focused executable-action-target `--agent-brief` test now proves generic `topChoiceKind=action-target`, path, rank, source, method, and selector shortcuts survive compact output alongside nested `actionTargetChoices`. | Watch for future brief trimming that would force executor loops to parse nested action target arrays before routing to JSON-LD or OpenSearch actions. | Focused CLI brief test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page/action-target handoff completion evidence. |
+| A86 | Generic source choice brief evidence | 100% | The focused official-source `--agent-brief` test now proves generic `topChoiceKind=source`, URL, host, command, command args, source type, and likely-official shortcuts survive compact output alongside `topSourceChoice*`. | Watch for future brief trimming that would force executor loops to parse nested source choices before opening attribution or official-source links. | Focused CLI brief test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page/source handoff completion evidence. |
+| A87 | Generic choice routing readiness evidence | 100% | `readiness:audit` now has a dedicated `brief-generic-choice-routing` evidence check requiring the result, source, form, and action-target `topChoiceKind` brief tests plus the compact top-choice/recommended helper. | Keep this audit as a completion gate for brief executor routing; add new choice kinds here if the routing model expands. | Typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens completion-audit evidence. |
+| A88 | Progress tracker readiness evidence | 100% | `readiness:audit` now has a dedicated `progress-tracker-current` evidence check requiring this progress file to keep estimates, forecast, active work, remaining work, research scope, and recent A80+ evidence. | Keep detailed status here, not in the root README, so long-running research remains resumable without recreating README bloat. | Typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens progress visibility and completion-audit evidence. |
+| A89 | Recommended command public type evidence | 100% | Public type tests now cover `recommendedCommand` alongside `recommendedCommandArgs`, proving TypeScript consumers can read either executable command form from search/page handoff summaries. | Watch for public API changes that expose runtime command strings without matching type-level evidence. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A90 | Recommended quality public type evidence | 100% | Public `recommendedRelevance` now matches the runtime relevance label type, and public type tests cover `recommendedRelevance`, `recommendedLikelyOfficial`, and `recommendedSelectionReason` so TypeScript consumers can branch on recommendation quality and rationale without nested result parsing. | Watch for new shallow recommendation scalars that need matching public type evidence. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A91 | Recommended result envelope type evidence | 100% | Public `AgentJsonEnvelope.recommendedResult` now uses the compact result-choice shape instead of `Record<string, unknown>`, so TypeScript consumers can safely read `path`, `rank`, `url`, `command`, `commandArgs`, and quality fields from the direct recommendation payload. | Watch for other envelope-level `Record` fields that already have stable compact public shapes. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A92 | Search envelope type evidence | 100% | Public `AgentJsonEnvelope.sourceSearch` and `searchResults` now use the compact source-search and result-choice shapes instead of generic records, so TypeScript consumers can follow selected source searches and ranked result commands without manual casting. | Watch remaining envelope records only when a stable compact public type already exists. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A93 | Suggested action envelope type evidence | 100% | Public `AgentJsonEnvelope.suggestedActions` now uses `AgentAction[]` instead of generic records, so TypeScript consumers can execute action commands and inspect action targets without manual casting. | Watch future action summary fields for public type parity when compact actions gain new routing metadata. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A94 | Page and verification envelope type evidence | 100% | Public `AgentJsonEnvelope.page`, `verification`, and `finds` now use compact page metadata, verification, and find-summary shapes instead of generic records, so TypeScript consumers can read metadata and verification evidence without manual casting. | Leave `pageCheck` generic until its larger compact structure has a dedicated public shape and fixture coverage. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API completion evidence. |
+| A95 | PageCheck envelope core type evidence | 100% | Public `AgentJsonEnvelope.pageCheck` now exposes typed core handoff fields for content evidence, readability, source/primary links, page actions, recommended actions, next steps, and metadata while leaving large structured arrays intentionally loose. | Tighten individual pageCheck arrays only when fixture coverage proves a stable compact public shape for that group. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page-check handoff API evidence. |
+| A96 | PageCheck action-target type evidence | 100% | Public `AgentPageCheck.actionTargets` now uses `AgentActionTargetChoice[]`, so TypeScript consumers can follow page-discovered search/OpenSearch/read/download targets without casting from generic records. | Keep `forms` loose until compact form output has a stable public type that covers both full and slim modes. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page action routing API evidence. |
+| A97 | PageCheck form type evidence | 100% | Public `AgentPageCheck.forms` now uses `AgentPageForm[]`, a compact-safe form shape that covers both full and slim form output while preserving query fields, URL templates, hidden fields, and input metadata for subagent routing. | Tighten form fields further only if runtime compacting starts preserving a narrower invariant across full and slim modes. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page form routing API evidence. |
+| A98 | PageCheck table and barrier type evidence | 100% | Public `AgentPageCheck.dataTables` and `barriers` now use compact-safe table and barrier shapes, so TypeScript consumers can inspect structured table samples and blocking signals without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens structured page-check API evidence. |
+| A99 | PageCheck resource and media type evidence | 100% | Public `AgentPageCheck.codeBlocks`, `citations`, `media`, and `resources` now use compact-safe shapes, so TypeScript consumers can inspect command-like code, citation URLs, images, and downloadable resources without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens inspect/open-target API evidence. |
+| A100 | PageCheck navigation type evidence | 100% | Public `AgentPageCheck.breadcrumbs`, `sections`, `pagination`, `toc`, and `authorLinks` now use compact-safe navigation shapes, so TypeScript consumers can follow page structure, next-page links, table-of-contents entries, and author URLs without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens navigation and structure API evidence. |
+| A101 | PageCheck structured-data type evidence | 100% | Public `AgentPageCheck.contactPoints`, `offers`, `identities`, `datasets`, `timeline`, and `faqs` now use compact-safe structured-data shapes, so TypeScript consumers can inspect contact, product, identity, dataset, date, and FAQ facts without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens structured fact API evidence. |
+| A102 | PageCheck technical-signal type evidence | 100% | Public `AgentPageCheck.hydration`, `apiEndpoints`, `clientState`, `runtime`, and `config` now use compact-safe technical signal shapes, so TypeScript consumers can inspect hidden app data, API endpoints, browser storage usage, workers, and configuration keys without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens hidden technical signal API evidence. |
+| A103 | PageCheck fact/provenance type evidence | 100% | Public `AgentPageCheck.topics`, `keyValues`, `metaFacts`, `provenance`, `httpPolicies`, and `schemaFacts` now use compact-safe fact shapes, so TypeScript consumers can inspect extracted topics, page facts, provenance IDs, HTTP policies, and JSON-LD facts without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens fact extraction API evidence. |
+| A104 | PageCheck app and embed type evidence | 100% | Public `AgentPageCheck.appHints`, `mobileHints`, `embeds`, and `transcripts` now use compact-safe app/media embed shapes, so TypeScript consumers can inspect manifests, mobile app hints, embedded players, and transcript/caption URLs without casting from generic records. | Keep optional text/selector semantics because aggressive compacting may omit verbose text or selectors. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens app/media inspection API evidence. |
+| A105 | Agent decision public type evidence | 100% | Public `AgentSummary.searchDecision` and `pageDecision` now use explicit decision shapes instead of generic records, so TypeScript consumers can branch on search/page routing decisions, confidence, counts, commands, and recommendation metadata without casting. | Keep `semanticSummary` separate because its nested accessibility-tree summary is larger and needs a dedicated typed pass. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens executor routing API evidence. |
+| A106 | Semantic summary public type evidence | 100% | Public `AgentSummary.semanticSummary` now uses an explicit accessibility-summary shape instead of a generic record, so TypeScript consumers can inspect outlines, keyboard shortcuts, links, controls, tables, fields, relations, choices, and raw ARIA state without casting. | Keep future semantic additions covered by the public type test when new nested shortcut groups are added. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens accessibility-tree handoff API evidence. |
+| A107 | Read value public type evidence | 100% | Public `AgentReadValue` now uses explicit inline payload and compact reference unions instead of `unknown`, so TypeScript consumers can inspect read handoff values or follow large-value references without manual casting. | Keep expanding the payload union when runtime read paths add new stable compact shapes. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes the last known public `unknown` read-value field. |
+| A108 | Read value shortcut preservation | 100% | Agent summaries now expose executor/handoff read-value path, type, count, and reference-path shortcuts in full, brief, and text output, so subagents can decide whether a handoff is inline or referenced without parsing nested `readValue`. | Keep shortcuts shallow; do not inline large read payloads into brief output. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens read-current handoff ergonomics without browser work. |
+| A109 | Primary browser fallback reason shortcut | 100% | Primary action shortcuts now preserve browser HTML reason and reason-code fields in full, brief, and text output, so subagents can explain capture/interact-browser decisions without joining top-level fallback metadata to nested primary action state. | Keep reason duplication shallow and limited to primary action routing; do not add verbose browser payloads to brief output. | Focused CLI tests, full CLI test, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens P4 fallback clarity without running browser-backed checks. |
+| A110 | Action fallback reason shortcut parity | 100% | Top and alternative action shortcuts now preserve browser fallback reason and reason-code fields when an action captures browser HTML or requires live browser interaction, so agents can branch on action-level fallback causes without reading nested actions or diagnostics. | Keep action fallback fields absent for ordinary read/open actions to avoid noisy brief output. | Focused CLI tests, full CLI test, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes an action-shortcut parity gap in P4. |
+| A111 | Executor/handoff browser fallback reason shortcuts | 100% | Executor and handoff shortcuts now preserve browser fallback reason and reason-code fields when their step carries browser HTML capture or live browser inspection metadata, so subagents can explain the exact execution-stage browser handoff without reading nested `browserHtml`, actions, or diagnostics. | Keep stage-level reason duplication shallow and emit it only when executor/handoff browser metadata exists. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes an execution-stage fallback clarity gap in P4. |
+| A112 | Browser fallback contract feature markers | 100% | The public agent contract now declares browser fallback shortcut and reason-code capability markers, including executor and handoff browser shortcut groups, so typed consumers can detect support without relying on incidental field presence. | Keep compact/brief contract output small; do not reintroduce the full feature list into brief payloads. | Public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens feature discovery for P4 fallback fields. |
+| A113 | Browser fallback shortcut scorer coverage | 100% | Static comparison scoring now checks executor and handoff browser fallback reason/code shortcuts against the top-level browser fallback reason, so future regressions cannot silently drop execution-stage fallback clarity while browser HTML capture still scores green. | Keep the scorer focused on shallow reason parity; avoid requiring large nested browser payloads in brief output. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens regression detection for P4 fallback fields. |
+| A114 | Read-value shortcut scorer coverage | 100% | Static comparison scoring now checks executor and handoff read-value path, type, count, and reference-path shortcuts against nested read-value handoff data, so future regressions cannot silently force subagents back into nested `readValue` parsing. | Keep the scorer aligned with compact read-value references; do not require large inline payloads in brief output. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens regression detection for read-current handoff ergonomics. |
+| A115 | Runbook read-value shortcut preservation | 100% | Runbook shortcuts now expose read-value path, type, count, and reference-path metadata in text, full, brief, public types, and static scoring, matching executor/handoff read-current handoff ergonomics without requiring nested `runbook.readValue` parsing. | Keep the runbook shortcut shallow and aligned with compact read-value references. | Focused CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a stage-parity gap in read-current handoff. |
+| A116 | Error runbook shortcut parity | 100% | Error/source-search recovery payloads now preserve shallow runbook decision, mode, operation, status, expected outcome, command args, URL, and read-value shortcuts, so failed-open recovery agents do not need nested `runbook` parsing before choosing browser capture or retry handling. | Keep error runbook shortcuts aligned with normal agent summaries and avoid adding verbose nested payloads to brief output. | Focused source-search failure CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes an error-path stage-parity gap. |
+| A117 | Progress tracker stale-plan cleanup | 100% | The forecast, remaining-work, active-work, and research-ledger rows now distinguish landed fallback/runbook/read-value shortcut parity from the still-open watch items for newly observed browser-only or interaction-required pages. | Keep future progress updates concise and move detail here instead of the root README. | Typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves auditability for the long-running research. |
+| A118 | Source-search failure path shortcut | 100% | Error/source-search recovery payloads now expose `sourceSearchFailurePath` in text, full, brief, public types, and static scoring so failed-open recovery agents can identify the failed payload path without joining against `sourceSearchSelectedPath` or nested `sourceSearch.selectedResult`. | Watch for deeper failure provenance only after a fixture shows path, URL, host, kind, retryability, and retry-after are insufficient. | Focused source-search failure CLI tests, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | Search handoff 90% -> 91%; improves failed-open recovery without browser work. |
+| A119 | Page-decision command shortcut | 100% | Page-check decisions now expose `pageDecisionCommand` in text, full, brief, public types, and static scoring, so executor loops can run an open-source/site-search/browser-retry page decision without parsing nested `pageDecision.command` or joining through primary action. | Keep page-decision command duplication shallow and only emit it when the decision itself carries a command. | Focused page-decision CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces page-check execution handoff parsing. |
+| A120 | Search-decision command shortcut | 100% | Search decisions now expose `searchDecisionCommand` in text, full, brief, public types, and static scoring, so result-selection and search-refinement loops can execute the decision without parsing nested `searchDecision.command` or joining through primary action. | Keep search-decision command duplication shallow and only emit it when the decision itself carries a command. | Focused search-decision CLI tests, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces search execution handoff parsing. |
+| A121 | Runbook command shortcut | 100% | Runbook summaries now expose `runbookCommand` in text, full, brief, public types, and static scoring, so top-level runbook consumers can execute the selected operation without parsing nested `runbook.command` or rebuilding from command args. | Keep runbook command duplication shallow and only emit it when the runbook itself carries a command. | Focused runbook CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces runbook execution handoff parsing. |
+| A122 | Execution-plan command shortcut | 100% | Execution-plan summaries now expose `executionPlanCommand` in text, full, brief, public types, and static scoring, so plan-level consumers can execute the chosen command without parsing nested `executionPlan.command` or rebuilding from command args. | Keep execution-plan command duplication shallow and only emit it when the plan itself carries a command. | Focused execution-plan CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces execution-plan handoff parsing. |
+| A123 | Executor command shortcut | 100% | Executor summaries now expose `executorCommand` in text, full, brief, public types, and static scoring, so executor-step consumers can run the selected command without parsing nested `executor.command` or rebuilding from command args. | Keep executor command duplication shallow and only emit it when the executor itself carries a command. | Focused executor CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces executor-step handoff parsing. |
+| A124 | Handoff command shortcut | 100% | Handoff summaries now expose `handoffCommand` in full, brief, public types, and static scoring, so final handoff consumers can run the selected command without parsing nested `handoff.command` or rebuilding from command args. | Keep handoff command duplication shallow and only emit it when the handoff itself carries a command. | Focused handoff CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces final handoff execution parsing. |
+| A125 | Answer-plan command shortcut | 100% | Answer-plan summaries now expose `answerPlanCommand` in text, full, brief, public types, and static scoring, so answer-first consumers can run the planned command without parsing nested `answerPlan.command` or rebuilding from command args. | Keep answer-plan command duplication shallow and only emit it when the answer plan itself carries a command. | Focused answer-plan CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces answer-stage execution parsing. |
+| A126 | Handoff after-interaction shortcut | 100% | Handoff summaries now expose `handoffAfterInteractionCommand` and args in text, full, brief, public types, and static scoring, so browser-capture handoff consumers can rerun after interaction without parsing nested `handoff.afterInteractionCommand`. | Keep after-interaction duplication shallow and only emit it when the handoff itself carries browser-followup command data. | Focused handoff CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces browser-capture handoff parsing. |
+| A127 | Executor followup command preservation | 100% | Executor steps now preserve `command`, `afterInteractionCommand`, and `executorAfterInteractionCommand*` shortcuts in text, full, brief, public types, and static scoring, so executor-stage consumers can run or rerun without falling back to `next` or nested followup fields. | Keep executor followup duplication shallow and tied to `next` command data. | Focused executor CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes an executor-stage preservation gap. |
+| A128 | Recommended command scorer coverage | 100% | Static comparison scoring and focused brief CLI coverage now check `recommendedCommand` beside `recommendedCommandArgs`, so recommended-result command strings cannot regress while args-only handoff still passes. | Keep recommended command scoring aligned with the compact recommended result payload. | Focused recommendation CLI test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens search recommendation regression coverage. |
+| A129 | Browser fallback command text coverage | 100% | Text-mode browser fallback tests now require copy-ready `browserHtmlCommand`, `executorCommand`, and `handoffCommand` lines beside command args, so human and text-only agents do not lose command strings while JSON shortcuts stay green. | Keep this as text evidence only; runtime/scorer coverage already guards the JSON shortcut surface. | Focused browser-fallback CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens text-mode fallback regression coverage. |
+| A130 | Browser fallback command readiness evidence | 100% | `readiness:audit` now has a dedicated `browser-fallback-command-text` evidence check requiring the focused text-mode fallback test to keep copy-ready browser, executor, and handoff commands beside their args. | Keep this audit narrow; expand it only if another fallback stage gains a separate command string. | Readiness audit, focused browser-fallback CLI text test, typecheck, README test, diff check, and process check pass. | No estimate increase; turns A129 into durable completion-audit evidence. |
+| A131 | Brief nested fallback command preservation | 100% | `--agent-brief` compact handoff and browser-HTML fallback objects now preserve `command` beside `commandArgs`, so subagents reading nested executor/handoff payloads do not need to rebuild shell commands. | Keep command duplication shallow and limited to existing fallback/execution objects. | Focused HTTP recovery CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a brief nested fallback ergonomics gap. |
+| A132 | Source-search date shortcut parity | 100% | Source-search selected and first alternate shortcuts now preserve normalized ISO and epoch date fields in full, brief, and text output, so failed-open recovery can compare freshness without parsing nested search result snippets. | Watch for deeper alternate-list ranking needs only if a fixture shows the first alternate is not enough. | Focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a search recovery shortcut parity gap. |
+| A133 | First-official date shortcut parity | 100% | Search-decision first-official shortcuts now preserve normalized ISO and epoch date fields in full, brief, and text output, so agents can compare official-result freshness against the recommendation without parsing nested search results. | Watch for first-official provenance needs only if official-source ambiguity remains after source score, relevance, command, and date shortcuts. | Focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a search-decision shortcut parity gap. |
+| A132 | Nested browser fallback reason preservation | 100% | Browser HTML capture objects now preserve a shallow `reason` in full and brief output, so nested executor/handoff consumers can explain why capture is needed without joining top-level fallback fields. | Keep exact reason codes on the existing top-level browser fallback shortcuts. | Focused HTTP recovery CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves fallback explainability for nested-only consumers. |
+| A133 | Nested browser fallback public type evidence | 100% | Public type tests now compile `AgentHandoff.browserHtml.reason` and copy-ready `command`, so TypeScript consumers can read nested browser fallback explanation and execution data without casts. | Keep public evidence aligned with stable nested capture fields only. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens API evidence for A131-A132. |
+| A134 | Nested browser fallback scorer coverage | 100% | Static comparison scoring now compares nested browser-HTML fallback `reason`, `command`, and `commandArgs`, so compact executor/handoff regressions cannot pass by preserving only file and capture-script fields. | Keep scorer checks shallow and aligned with `sameAgentBrowserHtml`. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens regression detection for A131-A133. |
+| A135 | Browser fallback score documentation | 100% | `docs/comparison-baseline.md` now explains that `averageAgentBrowserHtmlScore` covers capture file/script plus nested fallback reason, command, and command args across next/plan/executor/handoff payloads. | Keep score descriptions concise and tied to enforced gates. | Readiness audit, README test, diff check, and process check pass. | No estimate increase; improves auditability for browser fallback scorer results. |
+| A136 | Brief executor command scorer coverage | 100% | `averageAgentBriefExecutorStepScore` now compares brief executor/handoff `command` and `afterInteractionCommand` strings beside args, so compact command regressions cannot pass with args-only parity. | Keep command scoring limited to copy-ready executor/handoff strings. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens brief executor command regression detection. |
+| A137 | Page-link command scorer coverage | 100% | `averagePageLinkCommandScore` now requires compact primary/source page links to include copy-ready `command` plus `commandArgs`, not just a URL or args, matching the documented page-link execution contract. | Keep page-link command scoring focused on executable links only. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a scorer/documentation mismatch for page-link handoff. |
+| A138 | Page-link command public type evidence | 100% | Public type tests now compile `pageCheck.sourceLinks[0].command` beside `commandArgs`, proving TypeScript consumers can execute compact page source links without reconstructing shell commands. | Keep page-link public evidence aligned with compact source-link execution fields. | Focused public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens public API evidence for A137. |
+| A139 | Source-choice command scorer coverage | 100% | Static comparison scoring now requires `agent.sourceChoices` and handoff `sourceChoices` to preserve copy-ready `command` strings beside `commandArgs`, so source-link recovery cannot pass with args-only routing. | Keep source-choice scoring aligned with compact source-link execution fields. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens source-link handoff regression detection. |
+| A140 | Result-choice command scorer coverage | 100% | Static comparison scoring now requires `agent.resultChoices` and compact source-search alternate choices to preserve copy-ready `command` strings beside `commandArgs`, so search/open-result recovery cannot pass with args-only routing. | Keep result-choice scoring aligned with compact search execution fields. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens search result handoff regression detection. |
+| A141 | Search-decision source profile shortcuts | 100% | Recommended and first-official search-decision shortcuts now preserve source type and source hints in full, brief, and text output, so agents can compare official/source-profile rationale without opening nested result objects. | Watch for deeper official-source ambiguity only after source type, source score, source hints, relevance, dates, and command shortcuts are insufficient. | Focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a search-decision provenance shortcut gap. |
+| A141 | Brief handoff choice command scorer coverage | 100% | Brief executor scoring now requires handoff `resultChoices` and matching `sourceChoices` to carry executable `command` plus `commandArgs`, closing an args-only gap in compact handoff choice routing. | Keep the shared `hasExecutableCommand` helper limited to command-string presence checks. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens compact handoff regression detection. |
+| A142 | Compact handoff choice command preservation | 100% | Compact handoff result choices now use the command-preserving list helper, and compact source-choice refs keep `command` beside `commandArgs`, so stricter A141 scoring is backed by runtime output instead of only scorer intent. | Keep compact choice refs shallow while preserving executable command strings. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes the runtime gap exposed by A141. |
+| A143 | Form/action target command preservation | 100% | Site-search form choices and structured action-target choices now expose copy-ready `command` plus `commandArgs` in nested choices and top shortcuts when a concrete URL can be derived, so subagents can execute page-local search handoffs without rebuilding URLs from templates. | Keep commands derived from existing URL/template data; do not invent a query when no `--find` text exists. | Focused CLI/public type tests, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes an interaction handoff command gap. |
+| A144 | Text form/action command preservation | 100% | Text output now prints copy-ready form and action-target commands in both top shortcuts and nested choice lines, so text-only subagents can continue page-local searches without switching to JSON parsing. | Keep text output concise and only print commands already present on the choice payload. | Focused text CLI tests, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; extends A143 to text-only handoff. |
+| A145 | Nested browser fallback text commands | 100% | Text output now prints nested executor/handoff browser-HTML reason and command strings beside command args, so text-only agents reading the browser fallback block do not need to join top-level shortcuts before rerunning capture. | Keep nested text aligned with the existing browserHtml payload reason and command fields. | Focused browser fallback text test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens text-mode browser fallback explainability. |
+| A146 | Second table sample-cell shortcut | 100% | Runtime, compact, brief, text, public type, readiness, and static scoring paths now expose the second representative table/grid cell so agents can avoid parsing `semanticTopTableSampleCellRefs[1]`. | Keep this shortcut shallow and limited to the second emitted sample cell; do not expand arbitrary cell indexes until fixture evidence proves that is needed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a P5 table ergonomics gap. |
+| A147 | Second table header shortcut | 100% | Runtime, compact, brief, text, public type, readiness, and static scoring paths now expose the second representative table/grid header so agents can avoid parsing `semanticTopTableHeaderRefs[1]`. | Keep this shortcut shallow and limited to the second emitted header; do not expand arbitrary header indexes until fixture evidence proves that is needed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes another P5 table ergonomics gap. |
+| A148 | First owned table target text line | 100% | Text output now prints `semanticTopTableFirstOwned` from existing JSON shortcuts so text-only agents can target an `aria-owns`/virtualized rowgroup without scanning the long table summary. | Keep this text-only line aligned with existing `semanticTopTableFirstOwned*` JSON fields; do not add new JSON fields without fixture evidence. | Focused CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output ergonomics gap. |
+| A149 | Original hidden URL item paths | 100% | Original `pageCheck.hydration`, `apiEndpoints`, `runtime`, `appHints`, `mobileHints`, and `metaFacts` URL-bearing entries now expose `urlPath`/`urlQuery` in JSON and text output, and static comparison scoring includes those arrays. | Watch for non-HTTP/custom-scheme examples where path parsing is not useful enough and should be classified instead of expanded blindly. | Focused CLI tests, static fixture gate, and typecheck pass; readiness/README/diff/process checks remain before final handoff. | No estimate increase; reduces original read-target URL parsing. |
+| A149 | Table ref selector text parity | 100% | Text `semanticTopTable` aggregate refs now include header selectors plus sample-cell selectors and owned targets, so text-only agents can target later headers/cells without switching to JSON or relying only on first/second shortcuts. | Keep this limited to existing ref metadata; avoid adding larger nested table payloads to text output. | Focused CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode table targeting. |
+| A150 | Source-search source profile shortcuts | 100% | Source-search selected and first alternate shortcuts now preserve source type and source hints in full, brief, and text output, so failed-open recovery agents can compare source-profile rationale without opening nested result objects. | Watch for deeper alternate-list ranking needs only if first alternate profile shortcuts are insufficient. | Focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a source-search provenance shortcut gap. |
+| A150 | Source-search choice preview text parity | 100% | Source-search selected/alternate choice text lines now include host, snippet, and first sitelink preview metadata already present on the compact choice object, so text-only failed-open recovery can compare candidates without nested JSON. | Keep this limited to existing compact result metadata; do not add new search result fields without fixture evidence. | Focused CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode failed-open recovery. |
+| A151 | Source-search choice date text parity | 100% | Source-search selected/alternate choice text lines now include `dateText` already present on compact result objects, so text-only recovery can compare freshness without relying only on top-level first-alternate shortcuts. | Keep this limited to existing compact result metadata; avoid date inference changes without fixture evidence. | Focused CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode source candidate comparison. |
+| A152 | Browser fixture evidence summary guard | 100% | Browser fixture comparison summaries now include every `semanticTop*` shortcut used by check pass conditions, and a static CLI test fails if future semantic checks are not visible in the failure report. | Keep this as comparison observability only; do not add large nested summaries unless a failing fixture needs them for diagnosis. | Focused CLI static test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; makes browser parity failures faster to classify without extra browser runs. |
+| A152 | Source-search choice date detail text parity | 100% | Source-search selected/alternate choice text lines now include `datePrecision` and `dateSource` already present on compact result objects, so text-only recovery can judge freshness confidence and source without nested JSON. | Keep this limited to existing compact result metadata; avoid date inference changes without fixture evidence. | Focused CLI test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode source candidate comparison. |
+| A153 | Public original URL item path types | 100% | Public `AgentPage*` read-target types now include the `urlPath`/`urlQuery` fields already emitted by original URL-bearing `pageCheck` arrays, including nested breadcrumb/TOC items and dataset/identity URL-list helpers. | Keep public type fields aligned with runtime output when new read-target URL helpers are added. | Public type test and typecheck pass; readiness/README/diff/process checks remain before final handoff. | No estimate increase; removes TypeScript casts for original read-target URL routing. |
+| A154 | Embed poster/source URL item paths | 100% | `pageCheck.embeds` now exposes `posterUrlPath`/`posterUrlQuery` and `sourceUrlPaths`/`sourceUrlQueries` beside video/audio embed URLs, with text output, public types, and static scoring aligned. | Keep nested media URL helpers limited to existing embed metadata unless a fixture shows agents need deeper source attributes. | Focused CLI test, public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces media-source routing URL parsing. |
+| A153 | Result-choice date detail shortcut parity | 100% | Top result-choice shortcuts and result-choice text lines now include existing `datePrecision` and `dateSource` metadata, so text and compact search handoffs can judge freshness confidence without reading nested result objects. | Keep this limited to existing search-result metadata; avoid date inference changes without fixture evidence. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode and compact search candidate comparison. |
+| A154 | Generic top-choice date detail shortcut parity | 100% | Generic `topChoice*` shortcuts now include existing result-choice `dateText`, `datePrecision`, and `dateSource` metadata when the top choice is a search result, so generic executor loops can judge freshness without switching to result-specific fields. | Keep this limited to copied result-choice metadata; source/form/action-target choices stay unchanged unless they gain date evidence. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens generic top-choice search routing. |
+| A155 | Recommended-result date detail shortcut parity | 100% | `recommended*` shortcuts now include existing recommended-result `dateText`, `datePrecision`, and `dateSource` metadata, so agents can judge freshness from the recommended route without parsing `recommendedResult`. | Keep this limited to copied recommended-result metadata; avoid changing search result date inference without fixture evidence. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens recommended-result search routing. |
+| A156 | Search-decision date detail shortcut parity | 100% | `searchDecision` and `searchDecisionRecommended*` shortcuts now include existing recommended-result `dateText`, `datePrecision`, and `dateSource` metadata, so decision-only agents can judge freshness without joining against `recommendedResult`. | Keep this limited to copied search-decision metadata; avoid changing date inference without fixture evidence. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens decision-only search routing. |
+| A157 | Primary-target date detail shortcut parity | 100% | `primaryTarget*` shortcuts now include existing action-target `dateText`, `datePrecision`, and `dateSource` metadata, so agents can judge freshness of the exact primary execution target without parsing `primaryAction.target`. | Keep this limited to copied target metadata; avoid broad action-target expansion until fixture evidence shows it is needed. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens primary execution routing. |
+| A158 | Top-action target date detail shortcut parity | 100% | `topActionTarget*` shortcuts now include existing action-target `dateText`, `datePrecision`, and `dateSource` metadata, so agents can judge freshness of the first action candidate without parsing `actions[0].target`. | Keep this limited to copied target metadata; broaden to executor/handoff targets only if fixture evidence shows agents need it. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens action-candidate routing. |
+| A159 | Executor-target date detail shortcut parity | 100% | `executorTarget*` shortcuts now include existing executor target `dateText`, `datePrecision`, and `dateSource` metadata, so agents can judge freshness of the exact executor step without parsing `executor.target`. | Keep this limited to copied target metadata; broaden target shortcut groups only when fixture evidence shows agents need them. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens executor-step routing. |
+| A160 | Handoff-target date detail shortcut parity | 100% | `handoffTarget*` shortcuts now include existing handoff target `dateText`, `datePrecision`, and `dateSource` metadata, so receiving agents can judge freshness without parsing `handoff.target`. | Keep this limited to copied target metadata; avoid broader action-target expansion without fixture evidence. | Focused CLI tests, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens handoff-step routing. |
+| A161 | Alternative-action target date detail shortcut parity | 100% | `alternativeActionTarget*` shortcuts now include existing alternate action target `dateText`, `datePrecision`, and `dateSource` metadata, so agents can compare fallback freshness without parsing `alternativeActions[0].target`. | Keep this limited to copied target metadata; avoid broader action-target expansion without fixture evidence. | Focused CLI alignment test, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens fallback action routing. |
+| A162 | Action-target choice text shortcut parity | 100% | Text output now prints the existing `topActionTargetChoice*` kind/name/source/target URL/method/state/selector shortcuts, so agents using human-readable summaries can route JSON-LD and OpenSearch actions without falling back to nested choices. | Watch only for text-output fields that already exist as shallow JSON/brief shortcuts and affect action routing. | Focused CLI text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output routing parity gap. |
+| A163 | Form choice text shortcut parity | 100% | Text output now prints existing `topFormChoice*` method/action/form metadata, field counts, and selector shortcuts, so text-only agents can route site-search and submit forms without relying on nested form choice lines. | Keep this limited to existing shallow form choice metadata; add deeper field shortcuts only when a fixture proves routing needs them. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-output form routing parity gap. |
+| A164 | Generic top-choice command-args text parity | 100% | Text output now prints `topChoiceCommandArgs` beside the generic `topChoice` command, so text-only agents can execute the chosen result/source/form/action-target route without switching to kind-specific shortcut lines. | Keep this limited to the existing generic top-choice command args; avoid duplicating every nested choice payload in text output. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic text-mode executor routing. |
+| A165 | Search-decision command-args text parity | 100% | Text output now prints `searchDecisionCommandArgs` beside `searchDecisionCommand`, so decision-only text consumers can execute the chosen search result without joining against recommended or next-action fields. | Keep this limited to existing search-decision command metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves decision-only search routing. |
+| A166 | Page-decision command-args text parity | 100% | Text output now prints `pageDecisionCommandArgs` beside `pageDecisionCommand`, so page-check decision-only consumers can execute site-search or page follow-up commands without joining against next/top-choice fields. | Keep this limited to existing page-decision command metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves decision-only page routing. |
+| A167 | Primary action command text parity | 100% | Text output now prints `primaryCommand`, `primaryCommandArgs`, and primary after-interaction command shortcuts, so primary-action-only text consumers can execute the chosen follow-up without parsing nested action candidates. | Keep this limited to existing primary action execution metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves primary action text routing. |
+| A232 | Action target ISO/epoch freshness shortcuts | 100% | `topActionTarget*`, `executorTarget*`, `handoffTarget*`, `primaryTarget*`, and `alternativeActionTarget*` shortcuts now preserve existing `dateIso` and `dateUnixMs` target metadata in full, brief, text, public types, and static scorer checks, so agents can compare target freshness without parsing date text. | Keep this limited to already-normalized target date metadata; do not expand date inference without fixture evidence. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a target freshness routing shortcut gap. |
+| A168 | Primary action routing text parity | 100% | Text output now prints `primaryExecution`, `primaryUrl`, `primaryRank`, and `primaryOpenResult`, so primary-action-only text consumers can route or resume the chosen follow-up without joining against next/top-action fields. | Keep this limited to existing primary action routing metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves primary action text routing. |
+| A169 | Executor target text parity | 100% | Text output now prints existing `executorTarget*` URL/path/title/host/source/rank/date/relevance/selector shortcuts, so executor-only text consumers can identify the exact follow-up target without parsing nested executor payloads. | Keep this limited to existing executor target metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves executor text routing. |
+| A170 | Handoff target text parity | 100% | Text output now prints existing `handoffTarget*` URL/path/title/host/source/rank/date/relevance/selector shortcuts, so handoff-only text consumers can identify the exact follow-up target without parsing nested handoff payloads. | Keep this limited to existing handoff target metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves handoff text routing. |
+| A171 | Answer-plan URL text parity | 100% | Text output now prints `answerPlanUrl` beside answer-plan command shortcuts, so answer-plan-only text consumers can follow the planned URL without joining against `answerUrl`, next, or handoff fields. | Keep this limited to existing answer-plan routing metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves answer-plan text routing. |
+| A172 | Answer-evidence detail text parity | 100% | Text output now prints existing `topAnswerEvidenceTitle`, `topAnswerEvidenceUrl`, `topAnswerEvidenceConfidence`, and `topAnswerEvidenceReason` shortcuts, so text-only answer consumers can cite or open the top evidence without parsing nested evidence arrays. | Keep this limited to existing top answer-evidence metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves answer evidence text routing. |
+| A173 | Answer-evidence path/kind text parity | 100% | Text output now prints `topAnswerEvidencePath` and `topAnswerEvidenceKind` as separate shortcuts, so text-only answer consumers can branch on evidence location and type without parsing the aggregate evidence line. | Keep this limited to existing top answer-evidence metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves answer evidence text routing. |
+| A174 | Citation path/kind text parity | 100% | Text output now prints `topCitationPath`, `topCitationKind`, `topCitationConfidence`, and `topCitationScore` as separate shortcuts, so citation-only text consumers can branch on citation location and confidence without parsing the aggregate citation line. | Keep this limited to existing top citation metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves citation text routing. |
+| A175 | Source-search date detail shortcut parity | 100% | Source-search selected and alternate shortcuts now include existing `datePrecision` and `dateSource` beside `dateText`, so failed-open recovery can judge freshness confidence without parsing nested selected/alternate result objects. | Keep this limited to existing source-search result date metadata. | Focused CLI text test, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves source-search recovery routing. |
+| A176 | Nested choice command text parity | 100% | Result, source, form, and action-target choice text formatters now print prefix-specific `*Command` and `*CommandArgs` lines beside the existing nested command block, so text-only agents can grep executable routes from handoff and choice lists without indentation-sensitive parsing. | Keep existing nested `command` lines for compatibility and only mirror commands already present on the choice payload. | Focused CLI text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode handoff routing ergonomics. |
+| A177 | Action-candidate command text parity | 100% | Text output now prints command and after-interaction command details directly under each `actionCandidate`, plus prefix-specific `actionCandidate*Command*` lines, so agents comparing candidate actions can execute a chosen candidate without joining against primary or alternative shortcuts. | Keep this text-only mirror aligned with existing action payload command fields; do not add new action semantics. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode action comparison. |
+| A178 | PageCheck command text parity | 100% | The `pageCheck` text section now mirrors recommended-action and step command fields as `pageCheckCommand*` and `pageCheckStepCommand*` lines, so agents reading only the page-check block can execute follow-up steps without indentation-sensitive parsing. | Keep this text-only mirror aligned with existing recommended action and next-step command fields. | Focused CLI text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check-only text routing. |
+| A233 | Source-choice freshness shortcuts | 100% | Page source links now reuse existing result date hints, and source choices preserve `dateText`, `dateIso`, `dateUnixMs`, `datePrecision`, and `dateSource` in nested choices, top-source shortcuts, generic top-choice shortcuts, text output, public types, compact output, and static scorer checks. | Keep this limited to source link text/snippet date hints; do not add new date parsers without fixture evidence. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves source-link freshness comparison before browser handoff. |
+| A234 | Brief sitelink selector parity | 100% | Brief agent output now preserves first-sitelink selectors for top result choices and source-search selected/alternate choices, matching full compact output so subagents can target the exact DOM candidate without requesting full payloads. | Keep this to already-extracted sitelink selectors; do not expand the brief payload with nested result arrays unless a fixture proves it is needed. | Focused CLI tests, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a shallow full-vs-brief selector loss found by static comparison. |
+| A235 | Generic top-choice sitelink shortcuts | 100% | Generic `topChoice*` output now mirrors the first result sitelink title, URL, selector, command, and command args, so subagents can inspect or execute the strongest result sitelink without branching into result-specific shortcuts first. | Keep this to the first result sitelink; source/form/action-target choices should not gain synthetic sitelinks. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces branching in result handoff loops. |
+| A236 | Generic top-choice match signals | 100% | Generic `topChoice*` output now mirrors source hints, matched term, find-match text, and result sitelink count for result/source choices, so subagents can score the strongest choice without branching into result- or source-specific shortcut families. | Keep this limited to existing result/source match metadata; do not synthesize match signals for forms or action targets. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic choice scoring in search/page handoff loops. |
+| A237 | Generic form/action top-choice metadata | 100% | Generic `topChoice*` output now mirrors form/action target URLs, templates, query fields, query inputs, and disabled states, so subagents can understand page-check execution candidates without parsing `formChoices` or `actionTargetChoices` first. | Keep this limited to existing form/action metadata; do not invent a cross-kind schema beyond shallow copied fields. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic page-check choice routing. |
+| A238 | Generic action-target state shortcuts | 100% | Generic `topChoice*` output now mirrors action-target encoding type, pressed, expanded, popup, and controls state, so subagents can judge OpenSearch/JSON-LD action availability without parsing `actionTargetChoices` first. | Keep this limited to existing action-target state metadata; do not infer interaction state beyond extracted attributes. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic action-target handoff. |
+| A239 | Data-table text routing parity | 100% | Text-mode `topDataTable` output now includes header count and selector inline with the table path and size, so subagents can judge table shape and target the table without switching to JSON. | Keep JSON unchanged; this is a text-output parity improvement for existing `topDataTable*` fields. | Focused CLI text test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode table/grid handoff. |
+| A240 | Nested browser HTML reason codes | 100% | `browserHtml` capture objects now carry `reasonCode` alongside the capture script and rerun command, so subagents reading only `next`, `executionPlan`, `executor`, or `handoff` nested browser-capture payloads can understand why browser capture or inspection is required. | Keep the value aligned with top-level `browserHtmlReasonCode`; do not introduce a second reason taxonomy. | Focused CLI text/JSON test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces upward lookups in browser-capture handoff loops. |
+| A241 | Resource/media locator shortcuts | 100% | Top resource and media shortcuts now include `path` and `selector` fields, and text output prints them inline with URL and command context, so subagents can target feed/document/media evidence without parsing `pageCheck.resources` or `pageCheck.media` arrays first. | Keep this limited to existing path/selector metadata; do not synthesize selectors for resources or media that lack one. | Focused CLI JSON/brief/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check resource/media handoff. |
+| A242 | Embed/transcript/author locator shortcuts | 100% | Top embed, transcript, and author-link shortcuts now include `path` and `selector` fields, and text output prints them inline with URL and command context, so subagents can target embedded content, caption files, and author pages without parsing nested page-check arrays first. | Keep this limited to existing path/selector metadata; do not synthesize selectors for entries that lack one. | Focused CLI JSON/brief/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check embedded media and attribution handoff. |
+| A243 | FAQ/code locator shortcuts | 100% | Top FAQ and code-block shortcuts now include `path` and `selector` fields, and text output prints them inline with question/code context, so subagents can target FAQ answers and command snippets without parsing nested page-check arrays first. | Keep this limited to existing path/selector metadata; do not synthesize selectors for entries that lack one. | Focused CLI JSON/brief/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check structured text handoff. |
+| A244 | Breadcrumb locator shortcuts | 100% | Top breadcrumb shortcuts now include `selector`, and text output prints breadcrumb path, source, selector, and trail text in one line, so subagents can target navigation context without parsing `pageCheck.breadcrumbs` first. | Keep this limited to existing breadcrumb selector metadata from JSON-LD or HTML navigation. | Focused CLI JSON/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check navigation handoff. |
+| A245 | Pagination text routing parity | 100% | Text-mode `topPagination` output now prints kind, label, current state, selector, and URL inline with the page-check path, so subagents can choose next/previous/page links without switching to JSON. | Keep JSON unchanged; this is a text-output parity improvement for existing `topPagination*` fields. | Focused CLI pagination text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode navigation handoff. |
+| A246 | TOC text routing parity | 100% | Text-mode `topToc` output now prints title, item count, selector, first item label, first item URL, and summary text inline with the page-check path, so subagents can choose in-page section links without switching to JSON. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topToc*` fields. | Focused CLI TOC text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode table-of-contents handoff. |
+| A247 | Section text routing parity | 100% | Text-mode `topSection` output now prints heading level, selector, heading, and summary text inline with the page-check path, so subagents can target the strongest content section without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topSection*` fields. | Focused CLI section text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode content-section handoff. |
+| A248 | Provenance text locator parity | 100% | Text-mode `topProvenance` output now prints label, source, and selector inline with the provenance path, kind, value, and URL, so subagents can judge DOI/meta provenance and target the exact evidence node without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topProvenance*` fields. | Focused CLI provenance text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode provenance and source-trust handoff. |
+| A249 | Offer text locator parity | 100% | Text-mode `topOffer` output now prints offer name and selector inline with the offer path, price, availability, and URL, so subagents can connect price/buy evidence to the exact JSON-LD node without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topOffer*` fields. | Focused CLI offer text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode offer and purchase-evidence handoff. |
+| A250 | Dataset text locator parity | 100% | Text-mode `topDataset` output now prints encoding format, temporal/spatial coverage, creator, selector, first distribution URL, and license URL inline with the dataset path/name/URL, so subagents can judge dataset provenance and download/license routes without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topDataset*` fields. | Focused CLI dataset text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode dataset and data-download handoff. |
+| A251 | Identity text locator parity | 100% | Text-mode `topIdentity` output now prints source, selector, logo URL, and sameAs URL inline with the identity path/name/URL, so subagents can judge official organization/site identity and route logo/profile checks without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topIdentity*` fields. | Focused CLI identity text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode identity and official-source handoff. |
+| A252 | Timeline text freshness parity | 100% | Text-mode `topTimeline` output now prints label, normalized ISO date, epoch milliseconds, source, and selector inline with the timeline path/kind/value, so subagents can judge freshness evidence without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topTimeline*` fields. | Focused CLI timeline text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode freshness and publication-date handoff. |
+| A253 | Contact point text locator parity | 100% | Text-mode `topContactPoint` output now prints label, source, and selector inline with the contact path/kind/value/URL, so subagents can route email, phone, support URL, and address evidence without switching to JSON or brief output. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing `topContactPoint*` fields. | Focused CLI contact text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode contact and support-route handoff. |
+| A254 | Hidden app signal text locator parity | 100% | Text-mode `topHydration`, `topApiEndpoint`, `topClientState`, and `topAppHint` output now print labels, selectors, and client-state keys inline with the top hidden-signal path/kind/URL, so subagents can inspect app shells without switching to JSON or browser capture. | Keep JSON and brief output unchanged; this is a text-output parity improvement for existing hidden app signal fields. | Focused CLI hidden-signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode client-rendered app handoff. |
+| A255 | Generic hidden signal text summary | 100% | Text-mode output now prints a single `topHiddenSignal` summary line with group, path, kind, source, selector, URL, and text, so subagents can classify the best hidden signal even when it is not one of the dedicated top hydration/API/client-state/app-hint lines. | Keep the existing scalar `topHiddenSignal*` lines for compatibility; this is a text-output readability improvement. | Focused CLI hidden-signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic hidden data handoff. |
+| A256 | Hidden signal group count parity | 100% | Agent output now exposes shallow counts for runtime, config, mobile hints, topics, key-values, meta facts, HTTP policies, and schema facts, so subagents can judge hidden-data coverage from the agent summary instead of inspecting nested `pageCheck` arrays first. | Keep existing detailed `pageCheck.*` arrays unchanged; these are count shortcuts for routing and triage. | Focused CLI runtime/config count tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves hidden-data triage and browser-capture avoidance. |
+| A257 | Runtime and config top shortcuts | 100% | Agent output now exposes `topRuntime*` URL/selector/command shortcuts and `topConfig*` name/key/selector shortcuts, so subagents can fetch app runtime assets or inspect app config keys without parsing nested hidden-data arrays first. | Keep the full `pageCheck.runtime` and `pageCheck.config` arrays as the detailed source of truth. | Focused CLI runtime/config and text-output tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves app-shell handoff and browser-capture avoidance. |
+| A258 | Metadata top shortcut parity | 100% | Agent output now exposes `topMobileHint*`, `topTopic*`, `topKeyValue*`, and `topMetaFact*` shortcuts with labels, values, sources, selectors, and URLs/datetimes where available, so subagents can classify page intent and static metadata without parsing nested arrays first. | Keep full `pageCheck.mobileHints`, `topics`, `keyValues`, and `metaFacts` arrays as detailed evidence. | Focused CLI metadata tests, text-output test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves static metadata handoff beyond browser-tree-only signals. |
+| A259 | Policy and schema top shortcut parity | 100% | Agent output now exposes `topHttpPolicy*` and `topSchemaFact*` shortcuts with policy names/values/sources/selectors and first schema fact/type/count details, so subagents can judge security/indexing directives and JSON-LD structure without parsing nested arrays first. | Keep full `pageCheck.httpPolicies` and `pageCheck.schemaFacts` arrays as detailed evidence. | Focused CLI policy/schema and text-output tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves static policy/schema handoff beyond browser-tree-only signals. |
+| A260 | Resource and media detail shortcuts | 100% | Agent output now exposes top resource `rel`, `type`, and `hreflang` plus top media alt/caption/title/dimensions, so subagents can classify feeds, alternates, documents, images, and social preview media without parsing nested arrays first. | Keep full `pageCheck.resources` and `pageCheck.media` arrays as detailed evidence. | Focused CLI resource/media and text-output tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves resource/media handoff and image accessibility parity. |
+| A261 | Barrier text routing parity | 100% | Text-mode `topBarrier` output now prints source, diagnostic code, and selector inline with severity, kind, path, and text, so subagents can classify blocked-page handoffs without switching to JSON. | Keep existing `topBarrierSource`, `topBarrierSelector`, and `topBarrierDiagnosticCode` scalar lines for compatibility. | Focused CLI barrier text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves browser-fallback routing for login, paywall, challenge, and consent barriers. |
+| A262 | PageCheck barrier text detail parity | 100% | Text-mode `pageCheck` barrier rows now print source, diagnostic code, and selector inline for every barrier, so subagents comparing multiple login/paywall/challenge/consent signals can classify each row without switching to JSON. | Keep structured `pageCheck.barriers` unchanged; this is text-output readability for existing fields. | Focused CLI barrier text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves multi-barrier routing and blocked-page triage. |
+| A263 | PageCheck form text routing parity | 100% | Text-mode `pageCheck` form rows now print hidden-field count, action URL, query field, URL template, target, encoding type, accept charset, novalidate state, and selector inline, so subagents can choose and execute page forms without switching to JSON or the separate choice list. | Keep structured `pageCheck.forms` and `formChoice` output unchanged; this is text-output readability for existing form fields. | Focused CLI form text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-form execution handoff. |
+| A264 | PageCheck action-target text routing parity | 100% | Text-mode `pageCheck` action-target rows now print source, URL template, query input, method, encoding type, ARIA state, controls, and selector inline, so subagents can choose JSON-LD and OpenSearch actions without switching to JSON or the separate choice list. | Keep structured `pageCheck.actionTargets` and `actionTargetChoice` output unchanged; this is text-output readability for existing action-target fields. | Focused CLI action-target text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves search/action execution handoff. |
+| A265 | PageCheck app-shell text routing parity | 100% | Text-mode `pageCheck` hydration, API endpoint, client-state, runtime, config, and app-hint rows now print source, framework/build/route, method, key/name/key counts, label, and selector details inline, so subagents can inspect hidden app-shell evidence without switching to JSON. | Keep structured hidden-signal arrays and top shortcuts unchanged; this is text-output readability for existing app-shell fields. | Focused CLI hidden-signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens static app-shell evidence beyond browser accessibility trees. |
+| A266 | PageCheck metadata text routing parity | 100% | Text-mode `pageCheck` mobile hint, topic, key-value, meta fact, HTTP policy, and schema fact rows now print source, label/name, selector, type, and fact-count details inline, so subagents can inspect static metadata without switching to JSON. | Keep structured metadata arrays and top shortcuts unchanged; this is text-output readability for existing metadata fields. | Focused CLI hidden-signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens static metadata evidence beyond browser accessibility trees. |
+| A267 | PageCheck provenance text routing parity | 100% | Text-mode `pageCheck` provenance rows now print source, label, value, selector, and follow-up URL inline, so subagents can judge DOI/license/publisher provenance without switching to JSON. | Keep structured `pageCheck.provenance` and top provenance shortcuts unchanged; this is text-output readability for existing provenance fields. | Focused CLI provenance text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves source-trust and citation handoff. |
+| A268 | PageCheck offer text routing parity | 100% | Text-mode `pageCheck` offer rows now print source, product name, currency, price text/number, availability, brand/SKU/rating/review metadata, selector, and URL inline, so subagents can judge product/commercial evidence without switching to JSON. | Keep structured `pageCheck.offers` and top offer shortcuts unchanged; this is text-output readability for existing offer fields. | Focused CLI offer text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves commerce and price-evidence handoff. |
+| A269 | PageCheck identity text routing parity | 100% | Text-mode `pageCheck` identity rows now print source, name, logo URL, sameAs URLs, selector, and canonical URL inline, so subagents can judge organization/person/brand identity evidence without switching to JSON. | Keep structured `pageCheck.identities` and top identity shortcuts unchanged; this is text-output readability for existing identity fields. | Focused CLI identity text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves official-source and brand-identity handoff. |
+| A270 | PageCheck dataset text routing parity | 100% | Text-mode `pageCheck` dataset rows now print source, dataset name, encoding format, temporal/spatial coverage, creator, first distribution URL, license URL, selector, and landing URL inline, so subagents can judge data provenance and download/license routes without switching to JSON. | Keep structured `pageCheck.datasets` and top dataset shortcuts unchanged; this is text-output readability for existing dataset fields. | Focused CLI dataset text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves dataset and data-download handoff. |
+| A271 | PageCheck timeline text routing parity | 100% | Text-mode `pageCheck` timeline rows now print source, label, raw value, normalized ISO date, epoch milliseconds, and selector inline, so subagents can judge freshness and update evidence without switching to JSON. | Keep structured `pageCheck.timeline` and top timeline shortcuts unchanged; this is text-output readability for existing timeline fields. | Focused CLI timeline text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves freshness and publication-date handoff. |
+| A272 | PageCheck contact text routing parity | 100% | Text-mode `pageCheck` contact point rows now print source, label, normalized contact value, selector, and URL inline, so subagents can identify emails, phone numbers, addresses, and contact URLs without switching to JSON. | Keep structured `pageCheck.contactPoints` and top contact shortcuts unchanged; this is text-output readability for existing contact fields. | Focused CLI contact text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves contact and support-route handoff. |
+| A273 | PageCheck FAQ text routing parity | 100% | Text-mode `pageCheck` FAQ rows now print source, question, answer, and selector inline, so subagents can reuse Q/A evidence without parsing the combined display text or switching to JSON. | Keep structured `pageCheck.faqs` and top FAQ shortcuts unchanged; this is text-output readability for existing FAQ fields. | Focused CLI FAQ text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves support, docs, and answer-evidence handoff. |
+| A274 | PageCheck breadcrumb text routing parity | 100% | Text-mode `pageCheck` breadcrumb rows now print source, item count, first/last labels, crumb URLs, and selector inline, so subagents can judge page hierarchy and navigation routes without switching to JSON. | Keep structured `pageCheck.breadcrumbs` and top breadcrumb shortcuts unchanged; this is text-output readability for existing breadcrumb fields. | Focused CLI breadcrumb text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves hierarchy and navigation-evidence handoff. |
+| A275 | PageCheck section text routing parity | 100% | Text-mode `pageCheck` section rows now print heading level, heading text, excerpt count, first excerpt, and selector inline, so subagents can inspect document structure and section evidence without switching to JSON. | Keep structured `pageCheck.sections` and top section shortcuts unchanged; this is text-output readability for existing section fields. | Focused CLI section text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves heading and content-section handoff. |
+| A276 | PageCheck pagination text routing parity | 100% | Text-mode `pageCheck` pagination rows now print source, label, current-page state, selector, and URL inline, so subagents can choose next/previous/page navigation targets without switching to JSON. | Keep structured `pageCheck.pagination` and top pagination shortcuts unchanged; this is text-output readability for existing pagination fields. | Focused CLI pagination text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves navigation and multi-page evidence handoff. |
+| A277 | PageCheck TOC text routing parity | 100% | Text-mode `pageCheck` TOC rows now print title, item count, first/last labels, first/last heading levels, item URLs, and selector inline, so subagents can inspect in-page navigation and document outline routes without switching to JSON. | Keep structured `pageCheck.toc` and top TOC shortcuts unchanged; this is text-output readability for existing TOC fields. | Focused CLI TOC text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves document-outline and in-page navigation handoff. |
+| A278 | PageCheck code block text routing parity | 100% | Text-mode `pageCheck` code block rows now print source, language, command-like state, line count, and selector inline, so subagents can identify executable snippets and code evidence without switching to JSON. | Keep structured `pageCheck.codeBlocks` and top code block shortcuts unchanged; this is text-output readability for existing code block fields. | Focused CLI code block text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves command and code-evidence handoff. |
+| A279 | PageCheck citation text routing parity | 100% | Text-mode `pageCheck` citation rows now print source, title, quote, selector, and URL inline, so subagents can identify citation/reference evidence without switching to JSON. | Keep structured `pageCheck.citations` and top citation shortcuts unchanged; this is text-output readability for existing citation fields. | Focused CLI citation text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves citation and reference-evidence handoff. |
+| A280 | PageCheck media text routing parity | 100% | Text-mode `pageCheck` media rows now print alt text, caption, title, dimensions, selector, and URL inline, so subagents can inspect image/media evidence without switching to JSON. | Keep structured `pageCheck.media` and top media shortcuts unchanged; this is text-output readability for existing media fields. | Focused CLI media text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves image, caption, and visual-evidence handoff. |
+| A281 | PageCheck resource text routing parity | 100% | Text-mode `pageCheck` resource rows now print title, rel, type, hreflang, selector, and URL inline, so subagents can identify feeds, alternates, manifests, licenses, downloads, and documents without switching to JSON. | Keep structured `pageCheck.resources` and top resource shortcuts unchanged; this is text-output readability for existing resource fields. | Focused CLI resource text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves resource and download-route handoff. |
+| A282 | PageCheck embed text routing parity | 100% | Text-mode `pageCheck` embed rows now print title, type, poster URL, source URLs, sandbox/allow/loading attributes, selector, and URL inline, so subagents can inspect iframe/video/audio/embed evidence without switching to JSON. | Keep structured `pageCheck.embeds` and top embed shortcuts unchanged; this is text-output readability for existing embed fields. | Focused CLI embed text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves embedded-content and media-source handoff. |
+| A283 | PageCheck transcript text routing parity | 100% | Text-mode `pageCheck` transcript rows now print media kind, language, label, selector, and URL inline, so subagents can identify captions, subtitles, descriptions, chapters, and transcript resources without switching to JSON. | Keep structured `pageCheck.transcripts` and top transcript shortcuts unchanged; this is text-output readability for existing transcript fields. | Focused CLI transcript text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves caption and transcript-resource handoff. |
+| A284 | PageCheck author link text routing parity | 100% | Text-mode `pageCheck` author link rows now print source, name, rel, selector, and URL inline, so subagents can identify author/byline/profile routes without switching to JSON. | Keep structured `pageCheck.authorLinks` and top author link shortcuts unchanged; this is text-output readability for existing author link fields. | Focused CLI author link text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves authorship and profile-route handoff. |
+| A285 | PageCheck action text routing parity | 100% | Text-mode `pageCheck` action rows now print action type, selector, and visible text as named fields, so subagents can identify clickable controls without parsing positional text or selector parentheses. | Keep structured `pageCheck.actions` unchanged; this is text-output readability for existing action fields. | Focused CLI pageCheck text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves interactive-control handoff without browser work. |
+| A286 | PageCheck link text routing parity | 100% | Text-mode `pageCheck` primary link and source link rows now print kind, title, source host, rank, source type, score, hints, official state, selector, and URL as named fields, so subagents can compare source routes without parsing positional titles. | Keep structured `pageCheck.primaryLinks` and `pageCheck.sourceLinks` unchanged; this is text-output readability for existing link fields. | Focused CLI pageCheck/find text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves source-route and external-link handoff. |
+| A287 | PageCheck evidence text routing parity | 100% | Text-mode `pageCheck` content evidence rows now print rank, role, source, quality, score, selector, and evidence text as named fields, so subagents can cite page excerpts without parsing positional role/quality markers. | Keep structured `pageCheck.contentEvidence` unchanged; this is text-output readability for existing evidence fields. | Focused CLI pageCheck text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves answer-evidence and citation handoff. |
+| A288 | PageCheck data table text routing parity | 100% | Text-mode `pageCheck` data table rows now print rank, row count, column count, headers, caption, first sample row, selector, and summary text as named fields, so subagents can inspect tabular evidence without parsing size shorthands. | Keep structured `pageCheck.dataTables` unchanged; this is text-output readability for existing table fields. | Focused CLI data table text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves table/grid evidence handoff. |
+| A289 | PageCheck barrier text routing parity | 100% | Text-mode `pageCheck` barrier rows now print kind, severity, source, diagnostic code, selector, and barrier text as named fields, so subagents can decide when browser-captured HTML or live browser handling is required without parsing positional severity/kind tokens. | Keep structured `pageCheck.barriers` and top barrier shortcuts unchanged; this is text-output readability for existing barrier fields. | Focused CLI barrier text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves blocked-page and browser-handoff decisions. |
+| A290 | PageCheck form method text routing parity | 100% | Text-mode `pageCheck` form rows now print the submit method as `method=...` alongside field counts, action URL, query field, URL template, target, encoding, charset, novalidate state, selector, and form text, so subagents can execute forms without parsing positional uppercase methods. | Keep structured `pageCheck.forms` and `formChoice` output unchanged; this is text-output readability for existing form fields. | Focused CLI form text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves form execution handoff. |
+| A291 | PageCheck action target kind text routing parity | 100% | Text-mode `pageCheck` action target rows now print action target kind as `kind=...` alongside source, URL template, query input, method, encoding, ARIA state, controls, selector, target URL, and text, so subagents can choose JSON-LD/OpenSearch actions without parsing positional kind tokens. | Keep structured `pageCheck.actionTargets` and `actionTargetChoice` output unchanged; this is text-output readability for existing action target fields. | Focused CLI action target text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves structured action execution handoff. |
+| A292 | PageCheck hidden client signal text routing parity | 100% | Text-mode `pageCheck` hydration, API endpoint, client state, runtime, and config rows now print kind as `kind=...` beside source, selectors, URLs, and extracted metadata, so subagents can judge client-rendered behavior and hidden data routes without parsing positional kind tokens. | Keep structured hidden-signal arrays unchanged; this is text-output readability for existing hidden signal fields. | Focused CLI hidden signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves client-rendered and hidden-data handoff. |
+| A293 | PageCheck app/mobile/topic text routing parity | 100% | Text-mode `pageCheck` app hint, mobile hint, and topic rows now print kind as `kind=...` beside source, labels, selectors, URLs, and platform metadata, so subagents can identify installability, viewport/mobile hints, and topical metadata without parsing positional kind tokens. | Keep structured hint/topic arrays unchanged; this is text-output readability for existing fields. | Focused CLI hidden/app signal text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves app/mobile/topic metadata handoff. |
+| A294 | PageCheck provenance kind text routing parity | 100% | Text-mode `pageCheck` provenance rows now print kind as `kind=...` beside source, label, value, selector, URL, and provenance text, so subagents can judge DOI/license/publisher evidence without parsing positional kind tokens. | Keep structured `pageCheck.provenance` and top provenance shortcuts unchanged; this is text-output readability for existing provenance fields. | Focused CLI provenance text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves provenance and source-trust handoff. |
+| A295 | PageCheck identity and dataset kind text routing parity | 100% | Text-mode `pageCheck` identity and dataset rows now print kind as `kind=...` beside source, names, selectors, URLs, and extracted metadata, so subagents can distinguish organizations, people, datasets, and data downloads without parsing positional kind tokens. | Keep structured `pageCheck.identities`, `pageCheck.datasets`, and top shortcuts unchanged; this is text-output readability for existing fields. | Focused CLI identity/dataset text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves official-source and data-provenance handoff. |
+| A296 | PageCheck timeline and contact kind text routing parity | 100% | Text-mode `pageCheck` timeline and contact point rows now print kind as `kind=...` beside source, labels, values, selectors, URLs, and date metadata, so subagents can distinguish published/modified dates and email/phone/support routes without parsing positional kind tokens. | Keep structured `pageCheck.timeline`, `pageCheck.contactPoints`, and top shortcuts unchanged; this is text-output readability for existing fields. | Focused CLI timeline/contact text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves freshness and support-route handoff. |
+| A297 | PageCheck pagination kind text routing parity | 100% | Text-mode `pageCheck` pagination rows now print kind as `kind=...` beside source, label, current-page state, selector, URL, and text, so subagents can choose next/previous/page routes without parsing positional kind tokens. | Keep structured `pageCheck.pagination` and top pagination shortcuts unchanged; this is text-output readability for existing fields. | Focused CLI pagination text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves navigation and multi-page handoff. |
+| A298 | PageCheck media and resource kind text routing parity | 100% | Text-mode `pageCheck` media and resource rows now print kind as `kind=...` beside alt/title/rel/type/selector/URL metadata, so subagents can distinguish images, social previews, feeds, alternates, manifests, documents, and downloads without parsing positional kind tokens. | Keep structured `pageCheck.media`, `pageCheck.resources`, and top shortcuts unchanged; this is text-output readability for existing fields. | Focused CLI media/resource text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves visual evidence and resource-route handoff. |
+| A299 | PageCheck embed and transcript kind text routing parity | 100% | Text-mode `pageCheck` embed and transcript rows now print kind as `kind=...` beside title, media, language, label, selector, URL, and embed attributes, so subagents can distinguish iframes, embedded media, captions, subtitles, chapters, and transcript routes without parsing positional kind tokens. | Keep structured `pageCheck.embeds`, `pageCheck.transcripts`, and top shortcuts unchanged; this is text-output readability for existing fields. | Focused CLI embed/transcript text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves embedded-content and caption/transcript handoff. |
+| A300 | Semantic outline kind text routing parity | 100% | Text-mode agent `semanticOutline` rows now print `kind=...` and `text="..."` beside role, depth, parent, and selector metadata, so subagents can inspect accessibility-outline landmarks/headings without parsing `kind:text` tokens. | Keep structured `agent.semanticSummary.semanticOutline` and scalar top outline shortcuts unchanged; this is text-output readability for existing semantic fields. | Focused CLI semantic text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves accessibility-outline handoff. |
+| A301 | Semantic role/name text routing parity | 100% | Text-mode agent semantic keyboard, shortcut, heading, landmark, and interactive rows now print role/name/text as named fields, so subagents can inspect focusable controls, landmarks, headings, and button labels without parsing `role:name` or positional text tokens. | Keep structured `agent.semanticSummary` and scalar top semantic shortcuts unchanged; this is text-output readability for existing semantic fields. | Focused CLI semantic text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves accessibility-tree text handoff. |
+| A302 | Semantic top shortcut role/name text routing parity | 100% | Text-mode `semanticTopHeading`, `semanticTopLandmark`, `semanticTopNamedRole`, `semanticTopInteractive`, and `semanticTopFocusable` now print text, role, and name as named fields, so subagents can route top accessibility targets without parsing positional labels or `role:name` tokens. | Keep structured `agent.semanticSummary` and scalar top shortcut fields unchanged; this is text-output readability for existing semantic shortcuts. | Focused CLI semantic top shortcut tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves top accessibility-target handoff. |
+| A303 | Semantic top action target name text routing parity | 100% | Text-mode `semanticTopLink`, `semanticTopInPageLink`, `semanticTopButton`, and `semanticTopImage` now print `name="..."` and in-page link `kind=...`, so subagents can route clickable, skippable, button, and image targets without parsing positional labels or `kind:name` tokens. | Keep structured `agent.semanticSummary` and scalar top shortcut fields unchanged; this is text-output readability for existing semantic action targets. | Focused CLI semantic top button test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves clickable and visual accessibility-target handoff. |
+| A304 | Top form required-field shortcuts | 100% | Agent output now exposes `topFormChoiceRequiredField*` beside first-field shortcuts, so subagents can identify the first required input without parsing nested form fields first. | Keep the full `formChoices.fields` array as detailed evidence and add more field-selection shortcuts only when fixtures show a repeated need. | Focused CLI form text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves form execution handoff and accessibility-form parity. |
+| A305 | Top form invalid-field shortcuts | 100% | Agent output now exposes `topFormChoiceInvalidField*` for the first invalid field, so subagents can identify validation-error targets without parsing nested form fields first. | Keep the shortcut compact: name, type, label, invalid state, and selector. Add richer invalid-field detail only if fixtures show it is needed. | Focused CLI form text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves form error handoff and accessibility-form parity. |
+| A306 | Generic topChoice form field aliases | 100% | Generic `topChoice*` output now mirrors required and invalid form-field names/selectors when the selected choice is a form, so subagents using only the unified top-choice path can identify required input and validation targets. | Keep this as a compact alias of existing form-choice details; richer field metadata remains on `topFormChoice*`. | Focused CLI form text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic choice handoff. |
+| A307 | Generic topChoice URL path shortcuts | 100% | Generic `topChoice*` output now exposes `topChoiceUrlPath` and optional `topChoiceUrlQuery`, so subagents using the unified top-choice path can distinguish same-host result/source/form/action URLs without parsing full URLs. | Keep result/source-specific URL path shortcuts as detailed aliases; generic topChoice stays the unified first-read path. | Focused CLI source-choice text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic choice routing. |
+| A308 | Top action target URL path shortcuts | 100% | Agent output now exposes `topActionTargetUrlPath` and optional `topActionTargetUrlQuery`, so action-first loops can distinguish same-host primary targets without parsing full URLs. | Keep full `topActionTargetUrl` and nested `actions[].target` as the canonical detail; these are shallow routing shortcuts. | Focused CLI source-choice text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves primary action routing. |
+| A309 | Action-target choice URL-template path shortcuts | 100% | Action-target choices now expose `targetUrlPath`, `targetUrlQuery`, `urlTemplatePath`, `urlTemplateQuery`, plus matching `topActionTargetChoice*` shortcuts, so subagents can distinguish OpenSearch/JSON-LD action routes without parsing full URLs or templates. | Keep full target/template URLs as canonical execution detail; path/query fields are shallow routing aids for action-choice handoff. | Focused CLI action-target JSON/brief/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves structured action execution routing. |
+| A310 | Form choice URL path shortcuts | 100% | Form choices now expose `actionUrlPath`, `actionUrlQuery`, `urlTemplatePath`, `urlTemplateQuery`, plus matching `topFormChoice*` shortcuts, so subagents can distinguish search form routes without parsing full URLs or templates. | Keep full form action/template URLs and command args as canonical execution detail; path/query fields are shallow routing aids for form handoff. | Focused CLI form brief/text/submit-override tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves search-form execution routing. |
+| A311 | Recommended URL path shortcuts | 100% | Recommended search-result handoff now exposes `recommendedUrlPath`, optional `recommendedUrlQuery`, and matching `searchDecisionRecommendedUrlPath/Query`, so subagents can route recommended results without parsing full URLs. | Keep full recommended URLs and command args as canonical execution detail; path/query fields are shallow routing aids for search-result choice handoff. | Focused CLI search-decision/brief/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves recommended-result routing. |
+| A312 | Citation/evidence URL path shortcuts | 100% | Citation and answer-evidence handoff now exposes `topCitationUrlPath`, optional `topCitationUrlQuery`, `topAnswerEvidenceUrlPath`, and optional `topAnswerEvidenceUrlQuery`, so subagents can route evidence verification links without parsing full URLs. | Keep full evidence URLs and command args as canonical detail; path/query fields are shallow routing aids for citation and answer-evidence follow-up. | Focused CLI handoff/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves evidence verification routing. |
+| A313 | Identity/contact URL path shortcuts | 100% | Identity and contact handoff now exposes URL path/query shortcuts for canonical identity URLs, logos, first sameAs profile, and top contact URL, so subagents can compare official/source/contact targets without parsing full URLs. | Keep full URLs and command args as canonical detail; path/query fields are shallow routing aids for structured identity and contact follow-up. | Focused CLI identity/contact tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves official identity and contact routing. |
+| A314 | Hidden technical URL path shortcuts | 100% | Hidden technical handoff now exposes URL path/query shortcuts for hydration data, API endpoints, runtime/service-worker URLs, app manifests, mobile app hints, and the selected hidden signal, so subagents can route app-shell evidence without parsing full URLs. | Keep full URLs and command args as canonical detail; path/query fields are shallow routing aids for hidden app and API follow-up. | Focused CLI hydration/API/runtime/app/text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves hidden app-shell and API routing. |
+| A315 | Next read-value shortcut parity | 100% | `next` summaries now expose read-value path, type, count, and reference-path shortcuts in text, full, brief, public types, readiness, and static scoring, so first-step agents can decide whether to read inline or follow a referenced payload without parsing nested `next.readValue`. | Keep these fields shallow and do not inline large values into brief output. | Focused CLI text test, public type test, static fixture gate, typecheck, readiness audit, diff check, and process check pass. | No estimate increase; closes the remaining read-value stage-parity gap. |
+| A316 | Next read-value reference regression guard | 100% | Readiness audit and brief-output coverage now explicitly require `nextReadValueReferencePath`, so compact referenced payloads cannot silently lose the follow-path shortcut while path/type/count still pass. | Keep reference-path coverage paired with compact read-value scoring. | Focused CLI text/brief tests, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; strengthens regression detection for first-step read payload routing. |
+| A317 | Stage read-value compact reference parity | 100% | Compact full/brief output now derives runbook, next, executor, and handoff read-value reference-path shortcuts from the compacted payload shape, so agents keep a followable payload path even when nested stage objects are minimized or omitted. | Keep stage reference-path shortcuts aligned with `compactAgentReadValue` rules. | Focused CLI brief test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a compact-output stage parity gap. |
+| A318 | Text read-value reference parity | 100% | Text output now derives runbook, next, executor, and handoff read-value reference-path lines from the same compact read-value rules as JSON output, so text-only subagents can follow structured payload paths without nested parsing. | Keep text reference-path lines shallow and stage-specific. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; closes a text-mode read payload routing gap. |
+| A319 | Read-value scorer reference parity | 100% | Static comparison scoring now treats compact inline arrays/objects as followable reference paths, matching runtime `*ReadValueReferencePath` semantics for runbook, next, executor, and handoff shortcuts. | Keep scorer reference semantics synchronized with `compactAgentReadValueReferencePath`. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; prevents false shortcut regressions after compact reference parity. |
+| A320 | Read-value contract feature discovery | 100% | The public agent contract now declares `readValue.shortcuts` and `readValue.referencePaths`, so typed and runtime consumers can detect stage-level read-value routing support without inferring it from individual fields. | Keep compact/brief contract output on `featureCount`; add markers only for stable capability groups. | Public type test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves capability discovery for runbook/next/executor/handoff read payload routing. |
+| A321 | Page form URL path parity | 100% | `pageCheck.forms` now fills `actionUrlPath`, optional `actionUrlQuery`, `urlTemplatePath`, and optional `urlTemplateQuery`, matching the public `AgentPageForm` type and the existing agent form-choice path shortcuts. | Keep URL decomposition shallow; full URLs remain canonical execution detail. | Focused CLI form test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; fixes a discovered form-routing fixture gap while verifying agent contract output. |
+| A322 | Page form URL path gate coverage | 100% | Static comparison now scores `pageCheck.forms` action/template path and query fields directly, so regressions in page-check form routing cannot hide behind still-correct agent form-choice shortcuts. | Keep this as a small original-payload parity score beside the existing form-choice scorer. | Static fixture gate, comparison-gate test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page-check form routing regression coverage. |
+| A323 | Page action-target URL path gate coverage | 100% | Static comparison now scores `pageCheck.actionTargets` target/template path and query fields directly, so action-target routing regressions cannot hide behind still-correct agent action-target choice shortcuts. | Keep this beside the existing action-target choice scorer and avoid broad browser checks for this non-browser regression guard. | Static fixture gate, comparison-gate test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; strengthens page-check action execution regression coverage. |
+| A324 | In-page link text row parity | 100% | Text-mode agent output now mirrors in-page link fields as individual prefix rows, including path, kind, name, URL, path/query, target id, and selector. | Keep this as text-only exposure of existing `semanticTopInPageLink*` fields; add deeper anchor/TOC shortcuts only when fixture evidence shows first in-page link ambiguity. | Focused CLI in-page link text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves skip-link and anchor routing for text-only subagents. |
+| A325 | Top link text row parity | 100% | Text-mode agent output now mirrors top-link fields as individual prefix rows, including name, path, URL, path/query, target, relation, type, language, current state, download, and selector. | Keep this as text-only exposure of existing `semanticTopLink*` fields; add deeper link ranking shortcuts only when fixture evidence shows first-link ambiguity. | Focused CLI top-link text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves first-link and primary navigation routing for text-only subagents. |
+| A326 | Top button text row parity | 100% | Text-mode agent output now mirrors top-button fields as individual prefix rows, including name, path, description, type, state, disabled/pressed/expanded/popup state, controls target, form override metadata, and selector. | Keep this as text-only exposure of existing `semanticTopButton*` fields; add deeper button ranking shortcuts only when fixture evidence shows first-button ambiguity. | Focused CLI top-button text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves button execution and accessibility-control routing for text-only subagents. |
+| A327 | Top image text row parity | 100% | Text-mode agent output now mirrors top-image fields as individual prefix rows, including path, accessible name, URL, dimensions, loading/decoding metadata, srcset, sizes, and selector. | Keep this as text-only exposure of existing `semanticTopImage*` fields; add deeper image ranking shortcuts only when fixture evidence shows first-image ambiguity. | Focused CLI top-image text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves visual evidence and alt-text routing for text-only subagents. |
+| A328 | Top field text row parity | 100% | Text-mode agent output now mirrors top-field accessibility and state fields as individual prefix rows, including role, path, name, labelled-by/described-by/details/error-message text and selectors, required/readonly/invalid state, and selector. | Keep deeper HTML/value field metadata on the aggregate row unless fixture evidence shows text-only agents need more scalar rows. | Focused CLI top-field text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves form/input accessibility routing for text-only subagents. |
+| A329 | Source-search alternate choices text label | 100% | Text-mode agent output now prints a compact `sourceSearchAlternateChoices` summary line before per-choice details, so failed-open recovery agents can find the alternate-choice array by name without parsing singular choice rows. | Keep this line short; detailed candidates stay in the existing `sourceSearchAlternateChoice*` rows. | Focused source-search text test, source-search text-label guard, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves text-mode failed-open recovery. |
+| A330 | Citation and answer-evidence text row parity | 100% | Text-mode agent output now mirrors top citation and answer-evidence id/text/score fields as individual rows, so answer and verification subagents can cite or score evidence without parsing aggregate citation lines. | Keep these rows shallow and aligned with existing citation/evidence shortcut fields. | Focused citation/evidence text tests, text-label guard, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves text-mode answer evidence routing. |
+| A331 | Diagnostic and quality-gate text row parity | 100% | Text-mode agent output now mirrors top signal, top quality gate, problem signal, and failing quality gate shortcut fields as individual rows, so subagents can classify warnings/errors without parsing aggregate diagnostic lines. | Keep these rows shallow and aligned with existing compact/brief diagnostic fields. | Focused diagnostic text test, text-label guard, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves text-mode failure triage. |
+| A332 | Shortcut contract feature discovery | 100% | The public agent contract now declares stable shortcut capability groups for text, semantic, citation, answer-evidence, and diagnostics shortcuts, so subagents can detect these handoff surfaces without inferring support from individual scalar fields. | Keep compact/brief contract output on `featureCount`; add markers only for stable capability groups that are already covered by runtime and type tests. | Public type test, focused contract fixture test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves capability discovery for text-only and typed subagents. |
+| A333 | Browser fixture ledger IDs | 100% | Local browser parity check results now include `ledgerId` values that point to the documented G-row behind each comparison, making fixture output auditable without reading the whole progress ledger. | Keep new browser checks paired with a G-row before landing them; do not broaden browser execution just to satisfy the ledger. | Static browser-fixture ledger test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves traceability for browser parity research. |
+| A334 | Risky npm script inventory guard | 100% | Readiness audit now keeps an explicit allowlist of test, comparison, and `agent-browser` smoke scripts, so new broad runners cannot be added without updating the safety review. | Keep the allowlist small and review any new script for sequential execution and process cleanup requirements before adding it. | Agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Process safety 89% -> 90%; protects against repeat host overload. |
+| A335 | Source-search engine-attempt recovery shortcuts | 100% | Opened search-result handoffs now expose source-search engine attempt count, success/failure counts, first successful engine/result count, and first failed engine/error/status in agent output. | Watch whether future source-search fixtures need deeper per-engine arrays; keep the shallow fields unless recovery still requires nested parsing. | Focused CLI test, public type test, typecheck, readiness audit, README test, diff check, and process check pass. | Search handoff 91% -> 92%; helps failed/blocked auto-search recovery without browser inspection. |
+| A336 | Source-search engine-attempt static scoring | 100% | `compare-static` now understands source-search engine-attempt shortcut fields and scores them against `searchEngines` when that reference data is present. | Add a positive source-search comparison fixture only if future static gates emit source-search plus `searchEngines` together; keep current scoring reference-aware. | Static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; turns A335 into a guarded regression surface. |
+| A337 | Source-search engine-attempt fixture gate | 100% | The static agent fixture set now includes an offline auto-search/open-result target with one blocked engine, two successful engine attempts, source-search output, compact `searchEngines`, and source-search provenance read-target scoring preserved in agent/brief output. | Keep the fixture offline and sequential; broaden only when another recovery class needs coverage. | Focused CLI test, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; provides positive regression evidence for A335/A336. |
+| A338 | Source-search engine-attempt readiness guard | 100% | Readiness audit now requires the source-search engine-attempt shortcut fields across full, brief, and public shortcut evidence lists. | Keep this list aligned when adding source-search recovery shortcuts so regressions fail before broad comparison runs. | Agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; keeps A335-A337 from silently drifting. |
+| A339 | Source-search engine-attempt text-output guard | 100% | Text-mode search/open-result output now has a focused regression test for selected engine, attempt counts, success/failure counts, first successful engine/result count, and first failed engine/error/status rows. | Keep this as a mock-only CLI test; do not use live browser or network checks for text-row parity. | Focused CLI text-output test passes; broader gates pending. | No headline increase; protects text-only subagent recovery from parser drift. |
+| A340 | Source-search provenance gate-summary guard | 100% | The offline static fixture test now asserts the aggregate `averageAgentSourceSearchProvenanceScore`, not only the individual source-search fixture row, so summary aggregation drift is caught before broader gates. | Keep this tied to offline fixture comparisons; do not broaden to browser-backed checks for aggregate provenance coverage. | Static fixture gate passes; broader gates pending. | No headline increase; protects source-search read-target provenance reporting. |
+| A341 | Source-search selected-engine readiness guard | 100% | Readiness audit now requires `sourceSearchSelectedEngine` in full, brief, and public shortcut evidence lists, so auto-search handoff cannot lose the final chosen engine while retaining only requested engine/attempt counts. | Keep selected-engine coverage aligned with future source-search recovery fields. | Readiness audit passes; broader gates pending. | No headline increase; protects auto-search recovery routing. |
+| A342 | Source-search selected-result readiness guard | 100% | Readiness audit now requires selected-result host/source, command/args, source score, relevance, likely-official flag, and selection reason across full, brief, and public shortcut evidence lists, so subagents can open and judge the chosen result without nested parsing. | Keep this shallow selected-result guard aligned with source-search fixture coverage; add alternate-result guards only if a recovery fixture proves they are needed. | Readiness audit passes; broader gates pending. | No headline increase; protects source-search selection routing. |
+| A343 | Source-search alternate-result readiness guard | 100% | Readiness audit now requires alternate-result host/source, command/args, source score, relevance, likely-official flag, and reason across full, brief, and public shortcut evidence lists, so failed selected-result recovery can use the next result without nested parsing. | Keep this focused on the first alternate result; add deeper alternate arrays only if a fixture proves shallow recovery is insufficient. | Readiness audit and focused CLI fixture pass; broader gates pending. | No headline increase; protects source-search recovery routing. |
+| A344 | Static-readiness reason-code readiness guard | 100% | Readiness audit now requires static-readiness reason code plus read-target kind/count/score/primary/reason across full, brief, and public shortcut evidence lists, so browser fallback decisions stay machine-readable instead of relying on prose. | Keep this guard aligned with existing static-readiness fixtures; add new reason codes only when a fixture proves current categories are ambiguous. | Focused CLI text test and readiness audit pass; broader gates pending. | No headline increase; protects page-check/browser-fallback routing. |
+| A345 | Browser fallback command readiness guard | 100% | Readiness audit now requires top-level browser HTML fallback action, operation, URL, capture file/script, command, and command args across full, brief, and public shortcut evidence lists, so capture handoff stays copy-ready. | Keep after-interaction command guards separate until a fixture emits those top-level fields. | Focused fallback fixture and readiness audit pass; broader gates pending. | No headline increase; protects browser-capture routing. |
+| A346 | After-interaction command readiness guard | 100% | Readiness audit now requires next, execution-plan, executor, handoff, answer-plan, and primary after-interaction command shortcuts across full, brief, and public evidence lists, so live-browser follow-up routing cannot silently lose copy-ready rerun commands. | Keep this guard on shallow command/args fields that existing fixtures already emit. | Focused public type test and readiness audit pass; broader gates pending. | No headline increase; protects live-browser interaction recovery. |
+| A347 | Action command readiness guard | 100% | Readiness audit now requires top-action, primary-action, and alternate-action command shortcuts where matching command-args fields were already guarded, so copy-ready action execution cannot silently degrade into args-only metadata. | Keep this tied to existing shallow action command fields; add page-check text-only command guards only when a fixture emits them as agent payload fields. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects action execution routing. |
+| A348 | PageCheck browser-capture text guard | 100% | Text-mode challenge output now has a focused regression test for pageCheck command, command args, step command, step command args, and browserHtml command rows, so agents reading plain text can rerun the browser-capture path without parsing the larger agent block. | Keep this mock-only through stdin challenge HTML; do not use live browser checks for this parity guard. | Focused CLI text test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects text-only browser-capture rerouting. |
+| A349 | Recommended command readiness guard | 100% | Readiness audit now requires `recommendedCommand` alongside `recommendedCommandArgs` in full, brief, and public shortcut evidence lists, so copy-ready recommended actions cannot degrade into args-only metadata. | Keep this paired with existing recommended-command fixtures; expand only when a fixture proves another recommended-action shortcut is missing. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects recommended-action execution routing. |
+| A350 | Command shortcut symmetry guard | 100% | Readiness audit now checks that implemented `*CommandArgs` shortcuts have matching `*Command` coverage for source-search failure, top result/source/generic choices, search-decision official picks, next/executor/handoff routing, and structured page resources. | Keep this as a static audit symmetry check; add behavior tests only when a new command shortcut is emitted by a fixture without text/JSON coverage. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects copy-ready routing across search and page-check shortcut families. |
+| A351 | Comparison-gate metric readiness guard | 100% | Readiness audit now requires the comparison gate to keep source-search provenance, top source-choice shortcuts, recommended metadata, browser-advantage, hidden-command, search/page decision, semantic summary, barrier, and structured shortcut score floors. | Keep this aligned with `scripts/check-comparison-gates.ts`; add new gate metrics to readiness when they become part of the browser-accessibility-tree advantage claim. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects the benchmark proof that ax-grep is useful beyond raw accessibility-tree overlap. |
+| A352 | Automatic command shortcut symmetry audit | 100% | Readiness audit now scans its own guarded `*CommandArgs` shortcuts against implemented `*Command` fields in `src/cli.ts`, failing if a copy-ready command string is omitted while args are guarded. | Keep this static guard in readiness so future shortcut additions cannot regress into args-only coverage. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; prevents repeated command/args drift across search and page-check routing shortcuts. |
+| A353 | Automatic after-interaction shortcut symmetry audit | 100% | The same readiness symmetry audit now covers `*AfterInteractionCommandArgs` shortcuts, so live-browser follow-up commands cannot drift into args-only coverage when new routing fields are added. | Keep this paired with the command shortcut audit; expand suffix coverage only when another executable command family appears. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; prevents live-browser rerun command drift across search and page-check recovery shortcuts. |
+| A354 | Command symmetry audit self-guard | 100% | Readiness audit now requires its own command-symmetry audit call and both `CommandArgs` and `AfterInteractionCommandArgs` suffix scans, so the regression guard cannot be removed silently while the broader audit still passes. | Keep this as a self-guard in `scripts/check-agent-readiness.ts`; update it when executable shortcut families change. | Readiness audit, typecheck, README test, and diff check pass. | No headline increase; protects the guard that protects copy-ready search/page-check command routing. |
+| A355 | Command symmetry audit behavior test | 100% | `tests/agent-readiness.test.ts` now builds a minimal fake project where readiness guards `fooCommandArgs` while `src/cli.ts` implements `fooCommand`, and asserts the audit fails with the expected args-only shortcut message. | Keep this as a mock-only unit test; add more fixture cases only if another executable shortcut family needs custom symmetry behavior. | Focused agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; proves the self-guard catches real command/args drift. |
+| A356 | After-interaction symmetry behavior test | 100% | `tests/agent-readiness.test.ts` now also builds a fake project where readiness guards `fooAfterInteractionCommandArgs` while `src/cli.ts` implements `fooAfterInteractionCommand`, proving the audit catches args-only live-browser rerun shortcuts. | Keep this mock-only beside the generic command symmetry test. | Focused agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; proves live-browser recovery command drift is caught by behavior, not only static text. |
+| A357 | Unguarded command shortcut audit | 100% | Readiness audit now scans public `src/cli.ts` command shortcut fields and fails when a matching `*Command`/`*CommandArgs` or `*AfterInteractionCommand`/`*AfterInteractionCommandArgs` pair exists without readiness coverage. | Keep the tracked command shortcut coverage list aligned whenever a new public executable shortcut is added. | Focused agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; prevents new copy-ready command shortcuts from landing outside the progress/readiness guard surface. |
+| A358 | Browser fixture ledger validation guard | 100% | Browser fixture ledger tests now require every local browser parity check `ledgerId` to resolve to a documented observed-gap row with `Landed` status and validation evidence, and readiness audit requires that guard. | Keep local browser fixture checks paired with landed G-rows before using them as progress evidence. | Browser fixture ledger test, focused agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; strengthens the evidence chain behind browser accessibility-tree parity claims. |
+| A359 | Owned grid rowgroup browser parity fixture | 100% | Added a fifteenth local browser parity fixture for a `grid[aria-owns]` virtual rowgroup, proving browser-side semantic extraction and `--agent-brief` agree on owned rowgroup identity, owned cell text, row/column indices, and `ownedTarget` shortcuts. | Add deeper owned-row sampling only if a future comparison shows the first owned sample cell is insufficient. | Pre/post process check, browser fixture gate, browser fixture ledger test, readiness audit, typecheck, README test, diff check, and final process check pass. | P5 table/grid parity 85% -> 88%; overall remains 99.2%. |
+| A360 | Progress ledger freshness guard | 100% | Research milestones and scope ledger rows now use `Watch` for observed browser fallback, table/grid, search provenance, and browser-gap tracking work that has landed evidence, while readiness audit rejects the stale `Planned`/`In progress` wording for those rows. | Keep milestone status honest: use `Watch` for landed-but-expandable research, and reserve `In progress` for active unfinished implementation. | Focused agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves progress visibility for the long-running goal. |
+| A361 | Semantic target selector readiness guard | 100% | Readiness audit now requires the high-value semantic target selector and controls-target shortcut families for interactive, focusable, image, in-page link, field relation, value, choice, selected-choice, and state outputs so subagents keep browser-tree-like jump targets across compact, brief, and text surfaces. | Keep this guard focused on already-exposed shallow target fields; add new semantic signals only when a fixture shows agents still need nested payloads. | Focused agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; reduces regression risk for accessibility-tree targeting ergonomics. |
+| A362 | Nested source-search choice URL parts | 100% | Agent search-result targets now carry `urlPath` and `urlQuery`, and text output prints `sourceSearchAlternateChoiceUrlPath`/`UrlQuery` so subagents can compare second-or-later failed-open recovery choices without parsing full URLs. | Keep this limited to already-known result URLs; add deeper provenance only if a fixture shows path/query plus source score is still insufficient. | Focused CLI/public type tests, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves failed-open recovery ergonomics. |
+| A363 | Executor/handoff/primary target URL parts | 100% | `executorTarget*`, `handoffTarget*`, and `primaryTarget*` shortcuts now expose `UrlPath` and optional `UrlQuery` in full, brief, text, public types, and readiness guards, so subagents can distinguish same-host execution targets without parsing full URLs or nested target payloads. | Keep canonical URLs on the existing `*TargetUrl` fields; path/query fields are shallow routing aids only. | Focused CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; closes a small execution-stage handoff ergonomics gap. |
+| A364 | Active descendant field target shortcuts | 100% | Top semantic field shortcuts now expose `semanticTopFieldActiveDescendantTarget*` in full, brief, text, public types, and readiness guards, so autocomplete/searchbox agents can target the currently active option without parsing `relationItems` or browser accessibility-tree relations. | Keep this field-scoped; add interactive/focusable aliases only if fixtures show non-field active-descendant routing needs them. | Focused CLI relation/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves combo/listbox accessibility handoff. |
+| A365 | Flow-to relation target shortcuts | 100% | Agent summaries now expose `semanticTopFlowTo*` in full, brief, text, public types, and readiness guards, so subagents can follow `aria-flowto` reading/navigation hints even when another relation, such as `controls`, is the first relation item. | Keep this relation-specific; add more relation aliases only when fixtures show agents still need to parse nested relation arrays. | Focused CLI relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves accessibility-tree relation handoff ergonomics. |
+| A366 | Details and error-message relation shortcuts | 100% | Agent summaries now expose `semanticTopDetailsRelation*` and `semanticTopErrorMessageRelation*` in full, brief, text, public types, and readiness guards, so subagents can jump to explanatory or error targets on non-field controls without parsing nested relation arrays. | Keep field-specific details/error shortcuts separate from generic relation shortcuts; add more relation aliases only when browser-tree fixtures show repeated routing value. | Focused CLI relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves browser accessibility-tree relation parity for explanatory/error targets. |
+| A367 | Owns relation target shortcuts | 100% | Agent summaries now expose `semanticTopOwnsRelation*` in full, brief, text, public types, and readiness guards, so subagents can follow `aria-owns` from non-table controls to owned menu/listbox/region targets without parsing nested relation arrays. | Keep table/grid ownership shortcuts separate from generic relation shortcuts; add deeper owned-target sampling only when browser-tree fixtures show repeated value. | Focused CLI relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves browser accessibility-tree ownership parity for non-table controls. |
+| A368 | Described-by relation target shortcuts | 100% | Agent summaries now expose `semanticTopDescribedByRelation*` in full, brief, text, public types, and readiness guards, so subagents can jump from non-field controls to `aria-describedby` help text targets without parsing nested relation arrays. | Keep this generic shortcut separate from field-specific described-by fields; add labelled-by relation shortcuts only if fixture evidence shows agents need label-target navigation outside existing names. | Focused CLI relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves browser accessibility-tree description-target parity for non-field controls. |
+| A369 | Labelled-by relation target shortcuts | 100% | Agent summaries now expose `semanticTopLabelledByRelation*` in full, brief, text, public types, and readiness guards, so subagents can jump from non-field controls to `aria-labelledby` label targets without parsing nested relation arrays. | Keep this generic shortcut separate from field-specific labelled-by fields; add more label provenance only when fixtures show the computed accessible name is insufficient. | Focused CLI relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves browser accessibility-tree label-target parity for non-field controls. |
+| A370 | Active-descendant relation target shortcuts | 100% | Agent summaries now expose `semanticTopActiveDescendantRelation*` in full, brief, text, public types, and readiness guards, so subagents can jump from non-field composites like grids to the current `aria-activedescendant` target without parsing nested relation arrays. | Keep field-specific active-descendant shortcuts separate; add more composite aliases only if browser-tree fixtures show repeated non-field routing gaps. | Focused CLI active-descendant relation/text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; improves browser accessibility-tree active-target parity for composite widgets. |
+| A371 | Second owned table sample-cell shortcuts | 100% | Agent summaries now expose `semanticTopTableSecondOwnedSampleCell*` in full, brief, text, public types, and readiness guards, so subagents can compare a second virtualized/`aria-owns` table or grid cell without parsing `sampleCellRefs`. | Keep this to the second owned sample; add deeper owned-row arrays only if a browser fixture proves the shallow pair is insufficient. | Focused CLI owned-table text test, public type test, agent-readiness test, readiness audit, typecheck, README test, and diff check pass. | No headline increase; reduces P5 nested parsing for virtualized table/grid handoff. |
+| A372 | Second source-search alternate shortcuts | 100% | Agent summaries now expose `sourceSearchSecondAlternate*` in full, compact, brief, text, public types, and readiness guards so failed-open recovery agents can compare the next fallback candidate without parsing `sourceSearchAlternateChoices`. | Keep this limited to routing and quality fields; deeper date/sitelink shortcuts stay nested unless a fixture shows agents need them. | Focused source-search CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; strengthens failed-open recovery ergonomics without lengthening README. |
+| A373 | Second source-search alternate recency shortcuts | 100% | `sourceSearchSecondAlternate*` now also exposes date, matched-term, and find-match shortcuts across full, compact, brief, text, and public types; readiness guards require the second-alternate date fields so failed-open recovery can compare fallback freshness without parsing nested alternate choices. | Keep deeper second-alternate sitelinks nested unless a fixture shows command-level recovery still needs them. | Focused source-search CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves latest/current search recovery confidence. |
+| A374 | Second source-search alternate sitelink shortcuts | 100% | `sourceSearchSecondAlternateFirstSitelink*` now mirrors first-alternate sitelink title, URL parts, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so recovery agents can open a second fallback's relevant subpage without parsing nested sitelinks. | Keep this to the first sitelink already surfaced for selected/alternate results; add deeper sitelinks only if fixture evidence shows repeated need. | Focused source-search brief CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves failed-open subpage recovery ergonomics. |
+| A375 | Second page resource shortcuts | 100% | Agent summaries now expose `secondResource*` for the second page resource URL, URL parts, selector, type metadata, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare alternate feeds/AMP/documents without parsing `pageCheck.resources`. | Keep this to the second resource; add deeper resource arrays only if a fixture shows repeated need beyond the existing compact resource list. | Focused resource CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves page-check resource routing ergonomics. |
+| A376 | Second page media shortcuts | 100% | Agent summaries now expose `secondMedia*` for the second page media URL, URL parts, selector, alt/caption/title/dimensions, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare visual evidence without parsing `pageCheck.media`. | Keep this to the second media item; add deeper media arrays only if fixture evidence shows repeated need beyond compact media lists. | Focused media CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves page-check media routing ergonomics. |
+| A377 | Second page embed shortcuts | 100% | Agent summaries now expose `secondEmbed*` for the second embed URL, URL parts, title, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can recover from unavailable iframes by opening the next video/object/embed candidate without parsing `pageCheck.embeds`. | Keep this to routing metadata for the second embed; poster/source arrays remain in `pageCheck.embeds` until fixture evidence shows shallow aliases are needed. | Focused embed CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves embedded-content fallback routing. |
+| A378 | Second page transcript shortcuts | 100% | Agent summaries now expose `secondTranscript*` for the second caption/subtitle/transcript URL, URL parts, label, language, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare alternate language/audio transcript candidates without parsing `pageCheck.transcripts`. | Keep this to the second transcript item; deeper transcript arrays remain nested unless language-selection fixtures show repeated need. | Focused transcript CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves transcript and captions routing ergonomics. |
+| A379 | Second page pagination shortcuts | 100% | Agent summaries now expose `secondPagination*` for the second pagination URL, URL parts, label, current flag, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare prev/next/page candidates and open the next evidence route without parsing `pageCheck.pagination`. | Keep this to the second pagination item; deeper page lists remain nested unless navigation fixtures show repeated need. | Focused pagination CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves paginated search and page-check navigation. |
+| A380 | Second page provenance shortcuts | 100% | Agent summaries now expose `secondProvenance*` for the second citation/provenance identifier value, URL parts, source, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can cross-check DOI/PMID/arXiv-style evidence without parsing `pageCheck.provenance`. | Keep this to the second provenance item; deeper identifier lists remain nested unless citation verification fixtures show repeated need. | Focused provenance CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves citation and provenance trust checks. |
+| A381 | Second page author-link shortcuts | 100% | Agent summaries now expose `secondAuthorLink*` for the second author/profile URL, URL parts, source, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare co-author or alternate profile evidence without parsing `pageCheck.authorLinks`. | Keep this to the second author link; deeper author lists remain nested unless source/identity fixtures show repeated need. | Focused author-link CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves source and authorship trust checks. |
+| A382 | Second page offer shortcuts | 100% | Agent summaries now expose `secondOffer*` for the second structured offer price, URL parts, availability, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare product plans or alternate purchase routes without parsing `pageCheck.offers`. | Keep this to the second offer item; deeper offer lists remain nested unless pricing comparison fixtures show repeated need. | Focused offer CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves structured offer comparison and routing. |
+| A383 | Second page identity shortcuts | 100% | Agent summaries now expose `secondIdentity*` for the second structured identity name, URL/logo/sameAs path parts, commands, source, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare page/site ownership or publisher candidates without parsing `pageCheck.identities`. | Keep this to the second identity item; deeper identity lists remain nested unless source-credibility fixtures show repeated need. | Focused identity CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves source credibility and ownership routing. |
+| A384 | Second contact-point shortcuts | 100% | Agent summaries now expose `secondContactPoint*` for the second contact/support route label, value, URL path, command, source, and selector across full, compact, brief, text, public types, and readiness guards so agents can open support/contact URLs when the top contact fact is a phone or email. | Keep commands limited to HTTP(S) contact URLs; `mailto:` and `tel:` remain non-fetch command targets. | Focused contact CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves support/contact routing without browser handoff. |
+| A385 | Second timeline freshness shortcuts | 100% | Agent summaries now expose `secondTimeline*` for the second publication/update/event date path, kind, label, raw value, normalized ISO date, epoch milliseconds, source, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare published and modified dates without parsing `pageCheck.timeline`. | Keep this to the second timeline fact; deeper date lists remain nested unless freshness fixtures show repeated need. | Focused timeline CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves freshness and update-date routing. |
+| A386 | Second dataset download shortcuts | 100% | Agent summaries now expose `secondDataset*` for the second dataset/data-download name, URL, first distribution URL, license route, encoding format, coverage, creator, selector, command, and command args across full, compact, brief, text, public types, and readiness guards so agents can compare dataset downloads without parsing `pageCheck.datasets`. | Keep this to the second dataset item; deeper dataset lists remain nested unless data-source fixtures show repeated need. | Focused dataset CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves data-source and download routing. |
+| A387 | Second FAQ answer shortcuts | 100% | Agent summaries now expose `secondFaq*` for the second FAQ path, question, answer, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare adjacent Q/A pairs without parsing `pageCheck.faqs`. | Keep this to the second FAQ item; deeper FAQ lists remain nested unless FAQ-answer fixtures show repeated need. | Focused FAQ CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves FAQ answer routing. |
+| A388 | Second breadcrumb trail shortcuts | 100% | Agent summaries now expose `secondBreadcrumb*` for the second breadcrumb trail path, text, source, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare alternate JSON-LD/HTML navigation trails without parsing `pageCheck.breadcrumbs`. | Keep this to the second trail; deeper breadcrumb arrays remain nested unless navigation fixtures show repeated ambiguity. | Focused breadcrumb CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves navigation hierarchy handoff. |
+| A389 | Second content-section shortcuts | 100% | Agent summaries now expose `secondSection*` for the second heading-grouped section path, heading, level, text, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare adjacent content blocks without parsing `pageCheck.sections`. | Keep this to the second section; deeper section arrays remain nested unless long-form page fixtures show repeated ambiguity. | Focused section CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves long-form page evidence handoff. |
+| A390 | Second code-block shortcuts | 100% | Agent summaries now expose `secondCodeBlock*` for the second code block path, language, line count, text, and selector across full, compact, brief, text, public types, and readiness guards so agents can compare install/use command snippets without parsing `pageCheck.codeBlocks`. | Keep this to the second code block; deeper snippets remain nested unless command-heavy docs fixtures show repeated ambiguity. | Focused code-block CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves command and example handoff. |
+| A391 | Second citation shortcuts | 100% | Agent summaries now expose `secondCitation*` for the second citation id, path, kind, text, title, URL parts, command, confidence, reason, and score across full, compact, brief, text, public types, and readiness guards so agents can compare source evidence without parsing `agent.citations`. | Keep answer-evidence-specific second shortcuts separate because evidence ranking can differ from citation order. | Focused citation text test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves source verification handoff. |
+| A392 | Second answer-evidence shortcuts | 100% | Agent summaries now expose `secondAnswerEvidence*` for the second answer evidence id, path, kind, text, title, URL parts, command, confidence, reason, and score across full, compact, brief, text, public types, and readiness guards so answer subagents can compare selected evidence without parsing `agent.answerEvidence`. | Keep this evidence-ranked rather than citation-ranked; `secondCitation*` remains the generic source evidence comparison path. | Focused alignment CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves answer handoff evidence comparison. |
+| A393 | Second read-target shortcuts | 100% | Agent summaries now expose `secondReadTarget*` for the second read-target path, kind, count, score, primary flag, and reason across full, compact, brief, text, public types, and readiness guards so subagents can compare the next static read source without parsing `agent.readTargets`. | Keep this to the second target; deeper read routing remains in `agent.readTargets`. | Focused text-output CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves read-current fallback comparison. |
+| A394 | Second source-choice shortcuts | 100% | Agent summaries now expose `secondSourceChoice*` for the second page source-choice path, URL parts, command, rank, text, selector, source quality, official flag, primary flag, and reason across full, compact, brief, text, public types, and readiness guards so source-check subagents can compare a backup source link without parsing `agent.sourceChoices`. | Keep this to the second source choice; deeper source-link ranking remains in `agent.sourceChoices`. | Focused page-check CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves source verification fallback routing. |
+| A395 | Second result-choice shortcuts | 100% | Agent summaries now expose `secondResultChoice*` for the second search result-choice path, URL parts, command, rank/open-result, source quality, freshness, relevance, official flag, first sitelink, and reason across full, compact, brief, text, public types, and readiness guards so search subagents can compare a backup result without parsing `agent.resultChoices`. | Keep this to the second result choice; deeper result ranking remains in `agent.resultChoices`. | Focused brief alignment CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves search fallback routing. |
+| A396 | Second form-choice shortcuts | 100% | Agent summaries now expose focused `secondFormChoice*` fields for the second form path, action URL parts, URL template, query field, command, field counts, selector, submit control, and first input across full, compact, brief, text, public types, and readiness guards so page-check subagents can compare a backup search form without parsing `agent.formChoices`. | Keep this to execution-critical fields; full required/hidden-field detail stays in `agent.formChoices` unless fixture evidence shows agents need more top-level aliases. | Focused brief alignment CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves multi-form page-check routing. |
+| A397 | Second action-target shortcuts | 100% | Agent summaries now expose `secondActionTargetChoice*` for the second structured action target path, route URL parts, URL template, query input, method, encoding, command, state flags, controls, and selector across full, compact, brief, text, public types, and readiness guards so page-check subagents can compare backup JSON-LD/OpenSearch actions without parsing `agent.actionTargetChoices`. | Keep this to the second action target; deeper action metadata remains in `agent.actionTargetChoices`. | Focused action-target CLI tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves structured action fallback routing. |
+| A398 | Second barrier shortcuts | 100% | Agent summaries now expose `secondBarrier*` for the second page barrier kind, severity, source, path, text, selector, and diagnostic code across full, compact, brief, text, public types, and readiness guards so browser-fallback subagents can distinguish multiple blockers without parsing `pageCheck.barriers`. | Keep this to the second barrier; full barrier arrays remain in `pageCheck.barriers` for deeper diagnostics. | Focused barrier CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves blocked-page fallback reasoning. |
+| A399 | Second data-table shortcuts | 100% | Agent summaries now expose `secondDataTable*` for the second table path, caption, dimensions, headers, sample rows/cells, and selector across full, compact, brief, text, public types, and readiness guards so table-reading subagents can compare multiple data tables without parsing `pageCheck.dataTables`. | Keep this to table summary fields; detailed cell provenance remains in `pageCheck.dataTables` and semantic table refs. | Focused data-table CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves multi-table page-check routing. |
+| A400 | Second TOC shortcuts | 100% | Agent summaries now expose `secondToc*` for the second table-of-contents path, title, item count, text, first item URL parts, command, and selector across full, compact, brief, text, public types, and readiness guards so document-navigation subagents can compare alternate TOC blocks without parsing `pageCheck.toc`. | Keep this to the second TOC and first item route; deeper TOC item traversal remains in `pageCheck.toc`. | Focused TOC CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves multi-TOC page-check routing. |
+| A401 | Second API endpoint shortcuts | 100% | Agent summaries now expose `secondApiEndpoint*` for the second hidden API endpoint path, kind, method, URL parts, safe GET command, and selector across full, compact, brief, text, public types, and readiness guards so app-shell subagents can compare backup API routes without parsing `pageCheck.apiEndpoints`. | Keep command generation limited to HTTP(S) GET-like endpoints; non-GET methods remain visible but do not get replay commands. | Focused API endpoint CLI test, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves hidden app/API handoff without browser capture. |
+| A402 | Second hydration shortcuts | 100% | Agent summaries now expose `secondHydration*` for the second hydration/data endpoint path, kind, label, URL parts, command, and selector across full, compact, brief, text, public types, and readiness guards so app-shell subagents can compare preload/page-data routes without parsing `pageCheck.hydration`. | Keep this to fetchable hydration/data URLs; deeper hydration details remain in `pageCheck.hydration`. | Focused hydration CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves hidden app data handoff without browser capture. |
+| A403 | Second client-state shortcuts | 100% | Agent summaries now expose `secondClientState*` for the second client storage/cookie hint path, kind, operation, key, and selector across full, compact, brief, text, public types, and readiness guards so app-shell subagents can compare state dependencies without parsing `pageCheck.clientState`. | Keep this metadata-only; actual storage values are not inferred from static scripts. | Focused client-state CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves hidden state handoff without browser capture. |
+| A404 | Second runtime shortcuts | 100% | Agent summaries now expose `secondRuntime*` for the second runtime script hint path, kind, URL parts, command, and selector across full, compact, brief, text, public types, and readiness guards so app-shell subagents can compare worker/import/runtime routes without parsing `pageCheck.runtime`. | Keep this to discovered runtime URLs; runtime execution behavior still requires browser validation when static hints are insufficient. | Focused runtime CLI JSON/text tests, public type test, agent-readiness test, readiness audit, typecheck, README test, diff check, and process check pass. | Overall holds at 99.2%; improves hidden runtime handoff without browser capture. |
+| A179 | Verification command text parity | 100% | The `verification` text section now mirrors recommended follow-up command fields as `verificationCommand*` lines, so agents handling partial or missing verification can execute the recovery route without parsing generic `command` lines. | Keep this text-only mirror aligned with existing verification recommended-action command fields. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves verification recovery routing. |
+| A180 | Find-match quality text parity | 100% | The `finds` text section now prints existing match `source`, `score`, `quality`, and `qualityReason` metadata, so verification agents can judge whether a match came from semantic evidence or fallback text without switching to JSON. | Keep this limited to existing find match quality fields. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode verification confidence checks. |
+| A181 | Results quality text parity | 100% | The `results` text section now prints existing relevance, official, matched-term, find-match, date-detail, and selection-reason metadata, so agents comparing search results can understand ranking and trust signals without nested result JSON. | Keep this limited to existing result metadata. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-mode result comparison. |
+| A182 | PageCheck link quality text parity | 100% | The `pageCheck` `link` and `sourceLink` text lines now print existing source type, score, hints, and official flags, so agents reading only the page-check block can judge source-link trust without nested link JSON. | Keep this limited to existing page-link source metadata. | Focused CLI text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check-only source-link comparison. |
+| A183 | Top-level link metadata text parity | 100% | The top-level `links` text section now prints each link's role, selector, and snippet when available, so agents can choose visible targets without falling back to JSON or a browser accessibility tree. | Limited to metadata already present in `LinkSummary`; deeper link quality scoring remains separate. | Focused CLI text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-only link routing. |
+| A184 | PageCheck action selector text parity | 100% | `pageCheck` action text lines now include each action selector when available, so agents can identify interactive targets from the page-check block without switching to JSON or browser tree inspection. | Limited to selectors already present in `ActionSummary`; no new action ranking semantics. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page-check-only action targeting. |
+| A185 | Top-level action selector text parity | 100% | The top-level `actions` text section now includes each action selector when available, so agents can identify visible controls from plain text output without inspecting JSON or browser trees. | Limited to selectors already present in `ActionSummary`; no new action ranking semantics. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-only action targeting. |
+| A186 | Outline selector parity | 100% | Top-level `outline` JSON and text output now preserve heading selectors, so agents can locate structural headings without descending into semantic summary details or browser trees. | Limited to selectors already present on semantic heading nodes. | Focused CLI text/JSON tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text and JSON outline targeting. |
+| A187 | Top-level content selector text parity | 100% | The top-level `content` text section now includes each content item's role and selector when available, so agents can locate readable excerpts without switching to pageCheck evidence, JSON, or browser trees. | Limited to metadata already present in `ContentSummary`; content scoring remains unchanged. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-only content targeting. |
+| A188 | Search result sitelink selector parity | 100% | Search result sitelinks now preserve anchor selectors in JSON and text output, so agents can target result sublinks directly without reopening the browser tree. | Limited to selectors already available during search-card extraction; sitelink scoring remains unchanged. | Focused CLI JSON/text tests, public typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves search result sublink targeting. |
+| A189 | Sitelink selector shortcut parity | 100% | Top result/source-search first-sitelink shortcuts now include selector fields in full, brief, text, and public types, so agents can target sublinks without parsing nested sitelink arrays. | Limited to the first sitelink shortcut that already exposes title and URL. | Focused CLI/public type tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves shallow search/source handoff targeting. |
+| A190 | Choice sitelink selector text parity | 100% | Result/source-search choice text lines now print first-sitelink selectors, and brief source-search recovery shortcuts now preserve those selector fields, so text-only recovery loops can target sublinks without nested arrays. | Limited to the first sitelink selector already surfaced by A189. | Focused CLI JSON/text tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves choice-list handoff targeting. |
+| A191 | Source choice selector text parity | 100% | Page source-choice text lines now print source-link selectors, so agents reading choice lists can target the chosen source anchor without parsing nested source choices or browser trees. | Limited to selectors already present on `AgentSourceChoice`. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page source-link text targeting. |
+| A192 | List item ref selector text parity | 100% | `semanticTopList` text output now prints selectors for each shallow list item ref, so agents can target visible list/tree/menu items directly instead of relying only on the first-item shortcut or browser tree. | Limited to selectors already present on `semanticTopListItemRefs`. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-only list item targeting. |
+| A193 | Owned table ref selector text parity | 100% | `semanticTopTable` text output now prints selectors for shallow `ownedRefs`, so agents can target `aria-owns` rowgroups and virtualized table regions directly from text output instead of requiring JSON or browser-tree inspection. | Limited to selectors already present on `semanticTopTableOwnedRefs`. | Focused CLI text test, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves text-only owned table targeting. |
+| A194 | Top semantic structure selector shortcuts | 100% | Top heading, landmark, and named-role shortcuts now preserve selectors in full, brief, text, and public types, so agents can jump to the main structural accessibility targets without parsing nested semantic arrays or browser trees. | Limited to selectors already present on the selected top semantic items. | Focused CLI/public type tests, typecheck, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves shallow semantic structure targeting. |
+| A195 | Readiness audit semantic selector coverage | 100% | The readiness audit now requires top heading, landmark, and named-role selector shortcut evidence in both public types and CLI tests, so shallow semantic targeting regressions are caught by the standard readiness gate. | Limited to the A194 selector shortcuts already present in code and tests. | Readiness audit, focused readiness test, typecheck, README test, diff check, and process check pass. | No estimate increase; improves durability of selector shortcut coverage. |
+| A196 | Field idref target selector shortcuts | 100% | Top semantic field shortcuts now preserve selectors for `aria-labelledby`, `aria-describedby`, `aria-details`, and `aria-errormessage` targets in full, brief, text, and public types, so agents can jump from a field to its label/help/details/error nodes without browser-tree inspection. | Limited to the first resolved idref target selector already available in the static semantic map. | Focused CLI/public type tests, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; improves field relation targeting. |
+| A197 | Top state controls target shortcuts | 100% | Top semantic state shortcuts now preserve the controlled target role, name, and selector for resolved `aria-controls` relations in full, brief, text, and public types, so agents can jump from stateful controls to controlled panels/listboxes without parsing relation arrays or browser trees. | Limited to `controls` relations already resolved in `semanticSummary.relationItems`. | Focused CLI/public type tests, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; improves stateful control target routing. |
+| A198 | Top control relation target shortcuts | 100% | Top interactive, focusable, button, and field shortcuts now preserve resolved `aria-controls` target role, name, and selector in full, brief, text, and public types, so agents can jump from common controls to controlled regions without parsing relation arrays or browser trees. | Limited to controls relations already resolved in `semanticSummary.relationItems`. | Focused CLI/public type tests, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; improves controlled target routing consistency. |
+| A199 | Control target text coverage | 100% | Text-mode tests now assert controlled target role/name/selector output for top interactive/button and state shortcuts, so plain-text subagent handoff keeps the same `aria-controls` target routing guarantees as JSON and brief output. | Test-only guard for A197/A198 text output. | Focused CLI text tests, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; improves regression coverage for text-only control routing. |
+| A200 | Static declarative shadow DOM extraction | 100% | Static extraction now reads semantic children from declarative shadow DOM templates and prunes inert custom-element wrappers, so Web Component SSR can expose buttons, headings, regions, and control state without requiring a browser accessibility tree first. | Limited to declarative `shadowrootmode`/`shadowroot` templates in static HTML; normal inert templates remain pruned. | Focused static extractor test, readiness audit, agent-readiness test, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component page-check parity without browser load. |
+| A201 | Declarative shadow agent brief coverage | 100% | `--agent-brief` now has a focused regression test proving declarative shadow DOM headings, buttons, controlled target role/name/selector, and inert template pruning reach the shallow handoff fields used by subagents. | Test-only guard for A200 agent-facing output. | Focused CLI test, readiness audit, typecheck, README test, diff check, and process check pass. | No estimate increase; protects Web Component handoff behavior. |
+| A202 | Declarative shadow slot projection | 100% | Static extraction now projects slotted light-DOM elements into declarative shadow DOM and suppresses unused fallback or unprojected light children, so agent brief output better matches the browser composed accessibility tree. | Limited to element children assigned through declarative shadow slots; text-only slot projection can be revisited if fixtures show it matters. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component parity and avoids duplicate fallback controls. |
+| A203 | Slotted text accessible names | 100% | Static extraction now uses assigned slot text when computing declarative shadow DOM accessible names, so button/link names in agent brief output match the composed tree instead of stale fallback text. | Limited to text and element assignments available in static declarative shadow DOM. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component action naming parity. |
+| A204 | Slotted IDREF selector preservation | 100% | Static extraction now preserves IDREF target nodes that would otherwise be pruned, so slotted labels/descriptions can still provide `semanticTopField*Selector` shortcuts in agent brief output. | Applies to common ARIA idref attributes such as labelledby, describedby, details, errormessage, controls, owns, flowto, and activedescendant. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component field repair and targeting parity. |
+| A205 | Slotted label-for field names | 100% | Static extraction now computes `<label for>` text with the active declarative shadow slot assignment, so slotted field labels become agent-facing field names instead of fallback text. | Limited to static/declarative shadow DOM labels where the label element is available in HTML. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component form handoff parity. |
+| A206 | Declarative shadow host action names | 100% | Static extraction now computes custom-element host accessible names from the declarative shadow composed tree, so host-level buttons/actions use projected slot text and ignore unprojected light DOM or fallback text. | Limited to declarative shadow DOM available in static HTML. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves Web Component action target parity. |
+| A207 | Shadow host description separation | 100% | Static name-from-content now excludes `aria-describedby`, `aria-details`, and `aria-errormessage` target text while keeping those targets available as descriptions, so host action names do not absorb help/error text. | Limited to common description-style ARIA IDREF targets in static/declarative shadow DOM. | Focused static and CLI tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves action name/description parity. |
+| A208 | Web Component static fixture gate | 100% | The non-browser `agent-fixtures` static comparison set now includes a declarative shadow DOM Web Component gate, so composed-tree button/field handoff regressions affect fixture scoring instead of only focused unit tests. | Synthetic fixture only; no browser or external fetch. | Static fixture comparison test, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves gate coverage for Web Component parity. |
+| A209 | First sitelink command shortcuts | 100% | Top result and source-search first-sitelink shortcuts now include copy-ready command and command-args fields, so agents can open a relevant result sublink directly without parsing nested sitelink arrays or composing commands themselves. | Limited to the first sitelink already exposed as a shallow shortcut. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves search and failed-open recovery handoff. |
+| A210 | Full CLI regression gate restored | 100% | The complete CLI regression file now matches the current direct-URL command policy, selector output, form-choice routing, and structured action-target text, so broad CLI validation can run cleanly again after shortcut changes. | Test expectation cleanup only; no runtime behavior change. | Full CLI test, readiness audit, README test, diff check, and process check pass. | No headline increase; improves validation confidence for search/page handoff work. |
+| A211 | Nested sitelink command preservation | 100% | Sitelink objects inside compact agent result/search-source choices now carry direct command and command-args fields, so agents iterating choice arrays can open sublinks without joining top-level shortcut fields or composing commands. | Raw non-agent search result arrays remain source data; command fields are added to compact agent choices. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; strengthens multi-choice search handoff. |
+| A212 | Source-search failure retry command shortcuts | 100% | Failed opened-result output now exposes `sourceSearchFailureCommand` and `sourceSearchFailureCommandArgs`, so recovery agents can retry the failed selected result from the failure block without reinterpreting selected-result shortcuts. | Alias of the selected result command; alternate-result recovery remains exposed separately. | Focused CLI/public type tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; strengthens failed-open recovery handoff. |
+| A213 | Search decision first-official shortcuts | 100% | Search decision output now exposes the first likely official result rank/path/title/url/source/score/relevance and command fields, so agents can compare the recommended result with an official candidate without parsing result arrays. | Limited to the first `isLikelyOfficial` result already detected by search scoring. | Focused CLI/public type tests, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; reduces official-source ambiguity in search handoff. |
+| A214 | Top provenance follow-up command shortcuts | 100% | Page-check output now exposes `topProvenanceCommand` and `topProvenanceCommandArgs` for DOI/identifier provenance URLs, so agents can open source identifiers directly from shallow page-check fields. | Limited to the top provenance item when it has a URL. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves provenance verification handoff without browser work. |
+| A215 | Top resource follow-up command shortcuts | 100% | Page-check output now exposes `topResourceCommand` and `topResourceCommandArgs` for feed, PDF, manifest, license, and download resources, so agents can open the top resource directly from shallow fields. | Limited to the top resource item when it has a URL. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves resource follow-up handoff without browser work. |
+| A216 | Top author follow-up command shortcuts | 100% | Page-check output now exposes `topAuthorLinkCommand` and `topAuthorLinkCommandArgs` for HTTP(S) author/profile links, so agents can verify author/source context directly from shallow fields. | Limited to the top author link when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves author/source verification handoff without browser work. |
+| A217 | Top transcript follow-up command shortcuts | 100% | Page-check output now exposes `topTranscriptCommand` and `topTranscriptCommandArgs` for HTTP(S) caption, subtitle, and transcript URLs, so agents can open the top transcript-like resource directly from shallow fields. | Limited to the top transcript item when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves media/transcript follow-up handoff without browser work. |
+| A218 | Top pagination follow-up command shortcuts | 100% | Page-check output now exposes `topPaginationCommand` and `topPaginationCommandArgs` for HTTP(S) next/previous/page URLs, so agents can continue paginated navigation from shallow fields. | Limited to the top pagination item when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves multi-page navigation handoff without browser work. |
+| A219 | Top embed follow-up command shortcuts | 100% | Page-check output now exposes `topEmbedCommand` and `topEmbedCommandArgs` for HTTP(S) iframe/object/embed/media URLs, so agents can inspect embedded content from shallow fields without browser capture. | Limited to the top embed item when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves embedded-content handoff without browser work. |
+| A220 | Top dataset follow-up command shortcuts | 100% | Page-check output now exposes command and command-args shortcuts for the top dataset URL, distribution URL, and license URL, so agents can open dataset landing pages, data files, and license evidence directly from shallow fields. | Limited to the top dataset item and HTTP(S) URLs. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves dataset evidence handoff without browser work. |
+| A221 | Top offer follow-up command shortcuts | 100% | Page-check output now exposes `topOfferCommand` and `topOfferCommandArgs` for HTTP(S) offer URLs, so agents can open buy/detail links directly from shallow price and availability fields. | Limited to the top offer item when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves product/commerce handoff without browser work. |
+| A222 | Top identity follow-up command shortcuts | 100% | Page-check output now exposes command and command-args shortcuts for the top identity URL, logo URL, and first sameAs URL, so agents can verify canonical source, brand assets, and external profiles directly from shallow fields. | Limited to the top identity item and HTTP(S) URLs. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves identity/source verification handoff without browser work. |
+| A223 | Top media follow-up command shortcuts | 100% | Page-check output now exposes `topMediaCommand` and `topMediaCommandArgs` for HTTP(S) image/social preview/media URLs, so agents can open visual evidence directly from shallow media fields. | Limited to the top media item when it is fetchable HTTP(S). | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves visual evidence handoff without browser work. |
+| A224 | Top TOC follow-up command shortcuts | 100% | Page-check output now exposes `topTocFirstItemCommand` and `topTocFirstItemCommandArgs` for the first HTTP(S) table-of-contents item, so agents can jump directly to the first section link from shallow navigation fields. | Limited to the first item of the top TOC. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves in-page navigation handoff without browser work. |
+| A225 | Top contact point follow-up command shortcuts | 100% | Page-check output now exposes `topContactPointCommand` and `topContactPointCommandArgs` for HTTP(S) contact URLs, so agents can open contact/support pages directly from shallow contact facts. | Limited to the top contact point and only when its URL is fetchable HTTP(S); `mailto:` and `tel:` contact points deliberately do not produce commands. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves support/contact handoff without creating non-fetchable command shortcuts. |
+| A226 | Top citation and answer evidence follow-up commands | 100% | Agent output now exposes `topCitationCommand`/`topCitationCommandArgs` and `topAnswerEvidenceCommand`/`topAnswerEvidenceCommandArgs` for HTTP(S) evidence URLs, so subagents can open the strongest citation or answer evidence directly from shallow fields. | Limited to top citation/evidence items and fetchable HTTP(S) URLs. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves evidence verification handoff without browser work. |
+| A227 | Top API endpoint follow-up command shortcuts | 100% | Page-check output now exposes `topApiEndpointCommand` and `topApiEndpointCommandArgs` for fetchable GET-like API endpoints discovered in scripts, so agents can inspect hidden API evidence directly when the request is safe to replay as a simple fetch. | Limited to the top API endpoint, HTTP(S) URLs, and endpoints with no method or `GET`; `POST`, `DELETE`, and other non-GET methods deliberately do not produce commands. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves hidden API handoff beyond ordinary accessibility-tree visibility without unsafe method replay. |
+| A228 | Top hydration and app hint follow-up commands | 100% | Page-check output now exposes `topHydrationCommand`/`topHydrationCommandArgs` and `topAppHintCommand`/`topAppHintCommandArgs` for HTTP(S) hydration JSON, preload data, manifest, and app metadata URLs, so agents can inspect hidden app-shell resources directly. | Limited to top hydration/app hint items with fetchable HTTP(S) URLs. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; improves hidden app-data handoff beyond ordinary accessibility-tree visibility. |
+| A229 | Hidden command shortcut gate split-out | 100% | Static comparison summaries now expose and gate `averageAgentHiddenCommandShortcutScore`, separating hidden hydration/API/app-hint command coverage from broader hidden-signal shortcut scoring. | Covers hydration URL commands, safe GET-like API endpoint commands, and app hint URL commands in fixture/static gates. | Comparison gate tests, static fixture gate, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; makes hidden-resource command regressions fail loudly without browser-backed tests. |
+| A230 | Hidden metadata fixture hydration coverage | 100% | The synthetic hidden metadata comparison fixture now includes `__NEXT_DATA__` hydration JSON, so static fixture gates exercise hydration follow-up commands in addition to API and manifest hidden-resource commands. | Limited to browser-free fixture coverage; no runtime browser checks added. | Static fixture gate, comparison gate tests, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; strengthens evidence that hidden app-shell resources stay agent-actionable without agent-browser. |
+| A231 | Hidden command readiness audit hook | 100% | The readiness audit now requires `averageAgentHiddenCommandShortcutScore` to be documented in the comparison baseline, so the hidden-resource command gate cannot disappear silently from project evidence. | Documentation and audit coverage only; no runtime behavior change. | Agent-readiness test, readiness audit, README test, typecheck, diff check, and process check pass. | No headline increase; strengthens completion evidence for hidden-resource handoff gates. |
+| A232 | Media/source locator text readability | 100% | Text output now prints embed, transcript, and author-link locator fields as `id=`, `path=`, `kind=`, `source=`, and `url=` instead of positional tokens, including top-level page-check shortcuts. | Keep converting dense text rows only where agent handoff or human review is ambiguous. | Focused CLI test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves media/source follow-up readability. |
+| A233 | Resource/media locator text readability | 100% | Text output now prints resource and media locator fields as `id=`, `path=`, `kind=`, and `url=` instead of positional tokens, including top-level page-check shortcuts. | Continue converting dense text rows where the first tokens are ambiguous to subagents or hard to scan in logs. | Focused CLI test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves resource/media follow-up readability. |
+| A234 | Hidden app-shell locator text readability | 100% | Text output now prints hidden hydration, API endpoint, client-state, runtime, config, app-hint, mobile-hint, topic, key-value, meta-fact, HTTP-policy, schema-fact, and hidden-signal fields with named `id=`, `path=`, `kind=`, `method=`, `operation=`, and `url=` labels instead of positional tokens. | Continue converting remaining dense page-check text rows where positional values obscure agent handoff decisions. | Focused CLI test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves hidden-resource and app-shell handoff readability beyond accessibility-tree-visible content. |
+| A235 | Structured source locator text readability | 100% | Text output now prints provenance, offer, identity, dataset, timeline, and contact-point fields with named `id=`, `path=`, `kind=`, `name=`, `value=`, `price=`, and `url=` labels instead of positional tokens. | Continue converting remaining dense navigation/content rows where positional values slow down subagent handoff. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves source, commerce, freshness, and contact evidence handoff readability. |
+| A236 | Content navigation locator text readability | 100% | Text output now prints FAQ, code block, breadcrumb, section, pagination, table-of-contents, and citation fields with named `id=`, `path=`, `level=`, `heading=`, `label=`, `firstUrl=`, and `url=` labels instead of positional tokens. | Continue checking remaining text rows for positional locator ambiguity before declaring text-output cleanup done. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves answer, navigation, and citation handoff readability. |
+| A237 | Read-target action locator text readability | 100% | Text output now prints content evidence, data-table, barrier, form, and action-target rows with named `id=`, `path=`, `severity=`, `kind=`, `rows=`, `columns=`, and `url=` labels instead of positional locator tokens. | Remaining cleanup is mostly choice-summary text such as top search/action choices, which should be converted separately if ambiguity remains. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves primary read-target and executable action handoff readability. |
+| A238 | Choice routing locator text readability | 100% | Text output now prints top choice, answer evidence, result choices, source choices, form choices, and action-target choices with named `id=`, `path=`, `kind=`, `url=`, `actionUrl=`, and `targetUrl=` labels instead of positional tokens. | Continue scanning for isolated positional text rows after each routing change. | Focused CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves search/result/source/action routing handoff readability. |
+| A246 | Text locator format regression guard | 100% | Added a static text-output format test that fails if page-check rows or choice-routing rows go back to positional `id path` locator text instead of named `id=`/`path=` fields. | Extend the guard if future text-output formatter families add locator-bearing rows. | Text-output format test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; keeps recent text handoff readability improvements from silently regressing. |
+| A247 | Form field text shortcut parity | 100% | Text output now prints first hidden-field and first visible-field name/value/selected/selector shortcuts as separate `topFormChoiceFirst*` lines, so subagents can grep browser-submission inputs without parsing the aggregate form line. | Keep this to top form-choice submission-critical fields unless a fixture proves deeper field arrays need text shortcuts. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves form submission handoff without browser inspection. |
+| A248 | Form submit text shortcut parity | 100% | Text output now prints submit button type/name/value/disabled/selector and form-override shortcuts as separate `topFormChoiceSubmit*` lines, so subagents can reconstruct browser submission behavior without parsing the aggregate submit line. | Keep submit shortcut rows aligned with existing top form-choice fields; do not add synthetic submit metadata. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves form execution fidelity in text-only handoff. |
+| A249 | Nested action-target text shortcut parity | 100% | Nested `actionTargetChoice*` text output now mirrors target URL, URL template, query input, method, encoding, state, controls, and selector fields as separate prefix lines, so subagents can compare multiple executable action targets without parsing aggregate choice rows. | Keep nested shortcut rows limited to fields already present on compact action-target choices. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-action routing in text-only handoff. |
+| A250 | Nested form-choice text shortcut parity | 100% | Nested `formChoice*` text output now mirrors action URL, URL template, query field, field counts, first hidden field, form attributes, submit metadata, and selector fields as separate prefix lines, so subagents can compare multiple forms without parsing aggregate rows. | Keep nested shortcut rows limited to fields already present on compact form choices. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-form routing in text-only handoff. |
+| A251 | Nested result/source text shortcut parity | 100% | Nested `resultChoice*`, `handoffResultChoice*`, `sourceChoice*`, and `handoffSourceChoice*` text output now mirrors URL, title, rank, source, score, relevance, date, snippet, selector, and reason fields as separate prefix lines, so subagents can compare search/source candidates without parsing aggregate rows. | Keep nested shortcut rows limited to fields already present on compact result/source choices. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-result and source-link routing in text-only handoff. |
+| A252 | Nested result sitelink text shortcut parity | 100% | Nested `resultChoice*` text output now mirrors first-sitelink title, URL, selector, command, and command args as separate prefix lines, so subagents can open result sublinks without parsing aggregate rows or nested sitelink arrays. | Source-search result formatter uses the same prefix rows; add a focused source-search text fixture if a future gap shows nested source-search sitelinks are hard to inspect. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-result sublink routing in text-only handoff. |
+| A253 | Source-search sitelink text fixture coverage | 100% | Source-search recovery text tests now cover selected, alternate, alternate-choice, and handoff source-search first-sitelink title, URL, and command prefix rows. | Keep this as regression coverage for A252's shared source-search formatter path. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; protects failed-open recovery sublink routing in text-only handoff. |
+| A254 | Nested choice source-hints text parity | 100% | Nested `resultChoice*`, `sourceChoice*`, and `handoffSourceChoice*` text output now mirrors the full `sourceHints` array as `*SourceHints`, so agents comparing multiple search/source candidates do not lose provenance hints after leaving the top-choice shortcut path. | Keep this to already-computed source profile hints; add deeper per-hint scoring only if a fixture shows ordering or confidence is ambiguous. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-candidate provenance comparison in text-only handoff. |
+| A255 | Source-search result source-hints text parity | 100% | Nested source-search selected, alternate, alternate-choice, and handoff result text output now mirrors source profile hints as `*SourceHints`, so failed-open recovery agents can compare package-registry, documentation, mirror, or other provenance hints without parsing nested source-search objects. | Keep this aligned with existing source profile hints; do not add synthetic source categories without fixture evidence. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves failed-open recovery provenance comparison in text-only handoff. |
+| A256 | Source-search nested quality text parity | 100% | Nested source-search alternate-choice and handoff result text output now mirrors host, source, source type, source score, relevance, and likely-official fields as separate prefix rows, so failed-open recovery agents can compare result quality without parsing aggregate source-search rows. | Keep this to existing compact source-search result metadata; only add deeper ranking signals after fixture evidence. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves failed-open recovery candidate scoring in text-only handoff. |
+| A257 | Source-search nested preview text parity | 100% | Nested source-search alternate-choice and handoff result text output now mirrors URL, title, rank, open-result index, snippet, normalized date fields, first matched/find term, and selection reason as separate prefix rows, so recovery agents can compare candidates without parsing aggregate source-search rows. | Keep this to existing compact source-search result metadata; avoid adding full nested arrays unless a fixture proves the first-value shortcuts are insufficient. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves failed-open recovery preview comparison in text-only handoff. |
+| A258 | Generic top-choice text shortcut parity | 100% | Generic `topChoice*` text output now mirrors core result/source/form/action-target routing fields as separate prefix rows, including kind, path, URL, host, rank, open-result, recommendation/primary state, source quality, date, selector, reason, and command, so common executors can use the best choice without parsing the aggregate `topChoice:` row. | Keep this to fields already exposed in compact/public `topChoice*`; add deeper form/action fields only when a focused fixture shows aggregate parsing is still required. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves cross-kind top-choice routing in text-only handoff. |
+| A259 | Generic top-choice sitelink text parity | 100% | Generic `topChoiceFirstSitelink*` text output now mirrors title, URL, selector, and command as separate prefix rows, so result executors can open the strongest sublink without parsing the aggregate `topChoiceFirstSitelink:` row. | Keep this to the first sitelink already surfaced in compact/public output; deeper sitelink arrays stay nested until fixture evidence says otherwise. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves generic result sublink execution in text-only handoff. |
+| A260 | Top form first-field text parity | 100% | `topFormChoiceFirstField*` text output now mirrors label, placeholder, value, option list, autocomplete, input mode, pattern, min/max/step, length, required, checked, disabled, readonly, and invalid state as separate prefix rows, so form executors can choose safe input values without parsing the aggregate first-field line. | Keep this to the first executable field already surfaced in compact/public output; deeper field arrays stay in form-choice rows until a fixture proves per-field prefix expansion is needed. | Focused CLI text test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves static form execution fidelity in text-only handoff. |
+| A261 | Nested form-choice field text parity | 100% | Nested `formChoice*` text output now mirrors submit override attributes and first-field label, placeholder, value, options, autocomplete, input mode, validation, state, and selector prefix rows, so agents comparing multiple forms can choose and execute the right form without parsing aggregate rows or nested JSON. | Keep this aligned with fields already present on `AgentFormChoice`; add per-field arrays only if a multi-form fixture proves first-field shortcuts are insufficient. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-form execution fidelity in text-only handoff. |
+| A262 | Nested action-target identity text parity | 100% | Nested `actionTargetChoice*` text output now mirrors name, kind, source, and rank as separate prefix rows, so agents comparing multiple structured action targets can grep identity and priority without parsing the aggregate action-target row. | Keep this aligned with existing `AgentActionTargetChoice` identity fields; add deeper action metadata only when fixtures show aggregate parsing is still required. | Focused CLI text tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves multi-action-target comparison in text-only handoff. |
+| A263 | Readiness audit coverage for text shortcuts | 100% | `readiness:audit` now requires the recent form-field, form submit-override, and action-target identity text shortcut evidence, so future regressions in text-only multi-form/action handoff are caught by the project readiness gate rather than only by narrow focused tests. | Keep audit evidence limited to durable, user-facing shortcut names; avoid turning it into a duplicate of every formatter assertion. | Public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves durability of recent text-output parity work. |
+| A264 | Public type evidence for form/action shortcuts | 100% | Public type tests now assert the recent form first-field and action-target identity shortcut fields, and `readiness:audit` requires those public-type evidence names so TypeScript consumers keep the same handoff surface as CLI JSON/text users. | Keep public type assertions focused on stable API fields, not every text-only prefix. | Public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves external consumer confidence for multi-form/action handoff. |
+| A265 | Brief compact form/action handoff guards | 100% | Brief-mode regression tests now assert nested form field label/placeholder/options and structured action-target name/source/query/method/encoding/selector fields, and `readiness:audit` requires those evidence names. | This locks existing compact behavior rather than adding new output fields; add code only if a fixture shows brief output still drops execution-critical metadata. | Focused brief CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; prevents compact handoff regressions while keeping README detail out of the root README. |
+| A266 | Top data-table headers shortcut | 100% | `topDataTableHeaders` now exposes the full first-table header array in full, brief, text, public types, static scoring, and readiness evidence, so agents can compare column context without parsing `pageCheck.dataTables[0].headers`. | Keep this to the first table's already-extracted headers; deeper row/virtualized sampling still needs fixture evidence before adding broader payloads. | Focused table CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; closes a table/grid usability gap against accessibility-tree header lists. |
+| A267 | Top data-table second-row shortcut | 100% | `topDataTableSecondRow` and `topDataTableSecondCell` now expose the second extracted sample row in full, brief, text, public types, conditional static scoring, and readiness evidence, so agents can compare table options without opening nested `sampleRows`. | Keep this to the already-extracted second sample row; add wider row arrays only if a fixture proves two representative rows are insufficient. | Focused table CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves comparison-table usefulness while keeping payload growth bounded. |
+| A268 | Top result URL path shortcut | 100% | `topResultChoiceUrlPath` and optional `topResultChoiceUrlQuery` now expose parsed URL structure in full, brief, text, public types, static scoring, and readiness evidence, so agents can distinguish same-host result targets without parsing full URL strings. | Keep this to the top result choice until a fixture proves nested result/source choices need their own parsed URL parts. | Focused search CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves search-result routing precision. |
+| A269 | Top source-choice URL path shortcut | 100% | `topSourceChoiceUrlPath` and optional `topSourceChoiceUrlQuery` now expose parsed page source-link URL structure in full, brief, text, public types, static scoring, and readiness evidence, so agents can distinguish same-host source targets without parsing full URL strings. | Keep this to the top source choice until a fixture proves nested source choices need parsed URL parts. | Focused source-choice CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves page-check source routing precision. |
+| A270 | First sitelink URL path shortcuts | 100% | `topResultChoiceFirstSitelinkUrlPath`, `topChoiceFirstSitelinkUrlPath`, `sourceSearchSelectedFirstSitelinkUrlPath`, and `sourceSearchAlternateFirstSitelinkUrlPath` now expose parsed first-sitelink targets, with optional query fields when present, across full, brief, text, public types, and static scoring. | Keep this to the first sitelink already surfaced as a shortcut; nested/deeper sitelink arrays only get expanded if fixture evidence shows agents still need to parse them. | Focused CLI sitelink tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves sublink routing precision without making README longer. |
+| A271 | Semantic table/grid text row parity | 100% | Text-mode agent output now mirrors first table-header and selected table/grid-cell fields as individual prefix rows, including path, text, role, row/column indices when present, header arrays, selection/current state, selector, and owned target. | Keep this as text-only exposure of already-computed semantic table fields; do not synthesize missing row/column metadata without a focused extractor fixture. | Focused CLI table text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves table/grid parsing for text-only subagents. |
+| A272 | Selected list item text row parity | 100% | Text-mode agent output now mirrors selected/current list item fields as individual prefix rows, including text, role, level, position, set size, selected/current/expanded state, and selector. | Keep this as text-only exposure of existing `semanticTopSelectedListItem*` fields; nested list arrays stay compact unless a fixture shows the first selected/current item is insufficient. | Focused CLI list text test, typecheck, readiness audit, README test, diff check, and process check pass. | No headline increase; improves menu/list/tab routing for text-only subagents. |
+
+## Planned Work Detail
+
+| Order | Work | Why it matters | Entry condition | Done when | Risk |
+| ---: | --- | --- | --- | --- | --- |
+| 1 | Keep README short while this file carries progress detail. | Prevents another unreadable README buildup. | Any new long explanation is needed. | README tests pass and the detail lives under `docs/`. | Low. |
+| 2 | Watch deeper snippet provenance for search choices. | Agents can now read the selected result/source snippet without parsing nested arrays, but dedupe/provenance may still matter later. | A fixture shows top snippet shortcuts are not enough to choose or recover. | Add a focused provenance shortcut, or mark the gap as already covered. | Low; public type expansion only if justified. |
+| 3 | Tighten fallback policy categories. | Agents need to know when static inspection is enough and when browser capture is justified. | A fixture or real page produces unclear fallback guidance. | Reason codes and docs explain the decision without reading nested plans. | Medium; fetch/browser behavior can vary. |
+| 4 | Re-check table/grid semantics against observed gaps. | Tables are a major place where accessibility-tree context can beat plain HTML. | Gap record says current static output is insufficient. | New shortcut is added with public type/CLI/compact/test coverage, or the gap is documented as not worth implementing. | Medium; avoid adding noisy fields. |
+| 5 | Expand the browser parity gap ledger with only sequential evidence. | Progress becomes auditable instead of a broad percentage. | One small sequential fixture comparison is selected and process safety is clean. | At least one real fixture result is recorded as `implement`, `browser-only`, or `defer`. | Medium-high; browser-backed checks must remain limited. |
+
+## Milestone Tracker
+
+| Milestone | Estimate impact | Status | Evidence needed |
+| --- | ---: | --- | --- |
+| M1: README and docs stay controlled. | Keeps 90% docs hygiene stable. | In progress. | README tests pass and long operational detail remains under `docs/`. |
+| M2: No process leaks during validation. | Keeps 85% process containment stable. | In progress. | Sequential commands plus clean `pnpm check:processes` results before/after risky runs. |
+| M3: Static page output gives clear browser fallback reasons. | +2-4% page handoff. | Watch. | Reason codes cover observed no-content, fetch, retry, and browser-interaction cases; reopen only when a new page class is ambiguous. |
+| M4: Semantic tables/lists/controls have enough path shortcuts for agents to jump directly to evidence. | +3-6% semantic accessibility. | Watch. | Top shortcuts and nested refs include stable paths, selectors, states, resolved context, and local browser parity evidence for observed cases. |
+| M5: Browser parity gaps are tracked as research scope, not hidden churn. | +5-10% browser parity research. | Watch. | A maintained gap table records candidate signal, source, priority, decision, validation result, and ledger-backed local browser checks. |
+
+## Research Scope Ledger
+
+Research can grow while the work is underway. When a new useful
+accessibility-tree signal appears, add it here before implementing so progress
+estimates stay honest.
+
+| Signal or gap | Source | Priority | Decision | Status |
+| --- | --- | --- | --- | --- |
+| Table/grid ownership and cell navigation context beyond first-row shortcuts. | Static/browser fixture comparison. | P1 | Added header refs with path, role, row/column index, sort state, selector, table/grid ownership refs for `aria-owns`, selected/owned sample-cell shortcuts, and local browser parity for an owned virtual rowgroup; keep investigating deeper virtualized row sampling only if fixtures show it matters. | Watch. |
+| Browser-only fallback reasons for client-rendered, blocked, failed-open, or interaction-required pages. | Failed or low-content page checks, source-search failure recovery, executor/handoff browser steps, and barrier pages. | P1 | Added `staticReadiness`, `staticReadinessReasonCode`, `staticReadinessReason`, `staticReadinessReadFrom`, barrier-specific `browserHtmlReasonCode` values, action/executor/handoff browser fallback reason shortcuts, fallback contract markers, scorer coverage, runbook read-value shortcuts, and error-runbook recovery shortcuts. | Watch: landed for observed categories through A116; reopen only when a new page class has ambiguous browser-capture or live-interaction guidance. |
+| Brief output parity for page metadata. | Agent brief page handoff review. | P1 | Preserve top author-link name/source and best hidden read-target reason in `--agent-brief` so executor loops keep attribution and hidden-data routing rationale without requesting full output. | Landed. |
+| Brief output parity for table targeting. | Agent brief page handoff review. | P1 | Preserve top data-table path and selector in `--agent-brief` so structured table handoff can target the exact evidence without requesting full output. | Landed. |
+| Search-result provenance and failed-open reasons. | Search handoff review. | P2 | Added result/source choice host, quality, command, source-search selected/alternate/failure, engine-attempt, retryability, retry-after, and alternate-choice shortcuts so agents can compare provenance and recover failed opens without parsing nested payloads. | Watch. |
+| Source-choice official hint without nested parsing. | Search/source handoff review. | P2 | Added `isLikelyOfficial` to known official page source links and exposed `topSourceChoiceLikelyOfficial` in shallow full/text/compact/brief/public output. | Landed. |
+| Brief output parity for search provenance. | Agent brief search handoff review. | P2 | Preserve existing top result source hints, first sitelink title/URL, and top source-choice reason in `--agent-brief` so executor loops can choose or recover without requesting full output. | Landed. |
+| First list item accessibility state and selector without parsing arrays. | Semantic list/tree/menu handoff review. | P1 | Added first-item shortcuts for text, role, position, set size, selected/current/expanded state, and selector. Continue only if nested tree/list fixtures show agents need more. | Landed. |
+| Brief output parity for semantic selectors and state. | Agent brief handoff review. | P1 | Preserve existing top semantic selectors and core states in `--agent-brief` so executor loops can target elements without requesting the full payload. | Landed. |
+| Brief output parity for semantic table/list containers. | Agent brief handoff review. | P1 | Preserve top semantic table/list container selectors in `--agent-brief` so executor loops can target the actual container without requesting full output or deriving it from child selectors. | Landed. |
+| Brief output parity for actionable semantic targets. | Agent brief handoff review. | P1 | Preserve top link, in-page link, button, and image selectors in `--agent-brief` so executor loops keep direct click/inspect targets without requesting full output. | Landed. |
+| Brief output parity for table/list accessible context. | Agent brief handoff review. | P1 | Preserve top semantic table/list names and first sample-cell spans in `--agent-brief` so executor loops keep accessible labels and merged-cell structure without requesting full output. | Landed. |
+| Brief output parity for interactive/focusable targets. | Agent brief handoff review. | P1 | Preserve top interactive and focusable selectors in `--agent-brief` so keyboard and interaction routing keeps the exact first operable target without requesting full output. | Landed. |
+| Brief output parity for fetched-HTML usability. | Agent brief page handoff review. | P1 | Preserve `canUseFetchedHtml` in `--agent-brief` so executor loops can avoid unnecessary browser capture when fetched/static HTML is already usable. | Landed. |
+| Brief output parity for structured FAQ answers. | Agent brief page handoff review. | P1 | Preserve the top FAQ answer in `--agent-brief` so executor loops can use static question-answer evidence without requesting full output. | Landed. |
+| Brief output parity for code examples. | Agent brief page handoff review. | P1 | Preserve top code block text in `--agent-brief` so executor loops can use static command snippets and examples without requesting full output. | Landed. |
+| Brief output parity for resource metadata. | Agent brief page handoff review. | P1 | Preserve top resource titles in `--agent-brief` so executor loops can classify linked resources without requesting full output. | Landed. |
+| Brief output parity for keyboard accessibility targets. | Agent brief handoff review. | P1 | Preserve top keyboard shortcut selectors in `--agent-brief` so executor loops can target accesskey and aria-keyshortcuts elements without requesting full output. | Landed. |
+| Brief output parity for structured page evidence. | Agent brief page handoff review. | P1 | Preserve top media/section text plus provenance, offer, dataset, identity, timeline, contact, breadcrumb, pagination, TOC, embed, and transcript context in `--agent-brief` so executor loops can classify and target page evidence without requesting full output. | Landed. |
+| Local browser parity fixture gate. | Browser-side semantic tree vs `--agent-brief` comparison. | P1 | Added `pnpm compare:browser:fixture`; local fixtures cover heading/link/button/field/image, table header/cell context, form state/relation context, compacted list samples, keyboard shortcut targeting, expanded popup controls, current links, modal dialogs, and live regions. | Landed. |
+| Browser-side ARIA state extraction crash. | Local browser parity fixture. | P1 | Pass walk context into browser-side `getState` so `aria-controls`, `aria-live`, and `aria-valuetext` normalize without `ReferenceError`. | Landed. |
+| Compacted list item sample loss. | Local browser parity fixture. | P1 | Use link/button descendants as list sample fallback when semantic listitem wrappers compact away, so brief summaries still expose the useful list item text. | Landed. |
+| Brief output parity for non-first modal/live states. | Second local browser parity fixture. | P1 | Preserve top modal and live state shortcuts in `--agent-brief` so agents do not miss important browser-tree state just because the first state item is a button or link. | Landed. |
+| Browser parity for combobox active descendant and selected options. | Third local browser parity fixture. | P1 | Preserve selected/current option shortcuts in `--agent-brief` so agents do not miss the active selected choice just because the first option is unselected. | Landed. |
+| Browser parity for selected tab panel targets. | Fourth local browser parity fixture. | P1 | Preserve selected-choice controlled-target shortcuts in `--agent-brief` so agents can jump from the selected tab to its `tabpanel` without parsing relation arrays. | Landed. |
+| Browser parity for selected grid cells. | Fifth local browser parity fixture. | P1 | Preserve selected table/grid cell shortcuts in `--agent-brief` so agents can identify the active cell text, position, and selector without parsing sample-cell arrays. | Landed. |
+| Static fallback for semantic table/grid pages. | Fifth local browser parity fixture plus focused CLI fallback test. | P1 | Return `agent.semanticSummary` directly for low-prose semantic table/grid pages instead of escalating to browser interaction when static evidence is already structured and usable. | Landed. |
+| Brief output parity for primary execution shortcuts. | Agent brief handoff review. | P1 | Preserve shallow primary command args, after-interaction command args, rank/open-result, browser-interaction flag, readability reasons, and best read-target reason in `--agent-brief` so executor loops do not need full output or nested action parsing. | Landed. |
+| Brief output parity for decision and diagnostic shortcuts. | Agent brief handoff review. | P1/P2 | Preserve shallow search/page decision reasons, response status/OK/content-type/final-URL-change fields, verification counts, and barrier source/diagnostic code in `--agent-brief` so executor loops can classify failure, verification, and barrier cases without requesting full output. | Landed. |
+| Brief output parity for semantic counts and outline selector. | Agent brief semantic handoff review. | P1 | Preserve semantic description/value/relation/choice/state counts and `semanticTopOutlineSelector` in `--agent-brief` so executor loops can understand accessibility-tree shape and jump to the top outline target without full output. | Landed. |
+| Brief output parity for control-plane scalars. | Agent brief handoff review. | P1/P2 | Preserve routing intent, continuation mode, runbook/execution reasons, alternative-action count/reason, diagnostic codes, and diagnostic info count in `--agent-brief` so executor loops can classify the next step without nested payloads. | Landed. |
+| Source-search failure recovery classification. | Selected-result HTTP error fixture and source-search shortcut review. | P2 | Add `sourceSearchFailureKind` and `sourceSearchFailureRetryable` so agents can distinguish missing/client/server/fetch/timeout failures without parsing raw error code and status. | Landed. |
+| Table/grid selected-current cell gate coverage. | Semantic table fixture and static shortcut scoring review. | P1/P5 | Add selected/current cell state to the main semantic table fixture and require static gate scoring for first sample and selected table-cell shortcuts. | Landed. |
+| Live browser interaction reason codes. | Captured challenge brief/full fixture and browser fallback scorer review. | P1/P4 | Emit `browserHtmlReasonCode` for primary `interact-browser` handoffs even when no additional HTML capture is needed, preserving barrier-specific codes such as `challenge`. | Landed. |
+| Owned table rowgroup sampling. | Semantic table fixture with `aria-owns` rowgroup and static summary review. | P1/P5 | Include owned rowgroup descendants in top table row/cell counts, headers, header refs, sample cells, and sample-cell refs so static output better matches browser accessibility trees. | Landed. |
+| Rate-limited source-search recovery. | Opened search-result HTTP 429 fixture and source-search failure classifier review. | P2/P3 | Classify HTTP 429 selected-result failures as `rate-limited` and retryable, while HTTP 408 maps to `timeout`. | Landed. |
+| Source-search retry-after hints. | Opened search-result HTTP 429 fixture with `Retry-After` header. | P2/P3 | Preserve `Retry-After` as `sourceSearchFailureRetryAfter` so recovery loops can schedule same-URL retries without parsing response headers. | Landed. |
+| Additional browser accessibility-tree signals discovered during sequential comparison. | Future research. | P1/P2 after triage. | Add to this ledger, then classify as `implement`, `browser-only`, or `defer`. | Watch. |
+
+## Observed Gap Records
+
+Use this table to keep research scope honest. A row can raise or lower the
+percentage depending on whether it is implemented, documented as browser-only,
+or deferred.
+
+| ID | Observed gap | Evidence | Priority | Decision | Status | Validation / next command | Estimate impact |
+| --- | --- | --- | --- | --- | --- | --- | ---: |
+| G1 | `--open-result` failure recovery exposed selected/alternate result metadata, but not selected/alternate hosts as top-level shortcuts. Agents had to parse URLs or nested `sourceSearch` objects before choosing a replacement result. | `tests/cli.test.ts` missing-result fixture and source-search shortcut review. | P2 | Implement top-level `sourceSearchSelectedHost` and `sourceSearchAlternateHost`; keep failed-open reason categories as a later candidate. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check passed. | +1% search handoff. |
+| G3 | Table/grid `aria-owns` was visible only as a generic relation, not inside the table summary. Agents inspecting table/grid output had to parse relation items to detect browser-tree ownership or virtualized rowgroups. | Compact semantic fixture with `table[aria-owns]` and owned `rowgroup`. | P1 | Implement `ownedRefs`, `semanticTopTableOwnedCount`, `semanticTopTableOwnedRefs`, and first-owned shortcuts. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check passed. | +1% semantic accessibility. |
+| G4 | Blocked pages with challenge, login, or paywall diagnostics collapsed into generic `retry-action`/blocked fallback codes. Agents could not branch between simple rendered HTML capture and barrier-specific browser handling without reading diagnostics. | Challenge and login/paywall fixtures in `tests/cli.test.ts`. | P1 | Extend `browserHtmlReasonCode` with `challenge`, `login-required`, and `paywall` and prioritize diagnostics before generic retry codes. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check passed. | +1% page handoff. |
+| G5 | Failed opened search results required joining `error` with `sourceSearch.selectedResult` to know which selected result failed and why. This slowed recovery decisions and made failure provenance less direct. | `tests/cli.test.ts` selected-result HTTP error fixture. | P2 | Add `sourceSearchFailureCode`, `sourceSearchFailureStatus`, `sourceSearchFailureUrl`, and `sourceSearchFailureReason` shortcuts on source-search error payloads. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check passed. | +1% search handoff. |
+| G25 | Failed opened search results exposed raw failure code/status but still required agents to classify whether the selected URL was missing, blocked by a client error, transient, or worth retrying. | `tests/cli.test.ts` selected-result HTTP error fixture and `scripts/compare-static.ts` source-search shortcut gate. | P2 | Add `sourceSearchFailureKind` and `sourceSearchFailureRetryable` in full, compact, brief, public types, and static fixture scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, readiness audit, README test, diff check, and process check passed. | +1% search handoff. |
+| G26 | Selected/current table cell shortcuts existed, but the main static comparison gate did not score selected/current state for sample cells or selected-cell shallow shortcuts. A regression could drop browser-tree state parity while fixture gates stayed green. | Semantic table fixture and `scripts/compare-static.ts` shortcut scorer review. | P1/P5 | Add selected/current state to the semantic table fixture and score first-sample plus selected table-cell shortcuts in the static gate. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | +1% semantic accessibility. |
+| G27 | Captured challenge pages could require live browser interaction while `needsBrowserHtml=false`, so `browserHtmlReasonCode` stayed absent even though the agent still needed a browser step. Agents had to infer the barrier from diagnostics instead of shallow fallback fields. | Captured challenge `--agent` and `--agent-brief` fixtures. | P1/P4 | Emit browser/live-interaction reason text and reason code when the primary action is `interact-browser`, and keep barrier-specific codes such as `challenge`. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | +1% page handoff. |
+| G28 | Tables with `aria-owns` exposed owned rowgroups only as ownership metadata; row/cell/header samples ignored descendants of the owned target. Browser accessibility trees include owned rows/cells under the table, so static summaries could undercount virtualized table structure. | Compact agent semantic table fixture with `aria-owns="owned-rows"` and owned row/header/gridcell descendants. | P1/P5 | Traverse owned nodes when building table row/cell counts, header texts/refs, and sample cell refs. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | +1% semantic accessibility. |
+| G29 | Selected source-search result failures classified all 4xx statuses other than 404/410 as non-retryable client errors. Rate limits and request timeouts need different recovery guidance because retrying later can be correct. | Focused opened-result HTTP 429 fixture and source-search failure classifier review. | P2/P3 | Add `rate-limited` failure kind for HTTP 429, map HTTP 408 to `timeout`, and make both retryable through `sourceSearchFailureRetryable`. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, readiness audit, README test, diff check, and process check. | +1% failed-open recovery. |
+| G30 | Rate-limited selected source-search failures could be marked retryable, but the `Retry-After` value was discarded on HTTP errors. Agents still had to guess when to retry the selected URL. | Focused opened-result HTTP 429 fixture with `Retry-After: 120`. | P2/P3 | Preserve the response `Retry-After` header as `sourceSearchFailureRetryAfter` in full, compact, brief, public types, and tests. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, readiness audit, README test, diff check, and process check. | +1% failed-open recovery. |
+| G87 | Implicit `<button>` submit controls were summarized as type `submit`, but the fallback selector could be `button[type="submit"]`, which does not match buttons without an explicit `type` attribute. Form execution loops could identify the submit text but still fail to target the control. | Focused implicit-submit CLI fixture. | P1/P4 | Use `:nth-of-type` fallback selectors for unnamed implicit submit buttons while preserving name/type selectors when attributes exist. | Landed. | Typecheck, focused CLI test, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form execution targeting. |
+| G88 | Submit buttons can override form action, method, target, encoding, validation, and owner form, but form handoff used only the parent form defaults. Agents could submit the wrong URL or method when a submit control carries HTML5 form override attributes. | Focused submit-override CLI fixture. | P1/P4 | Apply submit `formaction`/`formmethod` to form handoff and expose submit override metadata in page forms, agent choices, top shortcuts, compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form execution correctness. |
+| G89 | Hidden form inputs such as CSRF tokens participate in browser form submission, but static form handoff dropped them completely. Agents could reconstruct the visible fields and submit target while missing required hidden parameters. | Existing form fixture with hidden `csrf` input. | P1/P4 | Preserve hidden form field count, compact hidden field refs, and first hidden field shortcuts across page forms, agent choices, top shortcuts, compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form submission completeness. |
+| G90 | Checkbox and radio controls decide whether their value is submitted, but page form handoff exposed value without checked state. Agents had to join semantic field state before deciding whether a form parameter would be included. | Focused checkbox form CLI fixture. | P1/P4 | Preserve form field `checked` and first-field checked shortcuts across page forms, agent choices, compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form submission correctness. |
+| G91 | Select controls expose option labels but not the selected/default option value that browsers submit. Agents could show the choices but still miss which value would be sent without inspecting option state elsewhere. | Existing select fixture plus focused selected-option CLI fixture. | P1/P4 | Preserve `selectedOption` and `selectedValue` for select fields and first-field shortcuts across page forms, agent choices, compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form submission correctness. |
+| G92 | Textarea default submission values live in element text content, not a `value` attribute. Static form handoff could expose the textarea field but miss the draft text a browser would submit. | Focused textarea form CLI fixture. | P1/P4 | Use textarea text content as the form field value and keep existing value/top-first-field handoff paths. | Landed. | Typecheck, focused CLI test, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form submission completeness. |
+| G93 | Image submit controls trigger form submission but were excluded from both normal fields and submit summaries. Agents could see the form but miss the actual image submit target. | Focused image-submit CLI fixture. | P1/P4 | Treat `input[type=image]` as a submit candidate, using accessible image text for submit text while preserving name/value/selector. | Landed. | Typecheck, focused CLI test, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form execution targeting. |
+| G94 | Form-level execution attributes such as `target`, `enctype`, `accept-charset`, `novalidate`, `id`, and `name` affect browser submission but were absent from form handoff unless a submit override existed. | Existing form fixture extended with form execution attributes. | P1/P4 | Preserve form-level execution metadata in page forms, agent choices, top shortcuts, compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens form execution fidelity. |
+| G95 | Action target choices preserved `encodingType` only inside nested choices. Top-action-target consumers could see method and template but still had to parse nested payloads to know the content type. | Existing JSON-LD search action fixture. | P2/P4 | Expose `topActionTargetChoiceEncodingType` in full/text/compact/brief output, public types, and static scoring. | Landed. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, and process check. | No headline increase; strengthens action target execution handoff. |
+| G6 | Top ranked result/source snippets were present only inside `resultChoices`/`sourceChoices`. Agents reading the shallow summary could see URL, host, command, and reason, but still had to parse nested arrays to inspect the first choice's snippet before opening or recovering. | Search result text fixture and public `AgentSummary` shortcut review. | P2 | Add `topResultChoiceSnippet` and `topSourceChoiceSnippet` with JSON, text, compact, public type, and focused test coverage. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff. |
+| G7 | Static readiness exposed status and a prose reason, but not a machine-readable reason code. Agents had to parse text to distinguish hidden app data, source-link recovery, thin content, and browser-required fallback. | Hidden hydration fixture, thin source-link fixture, blocked/no-inspectable/browser retry fixtures, and public `AgentSummary` shortcut review. | P1 | Add `staticReadinessReasonCode` with values for hidden data, source-link/form/action-target structured payloads, readable content, limited static payload, thin content, browser-required, interaction-required, and extraction errors. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G8 | Top interactive controls exposed their parsed state only through `semanticTopInteractiveState`. Agents had to parse a state string to know whether the first interactive target was pressed, expanded, opened a popup, or controlled another element, while button-specific shortcuts already exposed those fields. | Existing semantic button/control fixtures and public `AgentSummary` shortcut review. | P1 | Add `semanticTopInteractivePressed`, `semanticTopInteractiveExpanded`, `semanticTopInteractiveHaspopup`, and `semanticTopInteractiveControls` with JSON, text, compact, public type, and focused test coverage. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G9 | Top focusable controls exposed parsed state only through `semanticTopFocusableState`. Agents navigating by keyboard/focus order had to parse a state string to know disabled, pressed, expanded, popup, or controlled-target state. | Existing semantic focusable/control fixtures and public `AgentSummary` shortcut review. | P1 | Add `semanticTopFocusableDisabled`, `semanticTopFocusablePressed`, `semanticTopFocusableExpanded`, `semanticTopFocusableHaspopup`, and `semanticTopFocusableControls` with JSON, text, compact, public type, and focused test coverage. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G10 | Top list item refs exposed useful role/position/current/expanded data, but the first list item still required parsing `semanticTopListItemRefs[0]`. Agents routing through menus, trees, and list evidence need the first target's selector and state at the shallow summary level. | Existing semantic list fixture and public `AgentSummary` shortcut review. | P1 | Add `semanticTopListFirstItemText`, role, level, position, set size, selected/current/expanded state, and selector with JSON, text, compact, public type, and focused test coverage. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G11 | `--agent-brief` preserved many semantic shortcuts but dropped several selectors and core state booleans that full compact output already exposed. Brief executor loops could know a relation or choice existed but lose the exact element selector/state needed for browser-free routing. | Brief semantic relation/choice/state fixture review. | P1 | Add field/description/value/relation/choice/state selectors and hidden/disabled/selected/focused/required/readonly state shortcuts to brief output. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G12 | `--agent-brief` dropped some search-choice provenance available in full compact output: top result source hints, first sitelink title/URL, and top source-choice reason. Agents could see the URL but lose why it was preferred or which sitelink was the best follow-up. | Brief search handoff comparison against full compact shortcuts. | P2 | Preserve those existing top-level fields in brief output and extend the brief alignment test to fail on future losses. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff. |
+| G13 | `--agent-brief` kept top author URL and hidden read-target path/count, but dropped author name/source and the hidden read-target reason. Agents could follow a URL but lose source attribution, or see hidden app data without knowing why it was the best static read target. | Brief page metadata comparison against full compact shortcuts. | P1 | Preserve `topAuthorLinkName`, `topAuthorLinkSource`, and `bestHiddenReadTargetReason` in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G14 | `--agent-brief` kept data-table caption/count/first row but dropped `topDataTablePath` and `topDataTableSelector`. Agents could identify table content but not jump back to the exact static table target without full output. | Brief data-table handoff comparison against full compact shortcuts. | P1 | Preserve top data-table path and selector in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G15 | `--agent-brief` kept semantic table/list detail selectors such as first header, sample cell, and first list item, but dropped the table/list container selectors. Agents could target child evidence but still needed full output or selector inference to operate on the actual container. | Brief semantic table/list handoff comparison against full compact shortcuts. | P1 | Preserve `semanticTopTableSelector` and `semanticTopListSelector` in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G16 | `--agent-brief` kept top link, in-page link, button, and image names/metadata but dropped their selectors. Agents could identify an actionable semantic target, but still needed full output before clicking, inspecting, or routing to that exact element. | Brief actionable semantic handoff comparison against full compact shortcuts. | P1 | Preserve `semanticTopLinkSelector`, `semanticTopInPageLinkSelector`, `semanticTopButtonSelector`, and `semanticTopImageSelector` in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G17 | `--agent-brief` kept semantic table/list structure but dropped the table/list accessible names and first sample-cell row/column spans. Agents could see a table or list existed, but lost the label and merged-cell shape that often explain what the accessibility tree is exposing. | Brief semantic table/list context comparison against full compact shortcuts. | P1 | Preserve `semanticTopTableName`, `semanticTopListName`, `semanticTopTableFirstSampleCellRowSpan`, and `semanticTopTableFirstSampleCellColumnSpan` in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G18 | `--agent-brief` kept top interactive/focusable names and state but dropped their selectors. Agents could know the first keyboard or interaction target existed, but still needed full output before jumping to that exact operable element. | Brief interaction/focus handoff comparison against full compact shortcuts. | P1 | Preserve `semanticTopInteractiveSelector` and `semanticTopFocusableSelector` in brief output with focused fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G19 | `--agent-brief` exposed `needsBrowserHtml` and static-readiness details but dropped `canUseFetchedHtml`. Agents could see browser capture was not needed, but lacked the positive boolean that full compact output uses to decide static output is safe to read. | Brief static-readiness comparison against full compact shortcuts. | P1 | Preserve `canUseFetchedHtml` in brief output with focused hydration/static-readiness fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G20 | `--agent-brief` kept the top FAQ question but dropped the answer. Agents could identify that static FAQ evidence existed, but still needed full output before answering from the question-answer pair. | Brief FAQ handoff comparison against full compact shortcuts. | P1 | Preserve `topFaqAnswer` in brief output with focused FAQ fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G21 | `--agent-brief` kept top code block language and line count but dropped the code text. Agents could identify a command/example block existed, but still needed full output before using the static snippet. | Brief code-block handoff comparison against full compact shortcuts. | P1 | Preserve `topCodeBlockText` in brief output with focused code-block fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G22 | `--agent-brief` kept top resource kind and URL but dropped the title. Agents could see a linked resource existed, but had to request full output to distinguish a feed, report, or document by its label. | Brief resource handoff comparison against full compact shortcuts. | P1 | Preserve `topResourceTitle` in brief output with focused resource fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G23 | `--agent-brief` kept top keyboard shortcut metadata but dropped its selector. Agents could see an accesskey or aria-keyshortcuts target existed, but needed full output before jumping to the exact keyboard-accessible element. | Brief keyboard accessibility comparison against full compact shortcuts. | P1 | Preserve `semanticTopKeyboardShortcutSelector` in brief output with focused keyboard shortcut fixture coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility. |
+| G24 | `--agent-brief` still dropped several top structured page evidence details that full compact output exposed: media text, section text, breadcrumb source/path, pagination label/kind, TOC title/count, embed/transcript labels, and provenance/offer/dataset/identity/timeline/contact labels, sources, and selectors. Agents could see counts or URLs but often needed full output to classify evidence or jump to the exact static source. | Brief page-check shortcut comparison against full compact output and focused page evidence fixtures. | P1 | Preserve the missing structured page evidence shortcuts in brief output with focused provenance, offer, dataset, identity, timeline, contact, section, and media coverage. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff. |
+| G25 | Browser-side extraction crashed when a node had `aria-controls`, `aria-live`, or `aria-valuetext` because `getState` referenced `context` without receiving it. This made browser-tree comparison itself unreliable for stateful controls. | `pnpm compare:browser:fixture` first run failed with `ReferenceError: context is not defined`; focused extractor test covers ARIA state values. | P1 | Pass the walk context into browser-side `getState` in both source and generated extractor script. | Landed. | `pnpm exec vitest run tests/extractor.test.ts` and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility. |
+| G26 | In compact list output, list item wrappers can collapse to descendant links. Browser-side output still showed the useful list link, but static agent summaries had list name without sample item text, so a subagent could not prove the list contents from `--agent-brief`. | `pnpm compare:browser:fixture` showed `list:Release actions` and `link:Download report` in browser evidence but no `semanticTopListItems` in agent evidence. | P1 | Use link/button descendants as list sample fallback when listitem-like roles are absent. | Landed. | Focused CLI test and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility. |
+| G27 | The project lacked a small, auditable browser parity command that could be run safely after the prior server freeze incident. Existing browser comparison commands were broader and could touch external URLs or `agent-browser`. | Process-safety review plus current `package.json` script inventory. | P1 | Add `pnpm compare:browser:fixture`, a one-browser, one-local-fixture, no-network comparison gate. | Landed. | `pnpm compare:browser:fixture` passes and `pnpm check:processes` reports no leftovers before and after. | +5% browser parity research. |
+| G28 | A page can expose useful modal/live accessibility-tree state even when the first summarized state item is another control. In that case `--agent-brief` showed the current link and button state, but not the dialog modal state or live-region state without parsing nested `semanticSummary.stateItems`. | Second local browser fixture `stateful-overlay-links` showed `dialog:Settings panel`, `aria-modal=true`, and `aria-live=polite` in browser evidence while shallow agent fields only exposed the first button state. | P1 | Add `semanticTopModalState*` and `semanticTopLiveState*` shortcuts and preserve them in compact/brief output. | Landed. | Focused CLI/public type tests and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility, +5% browser parity research. |
+| G29 | Combobox/listbox pages can expose an active selected option that is not the first option. `--agent-brief` showed `semanticTopChoice*` for the first option, so a subagent still had to parse nested `semanticSummary.choiceItems` to identify the selected/current choice. | Third local browser fixture `combobox-active-descendant` showed `aria-activedescendant=report-option-2`, selected/current `option:Quarterly reports`, and position metadata while shallow choice fields pointed at unselected `Archive`. | P1 | Add `semanticTopSelectedChoice*` shortcuts and preserve them in compact/brief output. | Landed. | Focused CLI/public type tests and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility, +5% browser parity research. |
+| G30 | A selected tab can control the active `tabpanel`, but the shallow selected-choice shortcuts did not expose the resolved panel target. Agents could identify the selected tab but still had to parse relation arrays to jump to the panel content. | Fourth local browser fixture `tablist-selected-panel` showed selected `tab:Details` controls `panel-details` and `tabpanel:Details`; shallow selected-choice fields needed target role/name/selector. | P1 | Add `semanticTopSelectedChoiceControls*` target shortcuts and preserve them in compact/brief output. | Landed. | Focused CLI/public type tests and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility, +5% browser parity research. |
+| G31 | A selected `gridcell` can be the active point in a data grid, but shallow table shortcuts exposed sample cell text/position without selected/current state or a direct selected-cell target. Agents had to parse `semanticTopTableSampleCellRefs` or state arrays to know which cell was active. | Fifth local browser fixture `grid-selected-cell` showed browser-side `gridcell` text `Blocked`, `aria-selected=true`, and column index 2 while shallow agent fields needed selected-cell text, row/column index, state, and selector. | P1 | Add selected table/grid cell shortcuts and preserve cell selected/current state in sample-cell refs, compact output, brief output, text output, and public types. | Landed. | Focused CLI/public type tests and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | +1% semantic accessibility, +5% browser parity research. |
+| G32 | Low-prose semantic table/grid pages exposed useful accessibility-tree structure, but primary handoff still chose browser interaction because prose evidence was thin. This could cause unnecessary agent-browser use even when `agent.semanticSummary` already contained enough structured evidence. | Focused selected-gridcell brief fixture showed `canUseFetchedHtml=true`, `needsBrowserHtml=false`, and rich semantic grid evidence while the previous primary action was browser inspection. | P1 | Prefer `read-content` from `agent.semanticSummary` for non-blocked, non-search pages where semantic table/grid headers or sample cells exist and no source-link/barrier handoff should win. | Landed. | Typecheck, focused CLI test, `pnpm compare:browser:fixture`, and pre/post process checks pass. | +1% page handoff; reduces unnecessary browser escalation. |
+| G33 | Official page source links had `sourceType=official` in nested page links, but the shallow source-choice summary did not expose the boolean official hint. Agents could identify the URL but still had to inspect nested source choices or source profiles before trusting it as a likely official follow-up. | Focused `--agent-brief` source-link fixture with `https://openai.com/research`, plus public `AgentSummary` shortcut review. | P2 | Add source-link `isLikelyOfficial` for known official source profiles and expose `topSourceChoiceLikelyOfficial` in full/text/compact/brief/public output. | Landed. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff; overall 95% -> 96%. |
+| G34 | `--agent-brief` preserved primary action details only inside nested `primaryAction`/executor/handoff objects, while full compact exposed shallow primary command args, after-interaction command args, rank/open-result, browser-interaction flag, readability reasons, and best read-target reason. Agents reading the shallow brief summary could miss the exact command or reason and request full output unnecessarily. | Mechanical compact-vs-brief field comparison plus focused stdin fixtures for missing evidence, matched evidence, and captured challenge HTML. | P1 | Preserve the high-value shallow primary execution/read-target fields in brief output; keep bulky nested objects out of brief unless future fixtures prove they are needed. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% page handoff; overall remains 96%. |
+| G35 | `--agent-brief` still dropped shallow decision/diagnostic fields that full compact exposed: search/page decision reasons, HTTP response status/OK/content type/final URL change, verification counts, and barrier source/diagnostic code. Agents could see the next step but lose the reason and failure/barrier classification without full output. | Mechanical compact-vs-brief field comparison plus focused search, missing-evidence, matched-evidence, and captured-challenge brief fixtures. | P1/P2 | Preserve these small scalar diagnostics in brief output while continuing to omit bulky full decision/signal/quality-gate arrays. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% search handoff; overall remains 96%. |
+| G36 | `--agent-brief` dropped semantic relation/choice/state/value/description counts and `semanticTopOutlineSelector` even though full compact exposed them. Agents could identify a top semantic outline item but lose the selector and count-level shape of accessibility features without full output. | Mechanical compact-vs-brief field comparison plus focused relation brief fixture. | P1 | Preserve the small semantic count scalars and top outline selector in brief output while keeping full `semanticSummary` omitted. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | +1% semantic accessibility; overall remains 96%. |
+| G37 | After scalar cleanup, `--agent-brief` still missed small control-plane fields from full compact: routing intent, continuation mode, runbook/execution reasons, alternative-action count/reason, diagnostic codes, and diagnostic info count. The remaining non-scalar differences are large nested objects and should stay omitted until a fixture proves a need. | Mechanical compact-vs-brief field comparison and common brief alignment fixture. | P1/P2 | Preserve the scalar fields and document the remaining 13 compact-vs-brief differences as intentional nested-payload omissions. | Landed. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, and process check pass. | Closes current non-browser brief parity cleanup; overall remains 96%. |
+| G38 | Range/slider accessibility-tree value state is useful for agents deciding whether a control or progress indicator is at the expected value. The fixture needed to prove that `--agent-brief` preserves browser-side orientation, min/max/current value, value text, value item, and state selector without nested parsing. | Sixth local browser fixture `range-value-state` showed `slider:Release progress`, `aria-orientation=horizontal`, `aria-valuemin=0`, `aria-valuemax=100`, `aria-valuenow=40`, and `aria-valuetext=40 percent`; shallow agent fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing value/state shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 85% -> 88%; overall remains 96%. |
+| G39 | Busy live status regions let agents know whether dynamic content is still updating before trusting page text. The first draft fixture showed that unnamed `role=status` does not necessarily produce a named browser role, so the stable parity check should use an explicit accessible name and compare state rather than infer name from text. | Seventh local browser fixture `busy-status-state` showed `status:Indexing status`, `aria-busy=true`, and `aria-live=polite`; shallow state and live-state fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing busy/live shortcuts already matched the browser tree once the status region had an explicit name. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 88% -> 90%; overall remains 96%. |
+| G40 | Invalid fields with `aria-errormessage` are a common form-repair signal. Agents need the invalid reason and resolved error text without opening a browser accessibility tree or parsing nested relations. | Eighth local browser fixture `invalid-field-state` showed `textbox:Report code`, `aria-invalid=spelling`, and `aria-errormessage=code-error`; shallow field/state invalid fields, error-message id/text, and relation fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing invalid/error-message shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 90% -> 92%; overall remains 96%. |
+| G41 | Sorted column headers are a table-navigation signal that agents can use to decide whether a table is already in the needed order before interacting. The fixture needed to prove browser-side `aria-sort` state matches shallow table-header and state shortcuts. | Ninth local browser fixture `sorted-header-state` showed `table:Quarterly reports`, `columnheader:Quarter`, and `aria-sort=descending`; shallow first-header sort and top-state sort fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing header/state sort shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 92% -> 94%; overall remains 96%. |
+| G42 | Multiselect listboxes are a common filter UI where agents need to know that multiple options can be selected and which option is currently selected. The fixture needed to prove browser-side `aria-multiselectable` state and selected option position metadata match shallow state and selected-choice shortcuts. | Tenth local browser fixture `multiselect-listbox-state` showed `listbox:Report filters`, `aria-multiselectable=true`, `option:Open reports` selected, and position metadata; shallow multiselectable state and selected-choice fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing multiselectable and selected-choice shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 94% -> 95%; overall remains 96%. |
+| G43 | Drag/drop widgets can expose `aria-grabbed` and `aria-dropeffect` states that tell agents whether an item is ready to move. The fixture needed to prove browser-side drag/drop state matches shallow state shortcuts. | Eleventh local browser fixture `drag-drop-state` showed `button:Move report`, `aria-grabbed=true`, and `aria-dropeffect=move`; shallow button name, grabbed/dropEffect state, and selector fields matched. | P2 | Mark as covered evidence; no new implementation needed because existing drag/drop shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 95% -> 96%; overall remains 96%. |
+| G44 | Disabled and readonly fields are common form-state signals that tell agents whether a value can be edited or should only be read. The first draft using ARIA-only `aria-readonly` showed browser extractor parity was weaker there, so the stable fixture uses native field attributes for the browser-tree comparison. | Twelfth local browser fixture `disabled-readonly-field-state` showed `textbox:Archive code`, `disabled=true`, `readonly=true`, value `AR-42`, and selector `#archive-code`; shallow field/state/value fields matched. | P2 | Mark native disabled/readonly field state as covered evidence; keep ARIA-only readonly as a future candidate only if a real page needs it. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 96% -> 97%; overall remains 96%. |
+| G45 | Mixed checkbox state tells agents that a filter or setting is partially selected, which is materially different from checked or unchecked. The fixture needed to prove browser-side `aria-checked=mixed` matches shallow field/state checked shortcuts. | Thirteenth local browser fixture `mixed-checkbox-state` showed `checkbox:Include archived reports` and `checked=mixed`; shallow `semanticTopFieldChecked` and `semanticTopStateChecked` matched. | P2 | Mark mixed checkbox state as covered evidence; no new implementation needed because existing checked shortcuts already matched the browser tree. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 97% -> 98%; overall remains 96%. |
+| G46 | `aria-details` can point from a field to explanatory content that agents need before deciding how to fill or interpret the field. The fixture needed to prove browser-side field attributes and detail text match shallow field details and relation target shortcuts. | Fourteenth local browser fixture `field-details-relation` showed `searchbox:Archive filter`, `aria-details=archive-filter-details`, resolved details text, field selector, and target selector; shallow field details and relation fields matched. | P2 | Mark field details relation text as covered evidence; no new implementation needed because existing details shortcuts already matched the browser-side semantic output. | Landed. | Typecheck and `pnpm compare:browser:fixture` pass; pre/post process checks clean. | Browser parity research 98% -> 99%; overall remains 96%. |
+| G47 | Local fixtures alone do not prove real fetched pages can avoid browser capture, so a tiny real-page gate is needed before broader `agent-browser` comparison. | `readiness:real-page-smoke` checks `https://example.com` with `--agent-brief`; the run passed with static handoff usable, browser HTML not needed, named semantic roles present, and no browser continuation mode. | P1 | Keep the smoke gate as the smallest real-page guard; move next to one carefully bounded `agent-browser` comparison slice. | Landed. | Typecheck, `pnpm readiness:real-page-smoke`, readiness audit, README test, diff check, and process checks pass. | Overall 96% -> 97%; broader `agent-browser` comparison remains. |
+| G48 | A real `agent-browser` comparison is needed to prove the local/static evidence aligns with the actual reference tool on live pages. | `readiness:agent-browser-smoke` first proved `https://example.com` exact overlap, then expanded to `https://books.toscrape.com/`, `https://news.ycombinator.com`, and `https://www.gov.uk/foreign-travel-advice`; the GOV.UK run passed with overlap above `0.9`, readiness above `0.9`, and actionable/navigation recall at `1.0`, with clean pre/post process checks. | P1 | Keep this as the smallest browser-backed guard; next broaden only one documented target at a time. | Landed. | Typecheck, process checker test, `pnpm readiness:agent-browser-smoke`, readiness audit, README test, diff check, and process checks pass. | Overall 97% -> 99.2%; broader `agent-browser` comparison remains. |
+| G49 | Text-heavy document pages can make `agent-browser` StaticText/listmarker volume dominate named-role overlap even when actionable and navigation signals are usable. | `pnpm compare https://ko.wikipedia.org/wiki/대한민국` passed without process leaks; under the new score it reported overlap `0.446`, readiness `0.837`, actionable recall `0.823`, navigation recall `0.846`, structural content recall `0.864`, and text recall `0`. | P1 | Keep strict overlap/content/text recall visible, but use structural content recall in the weighted readiness score so StaticText volume does not hide useful action/navigation parity. | Landed as a separate text-heavy smoke gate. | Typecheck, text-heavy smoke with pre/post process checks, readiness audit, static fixture gate, README test, and main four-target smoke pass. | Overall holds at 99.2%; scope expanded and scoring policy is explicit. |
+| G50 | Text-heavy fetched pages could expose many focused paragraphs while `pageCheck.contentLength` only counted the first four preview snippets, making long static documents look thinner than they are and weakening readability/static-handoff scoring. | Non-browser analysis found `summarizePageCheck` capped `contentPreview` and `contentEvidence` at four items and reused preview length as total content length. | P1 | Count all focused content toward `contentLength` while keeping preview/evidence capped, so agents get accurate text-volume signals without a larger inline payload. | Landed. | Typecheck, focused CLI regression test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves text-heavy page handoff signal without adding browser load. |
+| G51 | Failed opened search results exposed only the first alternate source-search result as shallow top-level fields. If more than one fallback candidate existed, agents had to parse nested `sourceSearch.alternateResults` before choosing the next viable result. | Non-browser source-search fixture now includes two alternate candidates after a failed opened result. | P2 | Add `sourceSearchAlternateChoices` to top-level agent output, compact output, brief output, text output, and public types while keeping the existing first-alternate shortcut fields. | Landed. | Typecheck, focused CLI regression test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Search handoff 88% -> 89%; overall holds at 99.2%. |
+| G52 | Generic `topChoice` exposed only kind/path/label/url/command args, so an executor using the common choice fields could miss rank, source, method, selector, or selection reason and then fall back to kind-specific nested choices. | Existing agent alignment fixture covers result, source, form, and action-target top choices. | P2 | Add `topChoiceRank`, `topChoiceSource`, `topChoiceMethod`, `topChoiceSelector`, and `topChoiceReason` to full, compact, brief, text, and public type output. | Landed. | Typecheck, focused agent CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; reduces nested parsing for cross-kind choice routing. |
+| G53 | Recommended search result shortcuts exposed URL, title, rank, source, relevance, and command args, but not the direct payload path or source score. Agents had to infer `recommendedResult` or inspect nested result choices before citing or comparing source quality. | Existing out-of-window recommendation fixture covers a recommended result outside the compact search result slice. | P2 | Add `recommendedPath` and `recommendedSourceScore` to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused recommendation CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves recommended-result handoff without browser work. |
+| G54 | `topCitationScore` existed, but `topAnswerEvidenceScore` was missing. Agents could see the score on generic citations but had to inspect nested `answerEvidence` to judge the actual evidence selected for answer handoff. | Existing agent alignment fixture covers top answer evidence in full and brief-style output. | P2 | Add `topAnswerEvidenceScore` to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused agent CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves answer-evidence quality handoff without browser work. |
+| G55 | `searchDecision` and its shallow shortcuts exposed a recommended result's rank and URL, but not its payload path, title, source, source score, relevance, or likely-official flag. Decision-only consumers had to join against `recommendedResult` before trusting or opening the chosen result. | Existing search agent fixtures cover both the top result and an out-of-window recommended result. | P2 | Add recommended result metadata to `searchDecision`, shallow full/compact/brief/text output, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves search decision handoff without browser work. |
+| G56 | `pageDecision` carried readability, evidence, and source quality scalars, but shallow page-decision shortcuts exposed only decision/confidence/reason/readFrom/url/command args. Page-check loops using brief scalar output had to parse nested `pageDecision` before judging static-read confidence. | Existing page-check fixtures cover matched evidence and source-link decisions. | P2 | Add page-decision readability/evidence/source quality scalars to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves page-check decision handoff without browser work. |
+| G57 | Executor and handoff target shortcuts exposed only url/path/selector/text, while nested `target` could carry title, host, source, rank, source score, relevance, and likely-official metadata. Execution loops using scalar brief output had to parse nested target objects before trusting a result or source-link target. | Existing agent alignment fixture covers executor and handoff targets across search/page cases. | P2 | Add executor/handoff target title, host, source, rank, source score, relevance, and likely-official shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves execution-target handoff without browser work. |
+| G58 | `primaryAction.target` can carry the same quality metadata as executor/handoff targets, but scalar primary shortcuts only exposed action, command, URL, rank, and source-link ref. Primary-only loops had to parse nested action target objects before trusting a result or source-link target. | Existing agent alignment fixture covers primary actions with nested targets. | P2 | Add primary target title, host, source, rank, source score, relevance, likely-official, selector, and text shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves primary-action handoff without browser work. |
+| G59 | Alternative actions could point at a result or source-link target, but scalar alternative-action shortcuts exposed only action, source, execution, command, URL, and source-link ref. Recovery loops had to parse nested action targets before choosing a fallback. | Existing agent alignment fixture covers alternative actions when present. | P2 | Add alternative-action target title, host, source, rank, source score, relevance, likely-official, selector, and text shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves fallback-action handoff without browser work. |
+| G60 | `topAction` shortcuts exposed action, source, execution, command, URL, and source-link ref, but not the top action target's quality metadata. Loops that only consume the top-action scalars still had to parse the action list before judging the selected target. | Existing agent alignment fixture covers `actions[0]` across search/page cases. | P2 | Add top-action target title, host, source, rank, source score, relevance, likely-official, selector, and text shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves top-action handoff without browser work. |
+| G61 | Action objects and top/primary/alternative action shortcuts exposed execution and priority, while the expected outcome was only available through `expectedOutcome`, executor, or handoff summaries. Action-only loops had to join across payload sections before knowing whether an action should read evidence, open a result, capture HTML, inspect output, or stop. | Existing agent alignment fixture covers action lists, top action, primary action, and alternative action. | P2 | Add expected-outcome kind/message to action objects plus top, primary, and alternative action shortcuts in full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves action-only handoff without browser work. |
+| G62 | Top and alternative action shortcuts still missed priority reason, command string, after-interaction command, rank, and open-result fields that action objects could carry. Action-only loops could see intent but still had to parse nested actions for exact execution provenance. | Existing agent alignment fixture covers top and alternative actions, and static fixture scoring covers compact command behavior. | P2 | Add top/alternative priority-reason, command, after-interaction command, rank, and open-result shortcuts to full, compact, brief, text, public types, and static comparison scoring; keep brief command strings normalized to `--agent-brief`. | Landed. | Typecheck, focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Overall holds at 99.2%; improves action-only execution handoff without browser work. |
+| G63 | Source-search selected/alternate shortcuts exposed rank, title, URL, host, open-result index, command args, and reason, but missed command string, source, source score, relevance, and likely-official flag. Failed-open recovery loops still had to parse `sourceSearch.selectedResult` or `alternateResults[0]` before trusting a fallback result. | Source-search selected-result fixture, missing-result alternate fixture, public type coverage, and static fixture scoring. | P2 | Add selected/alternate source, command, source score, relevance, and likely-official shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused source-search CLI tests, public type test, readiness audit, static fixture gate, README test, diff check, and process check pass. | Search handoff 89% -> 90%; overall holds at 99.2%. |
+| G64 | `sourceSearchAlternateChoices` existed for multi-candidate failed-open recovery, but tests and static fixture scoring did not verify that each array item preserved source, command, source score, relevance, and likely-official fields. A regression could remove choice-level quality metadata while first-alternate shortcuts still passed. | Missing-result source-search fixture with two alternate candidates plus static fixture scoring review. | P2 | Extend focused CLI/public type tests and static comparison scoring to verify alternate-choice quality fields for the first three fallback candidates. | Landed. | Typecheck, focused source-search CLI tests, public type test, static fixture gate, README test, diff check, and process check pass. | No estimate increase; protects G51/G63 behavior from regression. |
+| G65 | Page source-choice shallow shortcuts exposed command args but not the command string, and static fixture scoring did not verify `topSourceChoiceLikelyOfficial`. Also, brief compaction could drop `command` from source-search alternate-choice array items even though recovery loops benefit from a copy-ready command. | Source-link fixtures, source-choice static scoring review, and brief source-search alternate-choice regression test. | P2 | Add `topSourceChoiceCommand` to full/text/compact/brief/public output, score command and likely-official in static comparison, and preserve command strings in compacted source-search result lists. | Landed. | Typecheck, focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; reduces nested parsing for page source-link and failed-open recovery. |
+| G66 | Generic `topChoice` still exposed command args without the command string, and result-specific shallow shortcuts had the same gap. A generic executor could see the best result/source choice but still parse `resultChoices[0]` or `sourceChoices[0]` to get a copy-ready command. | Existing agent alignment fixture covers result/source top choices; public type fixture covers `topChoice*`. | P2 | Add `topChoiceCommand` and `topResultChoiceCommand` to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, README test, diff check, and process check pass. | No estimate increase; improves generic top-choice executor handoff without browser work. |
+| G67 | Generic `topChoice` still missed source type, source score, relevance, and likely-official quality metadata that result/source choices already carried. A generic executor could identify and run the best choice but still parse nested choices before trusting source quality. | Existing agent alignment fixture covers result/source top choices; public type fixture covers `topChoice*`; static scoring review. | P2 | Add `topChoiceSourceType`, `topChoiceSourceScore`, `topChoiceRelevance`, and `topChoiceLikelyOfficial` to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic top-choice quality handoff without browser work. |
+| G68 | Generic `topChoice` still missed host and snippet preview context that result/source choices already carried. A generic executor could run and judge source quality, but still parse nested choices before seeing the target host or preview text. | Existing agent alignment fixture covers result/source top choices; public type fixture covers `topChoice*`; static scoring review. | P2 | Add `topChoiceHost` and `topChoiceSnippet` to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic top-choice context handoff without browser work. |
+| G69 | Generic `topChoice` still missed open-result index, recommended state, and primary state that result/source choices already carried. A generic executor could inspect the target and quality, but still parse nested choices before knowing which result index to open or whether the choice was primary/recommended. | Existing agent alignment fixture covers result/source top choices; public type fixture covers `topChoice*`; static scoring review. | P2 | Add `topChoiceOpenResult`, `topChoiceRecommended`, and `topChoicePrimary` to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves generic top-choice execution handoff without browser work. |
+| G70 | Top source-choice shortcuts still missed kind, rank, anchor text, selector, relevance, matched term, and find match even though `sourceChoices[0]` carried them. Page source-link loops could see URL/host/command/quality but still had to parse nested source choices before targeting or explaining the selected source link. | Existing source-link and agent alignment fixtures cover top source choices; public type fixture covers `topSourceChoice*`; static scoring review. | P2 | Add source-choice detail shortcuts to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page source-link handoff without browser work. |
+| G71 | Source-search selected and first-alternate shortcuts carried URL, host, command, score, relevance, and official status, but missed snippet, date text, matched term, find match, and sitelink preview metadata already present in `selectedResult` and `alternateResults[0]`. Failed-open recovery loops still had to parse nested source-search results before comparing candidate previews. | Source-search best-result and missing-result recovery fixtures plus static scoring review. | P2 | Add selected/alternate source-search preview shortcuts to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused source-search CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves failed-open recovery handoff without browser work. |
+| G72 | Read targets exposed path, count, score, primary, and reason, but not a machine-readable category. Executors had to parse paths such as `pageCheck.contentEvidence`, `agent.semanticSummary`, or `sourceSearch.alternateResults` before deciding whether to read evidence, semantic data, hidden app data, interaction metadata, or recovery search results. | Existing read-target static fixtures and public type fixture cover top read-target shortcuts. | P2 | Add `kind` to read target objects plus `topReadTargetKind` in full/text/compact/brief/public output and score it in static comparison. | Landed. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves read-current routing without browser work. |
+| G73 | `bestReadTarget` shortcuts still exposed path/count/score/primary/reason without kind. If the best target differed from the first read target, best-target routing still had to parse the path before deciding how to read the current payload. | Existing read-target static fixtures and public type fixture cover best read-target shortcuts. | P2 | Add `bestReadTargetKind` to full/text/compact/brief/public output and score it in static comparison. | Landed. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves best read-target routing without browser work. |
+| G74 | Stage-specific `runbookReadFrom`, `nextReadFrom`, `executorReadFrom`, and `handoffReadFrom` still lacked a kind shortcut. Agents could see the global top/best read target kind, but still had to parse a stage-specific path when a stage selected a different read source. | Existing agent alignment fixture and public type fixture cover stage readFrom shortcuts and nested readTarget kinds. | P2 | Add stage-specific read-target kind shortcuts to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves stage-specific read-current routing without browser work. |
+| G75 | Stage-specific read-target shortcuts still exposed only path and kind, while count, score, primary, and reason were available only in global top/best shortcuts or nested readTarget objects. A stage could select a different read source and still force agents to parse nested data before judging quality or why it was selected. | Existing agent alignment fixture, public type fixture, and static fixture scoring cover nested readTarget quality fields. | P2 | Add stage-specific read-target count, score, primary, and reason shortcuts to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves stage-specific read-current quality routing without browser work. |
+| G76 | `executionPlanReadFrom` and `answerPlanReadFrom` still lacked read-target kind and quality shortcuts. Agents could use runbook/executor/handoff shortcuts, but plan-first consumers still had to parse nested read targets before judging the read source. | Existing agent alignment fixture, public type fixture, and static plan/answer shortcut scoring. | P2 | Add plan-level read-target kind, count, score, primary, and reason shortcuts to full/text/compact/brief/public output and score them in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves execution/answer plan read-current routing without browser work. |
+| G77 | `pageDecisionReadFrom` and `staticReadinessReadFrom` still lacked read-target kind and quality shortcuts. Page/readiness decision consumers could see the path but still had to parse nested read targets before judging what kind of source was selected and why. | Existing agent alignment fixture, public type fixture, and page decision static scoring. | P2 | Add decision-level read-target kind, count, score, primary, and reason shortcuts to full/text/compact/brief/public output and score page-decision shortcuts in static comparison. | Landed. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves page/readiness decision routing without browser work. |
+| G78 | Generic `semanticTopKeyboardShortcut*` selected the first keyboard item, so an earlier `accesskey` link could hide the first actual `aria-keyshortcuts` target from shallow brief output. Agents then had to parse `semanticSummary.keyboardItems` before targeting the keyboard command. | Existing keyboard shortcut brief fixture with accesskey before `aria-keyshortcuts`. | P2 | Add `semanticTopAriaKeyShortcut*` to full/text/compact/brief/public output and score it in static comparison. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves keyboard command handoff without browser work. |
+| G79 | JavaScript app shells with root/app containers and script bundles were classified as generic `no-inspectable-content` or `browser-required`. Agents could not tell a likely client-rendered app from a blank response or extraction failure without reading prose. | Focused app-shell fixture with empty `#root` and script bundle. | P2 | Add `CLIENT_RENDERED`, `browserHtmlReasonCode: client-rendered`, and `staticReadinessReasonCode: client-rendered` for app shells while preserving generic empty-page behavior. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No estimate increase; improves browser fallback clarity without browser work. |
+| G80 | Owned rowgroup cells are included in table/grid sample arrays, but shallow consumers still have to parse `semanticTopTableSampleCellRefs` to find the first `aria-owns` virtualized cell. | Existing compact agent table fixture with `aria-owns="owned-rows"` and a virtual row at `aria-rowindex="50"`; fifteenth local browser fixture `owned-grid-rowgroup` verifies browser-side and `--agent-brief` parity for the owned virtual rowgroup. | P5 | Add `ownedTarget` to owned sample-cell refs and expose first-owned-sample-cell shortcuts in full, compact, brief, public types, static comparison scoring, and local browser fixture evidence. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, browser fixture gate, README test, diff check, and process check pass. | Semantic table/grid parity 85% -> 88%; overall holds at 99.2%. |
+| G81 | Form choices expose first-field name/type/label/required/selector shallowly, but placeholder, default value, and option labels remain nested under `formChoices[0].fields[0]`. Form-filling agents may need those hints before choosing a query or deciding whether a select-like first field is usable. | Existing form fixture with a search placeholder plus public type fixture for value/options. | P2 | Add first-field placeholder/value/options shortcuts to full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; strengthens form-filling handoff. |
+| G82 | Page form fields and first-field shortcuts still drop HTML input constraints such as `autocomplete`, `inputmode`, `pattern`, `min`, `max`, `step`, `minlength`, and `maxlength`, even though semantic field shortcuts expose similar hints. Form-filling agents then have to fall back to semantic field payloads or guess validation rules. | Existing form fixture with search input constraints. | P2 | Preserve these input constraints in page form fields, agent form choices, compact form refs, first-field shortcuts, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; strengthens form-fill reliability. |
+| G83 | Form choices and first-field shortcuts still drop disabled, readonly, and invalid state even when the form control is not safe to fill. Agents then need to join semantic field state before deciding whether to type, skip, or report a blocked form. | Existing form fixture with `aria-disabled`, `readonly`, and `aria-invalid`. | P2 | Preserve disabled/readonly/invalid in page form fields, agent form choices, compact form refs, first-field shortcuts, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; strengthens form-fill safety. |
+| G90 | First sitelink shortcuts exposed title, URL, and selector, but not copy-ready command fields. Search or failed-open recovery agents could identify a relevant sublink but still had to parse nested result data or compose a direct page-check command before opening it. | Existing search sitelink fixtures, source-search public type fixture, and static fixture scoring. | P2 | Add first-sitelink command and command-args shortcuts for top result, selected source-search result, and first alternate source-search result in full, compact, brief, text, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; improves direct sublink handoff without browser work. |
+| G91 | Top-level first-sitelink shortcuts help the first choice, but agents iterating `resultChoices`, handoff result choices, or source-search alternate choices still had to compose commands from each nested sitelink URL. | Existing sitelink search fixture plus public source-search choice type fixture. | P2 | Preserve direct page-check command and command-args fields on compact agent sitelink objects while leaving raw search result arrays unchanged. | Landed. | Focused CLI/public type tests, static fixture gate, readiness audit, typecheck, README test, diff check, and process check pass. | No headline increase; strengthens multi-choice sublink routing without browser work. |
+| G84 | Form choices expose submit text but not submit button type, name, value, disabled state, or selector. Form execution loops may need to target or skip a submit control without parsing semantic button output. | Existing form fixture with a named submit button plus public type fixture for disabled state. | P2 | Preserve submit button detail in page forms, agent form choices, top form shortcuts, compact form refs, public types, and static comparison scoring. | Landed. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check pass. | No headline increase; strengthens form execution handoff. |
+| G2 | Browser accessibility-tree comparison may reveal signals that static HTML cannot safely infer. | Future sequential fixture comparison only; no broad browser run allowed. | P1 after evidence | Track first, then classify as `implement`, `browser-only`, or `defer`. | Watch. | Add the smallest fixture command here before running any browser-backed check; run `pnpm check:processes` afterward. | Unknown until observed. |
+
+## Current Queue
+
+Use this queue when resuming the work. Do not start a browser-backed check while
+any earlier non-browser item can still reduce uncertainty.
+
+| Step | Status | What will change | Validation | Percent effect |
+| ---: | --- | --- | --- | ---: |
+| 1 | Done | Keep progress tracking detailed enough to show current work, planned work, completion signals, and scope-expansion rules. | `pnpm exec vitest run tests/readme.test.ts`, `git diff --check`, `pnpm check:processes`. | Keeps docs hygiene at 90%. |
+| 2 | Done | Expose top result/source snippets as shallow shortcuts so agents can inspect the first choice without nested parsing. | Focused CLI/public type tests, typecheck, static gates, README test, diff check, process check. | +1% search handoff. |
+| 3 | Done | Add machine-readable static-readiness reason codes for hidden data, low-content source-link recovery, thin content, and browser-required fallback. | Focused CLI/public type tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 4 | Done | Expose first list-item text, role, position, state, and selector as shallow shortcuts so agents do not parse `semanticTopListItemRefs[0]`. | Focused CLI/public type tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 5 | Done | Preserve top semantic selectors and core state booleans in `--agent-brief` so executor loops do not need the full payload for element targeting. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 6 | Done | Preserve top search result provenance in `--agent-brief` so executor loops do not need full output for source hints, first sitelink, or source-choice reason. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% search handoff. |
+| 7 | Done | Preserve top author attribution and hidden read-target rationale in `--agent-brief` so page-check executor loops keep attribution and fallback reasoning. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 8 | Done | Preserve top data-table path and selector in `--agent-brief` so table evidence can be targeted without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 9 | Done | Preserve top semantic table/list container selectors in `--agent-brief` so container-level actions do not require full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 10 | Done | Preserve top actionable semantic selectors in `--agent-brief` so link/button/image targets can be used without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 11 | Done | Preserve top table/list accessible names and sample-cell spans in `--agent-brief` so label and merged-cell context survive brief handoff. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 12 | Done | Preserve top interactive/focusable selectors in `--agent-brief` so keyboard and interaction targets can be used without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 13 | Done | Preserve `canUseFetchedHtml` in `--agent-brief` so static-usable pages do not need full output or browser capture. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 14 | Done | Preserve the top FAQ answer in `--agent-brief` so static FAQ evidence can answer without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 15 | Done | Preserve top code block text in `--agent-brief` so static command snippets can be used without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 16 | Done | Preserve top resource titles in `--agent-brief` so resource links can be classified without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 17 | Done | Preserve top keyboard shortcut selectors in `--agent-brief` so keyboard-accessible targets can be reached without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% semantic accessibility. |
+| 18 | Done | Preserve top structured page evidence text, labels, sources, and selectors in `--agent-brief` so page-check executor loops can classify and target evidence without full output. | Focused CLI tests, typecheck, static gates, README test, diff check, process check. | +1% page handoff. |
+| 19 | Done | Compare one semantic table/list/control fixture against browser-side output, sequentially only. | `pnpm check:processes`, `pnpm compare:browser:fixture`, `pnpm check:processes`. | +5% browser parity research. |
+| 20 | Done | Fix browser-side ARIA state extraction and compacted-list sample gaps found by the first parity fixture. | Focused extractor/CLI tests plus browser fixture gate. | +1% semantic accessibility. |
+| 21 | Done | Recalculate the overall estimate after landed browser parity evidence. | Update this file in the same commit as the evidence. | Overall 89% -> 90%. |
+| 22 | Done | Add the second local browser fixture for expanded popup controls, current links, modal dialogs, and live regions. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`. | Browser parity research 65% -> 70%. |
+| 23 | Done | Preserve top modal/live state shortcuts in full and brief agent output. | Typecheck, focused CLI/public type tests, browser fixture gate. | Semantic accessibility 89% -> 90%; overall 90% -> 91%. |
+| 24 | Done | Add the third local browser fixture for combobox active-descendant and selected/current option state. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`. | Browser parity research 70% -> 75%. |
+| 25 | Done | Preserve top selected-choice shortcuts in full and brief agent output. | Typecheck, focused CLI/public type tests, browser fixture gate. | Semantic accessibility 90% -> 91%; overall 91% -> 92%. |
+| 26 | Done | Add the fourth local browser fixture for selected tab and controlled panel targeting. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`. | Browser parity research 75% -> 80%. |
+| 27 | Done | Preserve selected-choice controlled-target shortcuts in full and brief agent output. | Typecheck, focused CLI/public type tests, browser fixture gate. | Semantic accessibility 91% -> 92%; overall 92% -> 93%. |
+| 28 | Done | Add the fifth local browser fixture for selected gridcell state and table/grid position. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`. | Browser parity research 80% -> 85%. |
+| 29 | Done | Preserve selected table/grid cell shortcuts in full and brief agent output. | Typecheck, focused CLI/public type tests, browser fixture gate. | Semantic accessibility 92% -> 93%; overall 93% -> 94%. |
+| 30 | Done | Use semantic table/grid evidence directly instead of browser handoff when static accessibility evidence is sufficient. | Typecheck, focused CLI test, browser fixture gate, process checks. | Page handoff 94% -> 95%; overall 94% -> 95%. |
+| 31 | Done | Expose likely-official source choice hints without requiring nested source-choice parsing. | Typecheck, focused CLI/public type tests, readiness audit, static fixture gate, README test, diff check, process check. | Search handoff 86% -> 87%; overall 95% -> 96%. |
+| 32 | Done | Preserve high-value primary execution/read-target shortcuts in `--agent-brief` after compact-vs-brief comparison showed shallow losses. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, process check. | Page handoff 95% -> 96%; overall remains 96%. |
+| 33 | Done | Preserve shallow decision and diagnostic fields in `--agent-brief` after compact-vs-brief comparison showed remaining scalar losses. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, process check. | Search handoff 87% -> 88%; overall remains 96%. |
+| 34 | Done | Preserve semantic count scalars and top outline selector in `--agent-brief` after compact-vs-brief comparison showed remaining semantic scalar losses. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, process check. | Semantic accessibility 93% -> 94%; overall remains 96%. |
+| 35 | Done | Preserve remaining small control-plane scalars in `--agent-brief` and classify the remaining compact-vs-brief differences as intentionally omitted nested payloads. | Typecheck, focused CLI tests, readiness audit, static fixture gate, README test, diff check, process check. | Closes current non-browser brief parity cleanup; overall remains 96%. |
+| 36 | Done | Add the sixth local browser fixture for range/slider value state after the non-browser scalar cleanup stayed closed. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 85% -> 88%; overall remains 96%. |
+| 37 | Done | Add the seventh local browser fixture for named busy live status after the six-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 88% -> 90%; overall remains 96%. |
+| 38 | Done | Add the eighth local browser fixture for invalid field and error-message state after the seven-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 90% -> 92%; overall remains 96%. |
+| 39 | Done | Add the ninth local browser fixture for sorted columnheader state after the eight-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 92% -> 94%; overall remains 96%. |
+| 40 | Done | Add the tenth local browser fixture for multiselect listbox state after the nine-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 94% -> 95%; overall remains 96%. |
+| 41 | Done | Add the eleventh local browser fixture for drag/drop state after the ten-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 95% -> 96%; overall remains 96%. |
+| 42 | Done | Add the twelfth local browser fixture for disabled/readonly field state after the eleven-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 96% -> 97%; overall remains 96%. |
+| 43 | Done | Add the thirteenth local browser fixture for mixed checkbox state after the twelve-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 97% -> 98%; overall remains 96%. |
+| 44 | Done | Add the fourteenth local browser fixture for field details relation text after the thirteen-fixture gate stayed clean. | Pre/post `pnpm check:processes`; `pnpm compare:browser:fixture`; one browser-backed command per run. | Browser parity research 98% -> 99%; overall remains 96%. |
+| 45 | Done | Add the smallest real-page static handoff smoke gate before any broader `agent-browser` comparison slice. | `pnpm readiness:real-page-smoke`; pre/post `pnpm check:processes`. | Overall 96% -> 97%; `agent-browser` comparison still remains. |
+| 46 | Done | Add one carefully bounded `agent-browser` comparison slice on `https://example.com` after process state is clean. | Pre/post `pnpm check:processes`; `pnpm readiness:agent-browser-smoke`. | Overall 97% -> 98%; broader comparison still remains. |
+| 47 | Done | Broaden the `agent-browser` smoke to one richer documented target, `https://books.toscrape.com/`, after process state is clean. | Pre/post `pnpm check:processes`; `pnpm readiness:agent-browser-smoke`. | Overall 98% -> 98.5%; broader comparison still remains. |
+| 48 | Done | Broaden the `agent-browser` smoke to the link-heavy listing target `https://news.ycombinator.com`, fix the numeric-label spacing normalization gap it exposed, and keep HN floors tolerant of live ranking/comment churn. | Pre/post `pnpm check:processes`; `pnpm readiness:agent-browser-smoke`. | Overall 98.5% -> 99%; broader comparison still remains. |
+| 49 | Done | Broaden the `agent-browser` smoke to the government index/search target `https://www.gov.uk/foreign-travel-advice`. | Pre/post `pnpm check:processes`; `pnpm readiness:agent-browser-smoke`. | Overall 99% -> 99.2%; broader comparison still remains. |
+| 50 | Done | Probe the Korean Wikipedia text-heavy document page and align live compare's comparable-role filtering with static compare after the page exposed StaticText/listmarker-dominated scoring. | Pre/post `pnpm check:processes`; one browser-backed command only. | Overall holds at 99.2%; gap is tracked as G49 instead of hidden. |
+| 51 | Done | Decide the first G49 policy: strict text/content recall remains reported, but weighted readiness score uses structural content recall. | Typecheck, readiness audit, static fixture gate, README test, and process checks. | Overall holds at 99.2%. |
+| 52 | Done | Re-measure `ko.wikipedia.org/wiki/대한민국` under the new score and add a separate text-heavy smoke gate instead of folding it into the main overlap smoke. | Pre/post `pnpm check:processes`; `pnpm readiness:agent-browser-text-heavy-smoke`. | Overall holds at 99.2%; text-heavy target is now re-checkable. |
+| 53 | Done | Decide whether `--agent` needs extra text chunk handoff for text-heavy pages beyond the structural readiness score; current fix keeps inline snippets capped but counts all focused content for readability/static-handoff scoring. | Typecheck, focused CLI regression test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%; text-heavy static handoff is less likely to be under-scored. |
+| 54 | Done | Preserve multiple source-search alternate candidates as a top-level compact/brief array so failed opened-result recovery does not require nested sourceSearch parsing. | Typecheck, focused CLI regression test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Search handoff 88% -> 89%; overall holds at 99.2%. |
+| 55 | Done | Preserve generic top-choice rank, source, method, selector, and reason so cross-kind executor loops do not need to parse result/source/form/action-target-specific payloads first. | Typecheck, focused agent CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 56 | Done | Preserve recommended result path and source score so agents can jump to `recommendedResult` and compare source quality without joining nested choices. | Typecheck, focused recommendation CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 57 | Done | Preserve top answer-evidence score so answer loops can judge evidence quality without parsing nested `answerEvidence`. | Typecheck, focused agent CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 58 | Done | Preserve recommended result metadata inside search decision shortcuts so decision-only loops can judge the chosen result without joining nested payloads. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 59 | Done | Preserve page-decision readability, evidence, and source quality scalars so page-check loops can judge static-read confidence without nested parsing. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 60 | Done | Preserve executor/handoff target metadata so execution loops can judge target quality without nested parsing. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 61 | Done | Preserve primary action target metadata so primary-only loops can judge target quality without nested parsing. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 62 | Done | Preserve alternative action target metadata so fallback loops can judge target quality without nested parsing. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 63 | Done | Preserve top action target metadata so top-action-only loops can judge target quality without nested parsing. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 64 | Done | Preserve action expected outcomes so action-only loops can execute/read/stop without joining executor or handoff summaries. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 65 | Done | Preserve top/alternative action execution details so action-only loops do not parse nested actions for command provenance, rank, or open-result index. | Focused CLI test, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Overall holds at 99.2%. |
+| 66 | Done | Preserve source-search selected/alternate source, command, source score, relevance, and likely-official shortcuts so failed-open recovery can choose a fallback without nested parsing. | Typecheck, focused source-search CLI tests, public type test, readiness audit, static fixture gate, README test, diff check, and process check. | Search handoff 89% -> 90%; overall holds at 99.2%. |
+| 67 | Done | Add regression coverage for source-search alternate-choice quality fields so multi-candidate recovery cannot silently lose source, command, score, relevance, or official status. | Typecheck, focused source-search CLI tests, public type test, static fixture gate, README test, diff check, and process check. | No estimate increase; protects current search handoff coverage. |
+| 68 | Done | Preserve top source-choice command strings and keep command strings in compacted source-search alternate choices so page source-link and failed-open recovery loops can execute without nested parsing. | Typecheck, focused CLI tests, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens page/search handoff. |
+| 69 | Done | Preserve generic top-choice and top-result command strings so common result/source executor loops can execute without kind-specific nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, README test, diff check, and process check. | No estimate increase; strengthens generic top-choice handoff. |
+| 70 | Done | Preserve generic top-choice source type, source score, relevance, and likely-official shortcuts so common result/source executor loops can judge quality without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens generic top-choice quality handoff. |
+| 71 | Done | Preserve generic top-choice host and snippet shortcuts so common result/source executor loops can inspect target context without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens generic top-choice context handoff. |
+| 72 | Done | Preserve generic top-choice open-result, recommended, and primary shortcuts so common result/source executor loops can execute the selected choice without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens generic top-choice execution handoff. |
+| 73 | Done | Preserve top source-choice kind, rank, text, selector, relevance, matched term, and find match so page source-link loops can target and explain the chosen link without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens page source-link handoff. |
+| 74 | Done | Preserve source-search selected/alternate snippet, date, matched term, find match, and first sitelink shortcuts so failed-open recovery can compare previews without nested parsing. | Typecheck, focused source-search CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens failed-open recovery handoff. |
+| 75 | Done | Preserve read-target kind on each read target and as `topReadTargetKind` so read-current loops can branch without path-string parsing. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens read-current routing. |
+| 76 | Done | Preserve `bestReadTargetKind` so best read-target loops can branch without path-string parsing when best and top targets differ. | Typecheck, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens best read-current routing. |
+| 77 | Done | Preserve `runbookReadTargetKind`, `nextReadTargetKind`, `executorReadTargetKind`, and `handoffReadTargetKind` so each stage can branch without path-string parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens stage-specific read-current routing. |
+| 78 | Done | Preserve stage-specific read-target count, score, primary, and reason shortcuts so each stage can judge read-source quality without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens stage-specific read-current quality routing. |
+| 79 | Done | Preserve `executionPlanReadTarget*` and `answerPlanReadTarget*` shortcuts so plan-first agents can branch and judge read-source quality without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens plan-level read-current routing. |
+| 80 | Done | Preserve `pageDecisionReadTarget*` and `staticReadinessReadTarget*` shortcuts so page/readiness decision consumers can branch and judge read-source quality without nested parsing. | Typecheck, focused agent CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens decision-level read-current routing. |
+| 81 | Done | Preserve the first `aria-keyshortcuts` target separately from the first generic keyboard item so accesskey-first pages keep shortcut keys and selector shallow. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; strengthens keyboard command handoff. |
+| 82 | Done | Classify JavaScript app shells as client-rendered browser handoff instead of generic no-inspectable/browser-required fallback. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, and process check. | No estimate increase; clarifies browser fallback routing. |
+| 83 | Done | Preserve first form-field placeholder, value, and options so form-filling loops do not parse nested form field payloads before choosing input text or select values. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens form handoff. |
+| 84 | Done | Preserve form input constraints in form field payloads and first-field shortcuts so form-filling loops can respect validation hints without joining semantic field payloads. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens form handoff. |
+| 85 | Done | Preserve form field disabled, readonly, and invalid state so form-filling loops can skip blocked controls without joining semantic field payloads. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens form handoff. |
+| 86 | Done | Preserve submit button detail so form execution loops can target or skip the submit control without joining semantic button payloads. | Typecheck, focused CLI test, public type test, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens form handoff. |
+| 87 | Done | Make unnamed implicit submit button selectors match real DOM instead of targeting a missing `type` attribute. | Typecheck, focused CLI test, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form execution targeting. |
+| 88 | Done | Apply submit button form overrides to form handoff so action URL, method, URL template, and submit metadata match the clicked submit control. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form execution correctness. |
+| 89 | Done | Preserve hidden form field count and first hidden field refs so browser-equivalent form submission keeps required hidden parameters. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form submission completeness. |
+| 90 | Done | Preserve checkbox/radio checked state in form fields and first-field shortcuts so submission handoff knows whether a value is included. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form submission correctness. |
+| 91 | Done | Preserve select selected/default option label and submission value in form fields and first-field shortcuts. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form submission correctness. |
+| 92 | Done | Preserve textarea default text content as the form field value. | Typecheck, focused CLI test, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form submission completeness. |
+| 93 | Done | Include image submit controls in submit handoff with accessible text, name, value, and selector. | Typecheck, focused CLI test, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form execution targeting. |
+| 94 | Done | Preserve form-level execution metadata such as target, enctype, accept-charset, novalidate, id, and name. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens form execution fidelity. |
+| 95 | Done | Expose top action-target encoding type so action execution can choose request content type without nested parsing. | Typecheck, focused CLI/public type tests, static fixture gate, README test, diff check, process check. | No headline increase; strengthens action target execution handoff. |
+| 96 | Done | Preserve selected table-cell headers and spans as shallow shortcuts so active table/grid cells keep accessible context without parsing sample-cell refs. | Typecheck, focused CLI/public type tests, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens table/grid handoff. |
+| 97 | Done | Preserve first sample-cell spans and first-owned sample cells in text output so text handoff matches JSON shortcut coverage. | Typecheck, focused text CLI tests, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; closes text-output parity gaps. |
+| 98 | Done | Preserve selected owned table-cell `ownedTarget` in full, brief, text, public types, readiness audit, and static scoring. | Typecheck, focused CLI/public type tests, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; strengthens virtualized table/grid handoff. |
+| 99 | Done | Preserve first sample-cell `ownedTarget` in text output when the first sample comes from an `aria-owns` rowgroup. | Typecheck, focused text CLI test, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; closes text-output parity gap. |
+| 100 | Done | Make text `semanticTopTable`, `semanticTopList`, and `semanticTopField` print role and name as named fields instead of positional `role:name` tokens. | Focused text CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves text-output readability for table/list/form handoff. |
+| 101 | Done | Make text semantic description, value, relation, choice, selected-choice, and state lines print role, name, relation, and target as named fields instead of positional tokens. | Focused text CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves text-output readability for relation/state handoff. |
+| 102 | Done | Make text table header and sample-cell shortcuts print cell/header content as `text="..."` instead of a positional token after the path. | Focused table text CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves table/grid text-output readability for accessibility-tree handoff. |
+| 103 | Done | Make text list item refs, first list item fields, and first owned table target fields use explicit named fields such as `text="..."`, `target=...`, and `name="..."`. | Focused list/table text CLI tests, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves list and virtualized-table text handoff readability. |
+| 104 | Done | Make aggregate text table `headerRefs`, `ownedRefs`, and `sampleRefs` use explicit named fields instead of compact `text:role` and `text@rNcM` tokens. | Focused table text CLI test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves aggregate table/grid handoff readability. |
+| 105 | Done | Make text `semanticTopUnavailable` print `tag=...` and quoted names as named fields instead of positional tag/name tokens. | Focused unavailable semantic text CLI test, public type test, typecheck, static fixture gate, readiness audit, README test, diff check, process check. | No headline increase; improves unavailable embedded/accessibility-target handoff readability. |
+
+When research expands:
+
+1. Add the new signal or gap to the ledger before implementation.
+2. Lower or hold the relevant percentage until the gap is classified.
+3. Record the smallest validation command that proves the decision.
+4. Keep browser-backed validation sequential and run `pnpm check:processes`
+   afterward.
+
+## Current Focus
+
+1. Browser parity research: compare static semantic output with browser
+   accessibility output on a small fixture set, one command at a time.
+2. Gap triage: classify each missing signal as `implement`, `document as
+   browser-only`, or `defer`.
+3. Shortcut selection: only add top-level fields when they reduce agent
+   routing work or reduce unnecessary browser handoff.
+4. Safety verification: keep process cleanup checks visible in the workflow so
+   the previous server freeze does not repeat.
+
+## Research Rules
+
+- Research goals may expand when a new useful browser accessibility-tree signal
+  is found.
+- New scope must be tracked here with priority, expected value, and status.
+- Percentages are estimates of current usefulness, not a promise that the whole
+  research space is closed.
+- Browser-backed checks must stay sequential and limited. Prefer static
+  fixtures, type checks, and non-browser gates first.
+- If a task requires browser execution, run the smallest possible case and
+  verify cleanup with `pnpm check:processes` afterward.
+
+## Done
+
+- Kept the root README short and moved detailed docs under `docs/`.
+- Added agent handoff fields for search results, forms, source links, barriers,
+  hidden signals, semantic links, buttons, fields, choices, and state.
+- Added action target state: disabled, pressed, expanded, popup, and controls.
+- Added field-level `aria-details` and `aria-errormessage` shortcuts with
+  resolved text.
+- Added list item refs with role, position, set size, current, selected, and
+  expanded state for list/tree/menu navigation.
+- Added top semantic list first-item shortcuts so agents can inspect the first
+  item's text, role, position, state, and selector without parsing item refs.
+- Preserved top semantic selectors and core state booleans in `--agent-brief`
+  so brief executor loops can target fields, choices, relations, and stateful
+  nodes without requesting full compact output.
+- Preserved top search-choice provenance in `--agent-brief` so brief executor
+  loops keep source hints, first sitelink, and source-choice reason.
+- Preserved top author attribution and hidden read-target rationale in
+  `--agent-brief` so brief page-check loops keep source and fallback context.
+- Preserved top data-table path and selector in `--agent-brief` so brief
+  page-check loops can target exact table evidence without full output.
+- Preserved top semantic table/list container selectors in `--agent-brief` so
+  brief executor loops can target semantic containers directly.
+- Preserved top link, in-page link, button, and image selectors in
+  `--agent-brief` so brief executor loops keep direct actionable targets.
+- Preserved top semantic table/list names and first sample-cell spans in
+  `--agent-brief` so brief executor loops keep labels and merged-cell context.
+- Preserved top interactive and focusable selectors in `--agent-brief` so
+  brief executor loops keep exact keyboard and interaction targets.
+- Preserved `canUseFetchedHtml` in `--agent-brief` so brief executor loops can
+  use static output directly when browser capture is unnecessary.
+- Preserved top FAQ answers in `--agent-brief` so brief executor loops can use
+  static FAQ question-answer evidence without full output.
+- Preserved top code block text in `--agent-brief` so brief executor loops can
+  use static command snippets and examples without full output.
+- Preserved top resource titles in `--agent-brief` so brief executor loops can
+  classify linked resources without full output.
+- Preserved top keyboard shortcut selectors in `--agent-brief` so brief
+  executor loops can target accesskey and aria-keyshortcuts elements directly.
+- Preserved top structured page evidence text, labels, sources, and selectors
+  in `--agent-brief` so brief executor loops can classify media, sections,
+  provenance, offers, datasets, identities, timelines, contacts, and navigation
+  metadata without full output.
+- Added `topAnswerEvidenceScore` so answer handoff can judge selected evidence
+  quality without parsing nested `answerEvidence`.
+- Added search-decision recommended result metadata so agents can trust the
+  selected search result without joining nested payloads.
+- Added page-decision readability/evidence/source quality shortcuts so brief
+  page-check loops can judge static-read confidence without nested parsing.
+- Added executor/handoff target metadata shortcuts so brief execution loops can
+  judge target quality without nested parsing.
+- Added top/primary/alternative action read-target metadata shortcuts so
+  action-only loops can see read target kind, count, score, primary state, and
+  reason without joining the full `readTargets` array.
+- Added primary action target metadata shortcuts so primary-only loops can judge
+  target quality without nested parsing.
+- Added alternative action target metadata shortcuts so fallback loops can judge
+  target quality without nested parsing.
+- Added top action target metadata shortcuts so top-action-only loops can judge
+  target quality without nested parsing.
+- Added action expected-outcome shortcuts so action-only loops can decide the
+  intended result without joining executor or handoff summaries.
+- Added top/alternative action execution-detail shortcuts so action-only loops
+  can keep command provenance, rank, and open-result context shallow.
+- Added source-search selected/alternate quality shortcuts so failed-open
+  recovery can compare command, source, source score, relevance, and
+  likely-official status without nested parsing.
+- Added regression coverage for source-search alternate-choice quality fields
+  so multi-candidate recovery keeps quality metadata shallow.
+- Added top source-choice command shortcuts and preserved command strings in
+  compacted source-search alternate choices.
+- Added generic top-choice and top-result command shortcuts for common
+  result/source executor loops.
+- Added generic top-choice quality shortcuts for source type, score, relevance,
+  and likely-official status.
+- Added generic top-choice host and snippet shortcuts for common result/source
+  executor loops.
+- Added generic top-choice selection shortcuts for open-result, recommended,
+  and primary state.
+- Added top source-choice detail shortcuts for kind, rank, text, selector,
+  relevance, matched term, and find match.
+- Added source-search selected/alternate preview shortcuts for snippet, date,
+  matched term, find match, and first sitelink metadata.
+- Added read-target kind shortcuts so executor loops can classify current-payload
+  read targets without parsing path strings.
+- Added best read-target kind shortcut for best-target routing.
+- Added table cell header refs so sampled table/grid cells keep resolved
+  `headers` context.
+- Added top semantic table first-header and first-sample-cell shortcuts so
+  agents can inspect row, column, span, header, and selector context without
+  parsing the full table refs array.
+- Added semantic table sample-cell ref paths so agents can jump back to the
+  exact `agent.semanticSummary.tableItems[*].sampleCellRefs[*]` item.
+- Added semantic table sample-cell row/column header shortcuts so agents can
+  distinguish browser-tree table context by header role, not only header text.
+- Added semantic table header refs and first-header shortcuts so agents can
+  inspect columnheader/rowheader path, role, row/column index, sort state, and
+  selector without parsing the full table payload.
+- Added semantic table ownership refs and first-owned shortcuts so agents can
+  detect `aria-owns` rowgroups or virtualized table ownership from the table
+  summary without parsing generic relations.
+- Added top data-table navigation shortcuts for header count, first header,
+  first row, first cell, and selector so agents can inspect table shape without
+  reading the full table payload.
+- Added top hidden signal selectors so agents can jump from hydration/API/client
+  hints back to the exact script or link source before deciding on browser
+  capture.
+- Added hidden signal group counts for hydration, API endpoints, client state,
+  and app hints so agents can see which static data channels exist without
+  scanning nested arrays.
+- Added top hidden signal group shortcuts for hydration, API endpoints, client
+  state, and app hints so agents can jump directly to the most useful static
+  data source in each group.
+- Added static-readiness shortcuts so agents can tell whether fetched HTML is
+  usable as direct content, structured data, hidden app data, thin output, or a
+  browser-required case before opening a browser.
+- Added `browserHtmlReason` so agents can explain why browser capture is needed
+  without digging through nested answer-plan gaps.
+- Added `browserHtmlReasonCode` so agents can branch on browser fallback causes
+  such as no inspectable content, HTTP errors, fetch failures, challenge,
+  login, paywall, retry actions, and browser interaction.
+- Added live browser interaction reason codes so captured barrier pages can
+  expose `challenge`/interaction fallback causes even when no new HTML capture
+  is required.
+- Added client-rendered fallback reason codes so JavaScript app shells are
+  distinct from blank no-inspectable pages.
+- Added top-level browser HTML capture shortcuts for action, operation, command
+  args, capture file, and capture script so agents can execute fallback without
+  parsing nested runbook/execution-plan objects.
+- Kept semantic unavailable target role/name/selector shortcuts in brief output
+  so agents can identify browser-inaccessible static targets without expanding
+  the full semantic summary.
+- Added top-level answer citation count and first citation id shortcuts so
+  agents can cite ready answers without parsing `answerPlan.useCitationIds`.
+- Added non-browser fixture gates and readiness audits for repeatable checks.
+- Added process-safety guidance: run tests and browser-backed checks
+  sequentially, and verify process cleanup.
+- Added top-level search decision count shortcuts for relevance, official
+  result, and find-match totals so agents can judge result quality without
+  parsing the nested `searchDecision` object.
+- Added result/source choice host shortcuts so agents can compare domains and
+  provenance without parsing URLs before opening a result.
+- Added source-search failure shortcuts so agents can see the selected result's
+  failed URL, error code, status, and reason without joining separate error and
+  source-search objects.
+- Added source-search failure kind and retryable shortcuts so agents can choose
+  alternate results or retry the selected URL without parsing raw error/status.
+- Added rate-limit-aware source-search failure classification so HTTP 429
+  selected-result failures are retryable instead of generic client errors.
+- Preserved source-search failure `Retry-After` hints so retryable selected
+  result failures can be scheduled without nested response parsing.
+- Added top-level `sourceSearchAlternateChoices` so failed opened-result
+  recovery can inspect multiple fallback candidates without parsing nested
+  `sourceSearch.alternateResults`.
+- Added static gate scoring for selected/current table-cell state so table/grid
+  parity cannot regress silently when matching browser accessibility trees.
+- Included `aria-owns` rowgroup descendants in table row/cell/header samples so
+  static table summaries better match browser accessibility-tree ownership.
+- Added generic `topChoice*` execution metadata for rank, source, method,
+  selector, and reason so cross-kind routing can avoid nested choice parsing.
+- Added recommended-result path and source-score shortcuts so recommendation
+  handoff can cite `recommendedResult` directly and compare source quality.
+- Counted all focused page content toward `pageCheck.contentLength` while
+  keeping previews/evidence capped, so text-heavy fetched pages are not
+  under-scored as thin content.
+- Tightened readiness audit coverage for controls target shortcuts so
+  interactive, focusable, button, field, selected-choice, and state outputs
+  must keep target role/name/selector fields wired through types and tests.
+- Added static declarative shadow DOM extraction so SSR Web Components can
+  expose shadow controls while ordinary inert templates remain pruned.
+- Added agent-brief coverage for declarative shadow DOM controls so the static
+  extractor gain is verified at the subagent handoff layer.
+- Added declarative shadow slot projection so slotted controls replace fallback
+  controls in static and agent-brief output.
+- Added slotted text handling for declarative shadow DOM accessible names so
+  static action labels match the composed accessibility tree.
+- Preserved referenced ID targets during pruning so slotted field labels keep
+  selector shortcuts in agent-brief handoff.
+- Made `label for` name lookup slot-aware so declarative shadow fields use
+  slotted label text in static and agent-brief output.
+- Made custom-element host action names use projected declarative shadow slot
+  text instead of raw light DOM or fallback text.
+- Kept description IDREF text out of name-from-content while preserving it in
+  semantic description shortcuts for host actions.
+- Added a synthetic Web Component target to the static fixture gate so composed
+  tree handoff regressions are covered by non-browser comparison scoring.
+- Added first-sitelink command and command-args shortcuts for top search result
+  and source-search recovery handoff.
+- Restored the full CLI regression gate by aligning old expectations with the
+  current direct-URL command and form/action-target handoff output.
+- Preserved command and command-args fields on compact agent sitelink objects
+  so choice-array consumers can open sublinks directly.
+- Added source-search failure command shortcuts so failed-open recovery can
+  retry the selected result directly from the failure context.
+- Added first-official search-decision shortcuts so agents can spot and open
+  the first likely official result without parsing nested search results.
+- Added top provenance follow-up command shortcuts so agents can open DOI or
+  identifier URLs directly from page-check summary fields.
+- Added top resource follow-up command shortcuts so agents can open feed, PDF,
+  manifest, license, and download resources from page-check summary fields.
+- Added top author follow-up command shortcuts so agents can open author or
+  profile pages directly from page-check summary fields.
+- Hardened browser-backed comparison cleanup so `compare.ts` and
+  `compare-static.ts` serialize `agent-browser` use and close opened sessions
+  from `finally` blocks even when snapshot or rendered-HTML fallback fails.
+- Added normalized timeline date shortcuts (`isoDate`, `unixMs`,
+  `topTimelineIsoDate`, `topTimelineUnixMs`) so page-check agents can judge
+  freshness and event timing without parsing date strings.
+- Added normalized offer price shortcuts (`priceAmount`,
+  `topOfferPriceAmount`) so page-check agents can compare product or plan
+  prices without parsing localized price text first.
+- Added dataset coverage and creator shortcuts
+  (`topDatasetTemporalCoverage`, `topDatasetSpatialCoverage`,
+  `topDatasetCreator`) so data-source handoff can judge scope and publisher
+  without opening nested dataset arrays.
+- Added normalized search-result date shortcuts (`dateIso`, `dateUnixMs`,
+  `recommendedDateIso`, `recommendedDateUnixMs`) so search agents can compare
+  freshness without parsing result snippets or titles.
+- Added normalized source-search selected/alternate date shortcuts so failed
+  opened-result recovery can compare freshness without nested result parsing.
+- Added normalized first-official search-decision date shortcuts so agents can
+  compare official-result freshness without nested search-result parsing.
+- Added recommended and first-official source type/hint shortcuts so agents can
+  compare source-profile rationale without nested search-result parsing.
+- Added source-search selected/alternate source type and hint shortcuts so
+  failed-open recovery can compare provenance without nested result parsing.
+- Added source-search selected/failure/alternate URL path/query shortcuts so
+  recovery loops can route and compare candidates without parsing full URLs.
+- Added selected/current list-item shortcuts so agents can identify active
+  nav, tab, menu, and list entries without parsing list item arrays.
+- Added current-link shortcuts so agents can identify active navigation links
+  even when the current link is not the first semantic link.
+- Added top/current semantic link URL path/query shortcuts so navigation
+  comparison and routing can avoid parsing full link URLs.
+- Added top in-page semantic link URL path/query shortcuts so anchor and
+  skip-link routing can compare targets without parsing full URLs.
+- Added submit-button `formaction` URL path/query shortcuts so form handoff can
+  compare overridden submit targets without parsing full URLs.
+- Added pagination and table-of-contents URL path/query shortcuts so navigation
+  follow-up handoff can compare page and section targets without URL parsing.
+- Added top resource URL path/query shortcuts so feed, PDF, manifest, and
+  license follow-up targets can be compared without parsing full URLs.
+- Added top media URL path/query shortcuts so image and social-preview
+  follow-up targets can be compared without parsing full URLs.
+- Added top embed and transcript URL path/query shortcuts so iframe, media,
+  caption, and transcript follow-up targets can be compared without parsing
+  full URLs.
+- Added top author and provenance URL path/query shortcuts so byline/profile
+  and DOI/source follow-up targets can be compared without parsing full URLs.
+- Added top dataset, distribution, and license URL path/query shortcuts so data
+  source, download, and license follow-up targets can be compared without
+  parsing full URLs.
+- Added top offer URL path/query shortcuts so product, plan, and purchase
+  follow-up targets can be compared without parsing full URLs.
+- Added citation and answer-evidence URL path/query shortcuts so verification
+  follow-up targets can be compared without parsing full URLs.
+- Added identity and contact URL path/query shortcuts so canonical site, logo,
+  profile, and support targets can be compared without parsing full URLs.
+- Added hidden technical URL path/query shortcuts so hydration, API, runtime,
+  manifest, mobile, and selected hidden-signal targets can be compared without
+  parsing full URLs.
+- Added `urlPath`/`urlQuery` directly to original `pageCheck.sourceLinks`,
+  `pageCheck.resources`, and `pageCheck.media` entries, with a static gate, so
+  agents reading the source arrays do not need to parse full URLs before routing
+  follow-up checks.
+- Extended the same original-array URL path/query contract to
+  `pageCheck.pagination`, `pageCheck.citations`, `pageCheck.embeds`,
+  `pageCheck.transcripts`, `pageCheck.authorLinks`, and
+  `pageCheck.provenance`, so read-target follow-up arrays preserve the same
+  routing evidence as their top shortcuts.
+- Extended original structured read-target arrays with URL path/query evidence
+  for `pageCheck.offers`, `pageCheck.datasets`, `pageCheck.identities`, and
+  `pageCheck.contactPoints`, including dataset distribution/license URLs and
+  identity logo/sameAs URLs, under the same static gate.
+- Added URL path/query evidence to nested `pageCheck.breadcrumbs[].items[]`
+  and `pageCheck.toc[].items[]`, so agents can inspect navigation trails and
+  in-page outlines without parsing item URLs or relying only on top shortcuts.
+- Added URL path/query evidence to original hidden and metadata URL arrays
+  (`hydration`, `apiEndpoints`, `runtime`, `appHints`, `mobileHints`, and
+  `metaFacts`), so agents reading those arrays get the same routing evidence as
+  top shortcuts.
+- Aligned public `AgentPage*` TypeScript types with original URL-bearing
+  `pageCheck` array output, so TypeScript consumers can read `urlPath` and
+  `urlQuery` without casts.
+- Added nested embed poster/source URL path/query evidence, so agents can
+  compare media fallback sources without parsing full video/audio URLs.
+- Added first-sitelink URL path/query shortcuts for top result, generic top
+  choice, and source-search selected/alternate handoff targets, so sublink
+  routing no longer requires parsing full first-sitelink URLs.
+- Added individual text rows for semantic first table headers and selected
+  table/grid cells, so text-only agents can read table position and header
+  context without parsing aggregate table rows.
+- Added individual text rows for selected/current semantic list items, so
+  text-only agents can route menu, tree, tab, and list state without parsing
+  aggregate list rows.
+- Added individual text rows for semantic current links, so text-only agents
+  can route current-page navigation state without parsing aggregate link rows.
+- Added individual text rows for semantic in-page links, so text-only agents
+  can route skip links and anchors without parsing aggregate in-page link rows.
+- Added individual text rows for semantic top links, so text-only agents can
+  route first-link navigation state without parsing aggregate link rows.
+- Added individual text rows for semantic top buttons, so text-only agents can
+  route button state, controlled panels, and form overrides without parsing
+  aggregate button rows.
+- Added individual text rows for semantic top images, so text-only agents can
+  route visual evidence, alt text, and responsive image metadata without parsing
+  aggregate image rows.
+- Added individual text rows for semantic top fields, so text-only agents can
+  route input labels, descriptions, errors, state, and selectors without parsing
+  aggregate field rows.
+- Added individual text rows for semantic top choices and selected choices, so
+  text-only agents can route active options, tabs, current state, and controlled
+  panels without parsing aggregate choice rows.
+- Added individual text rows for semantic top state, modal state, and live
+  state, so text-only agents can route control state and announcements without
+  parsing aggregate state rows.
+- Added individual text rows for semantic top descriptions, values, and
+  relations, so text-only agents can route descriptions, valuetext, and target
+  references without parsing aggregate relation rows.
+- Added individual text rows for semantic unavailable targets, so text-only
+  agents can identify iframe/static-HTML gaps without parsing aggregate
+  unavailable rows.
+- Added individual text rows for semantic top heading, landmark, named-role,
+  interactive, and focusable targets, so text-only agents can route primary
+  accessibility targets without parsing aggregate target rows.
+- Added individual text rows for semantic keyboard shortcuts and ARIA key
+  shortcuts, so text-only agents can target keyboard commands without parsing
+  aggregate keyboard rows.
+- Added second semantic list-item shortcuts in full, brief, text, public type,
+  readiness, and static scoring paths, so agents can compare representative
+  list/menu/tree entries without parsing `semanticTopListItemRefs[1]`.
+- Added individual text rows plus static/readiness scoring for first semantic
+  list-item shortcuts, closing the remaining text-only gap for
+  `semanticTopListItemRefs[0]`.
+- Added individual text rows for second semantic table headers, so text-only
+  agents can read the second representative header without parsing the
+  aggregate `semanticTopTableSecondHeader` line.
+- Added individual text rows for first semantic table owned targets, so
+  text-only agents can target `aria-owns`/virtualized rowgroups without parsing
+  the aggregate `semanticTopTableFirstOwned` line.
+- Added individual text rows for first semantic table sample cells, so
+  text-only agents can read representative cell path/text/span/header/selector
+  fields without parsing the aggregate sample-cell row.
+- Added individual text rows for second semantic table sample cells, so
+  text-only agents can inspect the second representative cell without parsing
+  aggregate sample-cell rows.
+- Added individual text rows for first owned semantic table sample cells, so
+  text-only agents can inspect virtualized/`aria-owns` cells without parsing
+  aggregate owned sample-cell rows.
+- Added individual text rows for semantic top field metadata and state, so
+  text-only agents can inspect form names, placeholders, autocomplete,
+  constraints, control targets, and value state without parsing aggregate field
+  rows.
+- Added individual text rows for semantic top table summary metadata, so
+  text-only agents can choose tables by role, name, dimensions, headers,
+  samples, and selector without parsing aggregate table rows.
+- Added individual text rows for semantic top list summary metadata, so
+  text-only agents can choose list containers by role, name, item count,
+  samples, and selector without parsing aggregate list rows.
+- Added individual text rows for semantic top outline metadata, so text-only
+  agents can jump to the first landmark/heading by path, kind, role, depth,
+  parent, and selector without parsing aggregate outline rows.
+- Added direct text rows for remaining semantic top role/ref fields, so
+  text-only agents can read top role counts and shallow table/list ref arrays
+  without parsing aggregate table/list/role rows.
+- Added a static CLI regression test that fails when a `semanticTop*` shortcut
+  exists in the agent contract but lacks a labelled text-output row.
+- Added `secondConfig*` agent shortcuts for the second detected app config,
+  including path, kind, name, key count, key names, and selector. This keeps
+  common app-shell config handoff shallow while the full `pageCheck.config`
+  array remains the detailed source of truth.
+- Added `secondAppHint*` agent shortcuts for the second detected app hint,
+  including URL path/query and fetch command args when the hint is fetchable,
+  so agents can inspect secondary icons/install metadata without parsing the
+  full app-hint array first.
+- Added `secondMobileHint*` agent shortcuts for the second detected mobile
+  hint, including value, platform, URL path/query, and selector, so agents can
+  inspect viewport-adjacent mobile rendering and app-link signals without
+  parsing the full mobile-hint array first.
+- Added `secondTopic*` agent shortcuts for the second detected topic, including
+  kind, label, value, source, and selector, so agents can compare alternate
+  relevance terms for search refinement without parsing the full topic array.
+- Added `secondKeyValue*` agent shortcuts for the second detected key-value
+  fact, including label, value, datetime, source, and selector, so agents can
+  compare adjacent definition-list/time facts without parsing the full
+  key-value array.
+- Added `secondMetaFact*` agent shortcuts for the second detected metadata
+  fact, including label, value, URL, source, and selector, so agents can inspect
+  robots/Open Graph/canonical-adjacent facts without parsing the full metadata
+  array.
+- Added `secondHttpPolicy*` agent shortcuts for the second detected HTTP policy,
+  including name, value, source, and selector, so agents can inspect CSP,
+  X-Robots-Tag, and referrer policy combinations without parsing the full
+  policy array.
+- Added `secondSchemaFact*` agent shortcuts for the second detected JSON-LD
+  schema fact group, including types, first fact, fact count, and selector, so
+  agents can compare alternate structured entities without parsing the full
+  schema-fact array.
+- Added `secondHiddenSignal*` agent shortcuts for the second static-only hidden
+  signal, so agents can compare alternate hydration/API/runtime/config hints
+  before escalating to a browser.
+
+## In Progress
+
+- Compare static semantic output against browser accessibility output for a
+  small, sequential fixture set and record any newly discovered signal gaps.
+- Keep browser-backed validation resource-safe by auditing comparison-script
+  locks, finally-based session close paths, and clean `pnpm check:processes`
+  results before and after risky runs.
+
+## Next Candidates
+
+- Add missing top-level shortcuts only when they improve agent routing or
+  reduce the need for browser handoff.
+- Continue checking whether action shortcuts need any remaining evidence,
+  citation, or source quality fields before declaring the handoff surface done.
+- Improve table/grid summaries where browser trees expose useful ownership or
+  navigation context not already covered by row/column/header refs.
+- Track cases where static HTML should stop and recommend browser capture
+  instead of guessing.
+
+## Safety Rule
+
+Use one validation command at a time. Before and after risky work, run:
+
+```sh
+pnpm check:processes
+```