yunti-browser-runtime 0.1.2 → 0.2.0

package/docs/PROJECT_STATUS.md

@@ -2,10 +2,331 @@
 
 ## Current Phase
 
-Patch release `yunti-browser-runtime@0.1.2` is complete. The local loopback
-bridge no longer requires a token by default, the extension popup no longer
-shows token as a first-run field, and the published package is verified with
-`npm run release:verify-published`.
+Patch release `yunti-browser-runtime@0.1.3` is complete. The project is now
+closing the `0.2.0 Best Browser Automation Runtime` npm/unpacked release
+candidate, documented in `docs/NEXT_MAJOR_PLAN.md`.
+
+Current implementation focus: P7.1 `0.2.0` npm/unpacked release closure.
+P6.6.1 browser extension distribution readiness, P6.6.2 store-facing
+permission/privacy copy, and P6.6.3 pre-store permission strategy decision are
+complete; P6.5 optional local runtime console is complete through real-browser
+validation. P6.6.4 store-candidate permission UX design is deferred to the
+post-0.2 store-candidate track.
+
+Latest visible 0.2.0 phase split:
+
+- P6.1.1 completed: `yunti_observe_page` schema, tool hints, and bridge
+  routing.
+- P6.1.2 completed: content-script DOM observer, compact text tree, field
+  state hints, scroll metadata, and minimum redaction.
+- P6.1.3 completed: fresh observe uids feed existing click/hover/fill uid
+  actions while preserving the snapshot compatibility path.
+- P6.1.4 completed: real-browser `observe -> click uid -> observe/verify`
+  closure validation passed with the local `pytest-playwright` environment
+  plus temporary Node Playwright 1.58.0.
+- P6.2.1 completed: additive structured action result contract and main action
+  path coverage.
+- P6.2.2 completed: `yunti_select` uid/selector/value/text semantics and
+  structured failure diagnostics.
+- P6.2.3 completed: `yunti_fill` / `yunti_fill_form` field-state,
+  contenteditable, failure aggregation, and post-fill value retention
+  diagnostics.
+- P6.2.4 completed: `yunti_scroll` uid, coordinate, no-movement, partial
+  movement, missing-target, and document-fallback diagnostics.
+- P6.2.5 completed: async UI recovery guidance for
+  `yunti_wait_for -> yunti_observe_page -> fresh uid`.
+- P6.2.6 completed: `yunti_wait_for` now preserves compatibility fields such
+  as `found`, `text`, `selector`, `condition`, `value`, and `waitedMs` while
+  adding `action: "wait_for"`, `target`, `ok`, `recoverable`, `nextStepHint`,
+  and timeout diagnostics with `code: "WAIT_TIMEOUT"` plus `recoveryHint`.
+- P6.2.7 completed: structured action result coverage is now captured in
+  `docs/ACTION_RESULT_COVERAGE.md` and checked by
+  `npm run check:action-results`, which is part of `npm run release:check`.
+- P6.3.1 completed: `docs/AGENT_WORKFLOW_CONTRACT.md`,
+  `yunti_get_tool_usage_hints` workflow hints, Tool Guide, and the packaged
+  skill now share the same default agent workflow contract and copyable prompt.
+- P6.3.2 completed: Tool Guide, packaged skill, and `yunti_get_tool_usage_hints`
+  now expose minimal workflows for click, form fill, scroll-to-find, tab
+  switching, and async wait.
+- P6.4.1 completed: strict DOM redaction now covers page title, labels, names,
+  visible text, placeholders, value previews, select selected value/text,
+  option value/text, scrollable container names, and likely email, phone,
+  Luhn-valid payment-card-like, and address-like content while keeping balanced
+  credential-like protection as the default.
+- P6.4.2 completed: shared MCP redaction now sanitizes likely Bearer tokens,
+  JWTs, private keys, long token-like values, emails, phones, Luhn-valid
+  payment-card-like values, and address-like text before learning memory writes
+  and console diagnostic caching.
+- P6.4.3 completed: raw CDP and screenshot tools now expose explicit guidance
+  for non-default-redaction diagnostics, sanitized-tool preference, scope
+  minimization, safe summarization, and CDP cleanup.
+- P6.5.1 completed: optional local runtime console minimum loop. The bridge now
+  serves `/console`, `/console/state`, and `/console/cancel-pending`; the CLI
+  exposes `yunti-browser-runtime console`; state uses sanitized summaries by
+  default and remains protected when bridge auth is enabled.
+- P6.5.2 completed: local console diagnostic polish now includes `HEAD
+  /console`, runtime/expected-extension/session-extension version summaries,
+  extension version mismatch warnings, visible warning cards, doctor console URL
+  output, and stronger no-page recovery guidance.
+- P6.5.3 completed: real-browser console validation now checks `HEAD
+  /console`, `/console`, `/console/state`, runtime/extension version alignment,
+  connected page state, and absence of no-page or version-mismatch warnings in
+  the opt-in Playwright extension smoke.
+- P6.6.1 completed: browser extension distribution readiness audit and public
+  doc links.
+- P6.6.2 completed: store-facing permission and privacy copy.
+- P6.6.3 completed: pre-store permission strategy decision.
+- P6.6.4 deferred: store-candidate permission UX design.
+- P7.1 in progress: `0.2.0` npm/unpacked release closure.
+
+Latest detailed P6.2 status:
+
+- Completed: action result main-path coverage, `yunti_select`
+  selector/value + uid/value + uid/text, select uid/selector failure
+  diagnostics, contenteditable fill, selector/uid fill failure diagnostics,
+  `yunti_fill_form` aggregation diagnostics, uid scroll, and scroll
+  no-movement recovery diagnostics.
+- Completed: uid/selector fill diagnostics for non-editable, hidden, disabled,
+  or readonly targets while preserving existing successful fill/select/scroll
+  behavior and compatibility fields.
+- Completed: uid fill post-value verification for keyboard/contenteditable
+  paths. When the value does not remain, `yunti_fill` returns
+  `VALUE_NOT_APPLIED` with length-only diagnostics and recovery guidance instead
+  of echoing the raw field value.
+- Completed: scroll partial-movement diagnostics. When a scroll moves less than
+  requested, results keep `ok: true` and include `partialMovement` with
+  requested/actual deltas, affected axes, `edgeHint`, and
+  `decision: "observe-before-continuing-scroll"`.
+- Completed: uid scroll missing/stale target diagnostics. When a scroll uid
+  cannot be resolved, `yunti_scroll` returns `scrolled: false`,
+  `UID_NOT_FOUND` or `UID_COORDINATES_UNAVAILABLE`, and recovery guidance to
+  refresh observation before retrying.
+- Completed: coordinate scroll hit/fallback diagnostics. When `x` / `y` is
+  provided, `yunti_scroll` can report `coordinateTarget`,
+  `scrollContainerFound`, and `coordinateScrollFallback: "document"` so agents
+  can detect document fallback instead of assuming a nested panel moved.
+- Completed: coordinate scroll document fallback recovery hints. Successful
+  document-fallback scrolls keep `ok: true` but can include
+  `coordinateFallbackHint` and a clearer `nextStepHint` pointing agents back to
+  `yunti_observe_page` and fresh `scrollableContainers[]` uids.
+- Completed: wait-observe recovery guidance for async UI transitions. Tool
+  usage hints, Tool Guide, and the packaged skill now tell agents to use
+  `yunti_wait_for`, then `yunti_observe_page`, then fresh uids for async
+  fill/select/scroll recovery instead of reusing old targets.
+- Completed: structured `yunti_wait_for` results. Successful waits now include
+  `action: "wait_for"`, `target`, `ok: true`, `recoverable: false`, and
+  `nextStepHint`; timeout waits return `found: false`, `ok: false`,
+  `recoverable: true`, `code: "WAIT_TIMEOUT"`, `recoveryHint`, and
+  `nextStepHint`.
+- Latest P6.2.6 validation: `git diff --check` passed; token residue grep
+  returned no matches; `node --test tests/tool-handlers.test.js` passed 42
+  tests; `node --test tests/bridge.test.js` passed 66 tests; `YUNTI_E2E=1 npm
+  run test:e2e` skipped because Playwright/Chromium is not installed in the
+  current environment; `npm run release:check` passed with 116 node:test cases
+  total, 115 passing and 1 default real-browser smoke skipped, plus npm package
+  and extension zip content checks.
+- Completed: `yunti_observe_page` field-state hints, select selected-option
+  hints, select `options[]` summaries, and structured disabled-option
+  diagnostics for `yunti_select`.
+- Completed: P6.2.7 coverage audit gate. `docs/ACTION_RESULT_COVERAGE.md`
+  now records the 11 core action surfaces and their success paths, failure
+  diagnostics, preserved compatibility fields, and remaining real-browser
+  validation notes. `scripts/check-action-result-coverage.js` validates the
+  matrix and is exposed as `npm run check:action-results`.
+- Latest P6.2.7 validation: `npm run check:action-results` passed with 11
+  coverage rows; `npm run check` passed; `git diff --check` passed; token
+  residue grep returned no matches; `YUNTI_E2E=1 npm run test:e2e` skipped
+  because Playwright/Chromium is not installed in the current environment;
+  `npm run release:check` passed with 116 node:test cases total, 115 passing
+  and 1 default real-browser smoke skipped, plus npm package contents
+  validation with 44 files and extension zip contents validation with 13 files.
+- Completed: P6.3.1 agent workflow contract. `docs/AGENT_WORKFLOW_CONTRACT.md`
+  records the default browser operation loop, recovery rules, confirmation
+  boundary, copyable prompt, and non-goals. `yunti_get_tool_usage_hints` now
+  accepts `topic: "workflow"` and returns `defaultPageOperation`,
+  `confirmationBoundary`, and `copyableAgentPrompt`. Tool Guide and packaged
+  skill now include the same default page operation contract.
+- Latest P6.3.1 validation: `node --test tests/bridge.test.js` passed 67
+  tests; `git diff --check` passed; token residue grep returned no matches;
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment; `npm run release:check` passed with
+  117 node:test cases total, 116 passing and 1 default real-browser smoke
+  skipped, plus npm package contents validation with 45 files and extension zip
+  contents validation with 13 files.
+- Next: implement P6.3.2 minimal use cases in Tool Guide and skill, or close
+  P6.1.4 first if Playwright/Chromium is available for real-browser closure.
+- Completed: P6.3.2 minimal use cases. Tool Guide and packaged skill now cover
+  Click, Fill Form, Scroll To Find, Switch Tab, and Wait For Async Result.
+  `yunti_get_tool_usage_hints` now returns matching `minimalUseCases` under
+  the `workflow` topic, and bridge tests lock the five keys plus critical
+  guidance such as fresh observation, field state, `scrollableContainers[]`,
+  `Target.activateTarget`, and `WAIT_TIMEOUT`.
+- Latest P6.3.2 validation: `node --test tests/bridge.test.js` passed 68
+  tests; `git diff --check` passed; token residue grep returned no matches;
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment; `npm run release:check` passed with
+  118 node:test cases total, 117 passing and 1 default real-browser smoke
+  skipped, plus npm package contents validation with 45 files and extension zip
+  contents validation with 13 files.
+- Completed: P6.4.1 DOM redaction policy deepening. `strict` mode now redacts
+  PII-like text surfaces in `yunti_observe_page`, including page title, labels,
+  names, visible text, placeholders, value previews, select selected value/text,
+  option value/text, scrollable container names, email, phone, Luhn-valid
+  payment-card-like values, and address-like text. `balanced` remains the
+  default credential-like protection mode; `off` remains explicit local
+  debugging only.
+- Latest P6.4.1 validation: `node --check extension/dom-observer.js` passed;
+  `node --test tests/dom-observer.test.js` passed 6 tests covering strict PII
+  categories, select option text redaction, page title redaction, and balanced
+  ordinary business text behavior; `node --test tests/bridge.test.js` passed 68
+  tests; `git diff --check` passed; token residue grep returned no matches;
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment; `npm run release:check` passed with
+  119 node:test cases total, 118 passing and 1 default real-browser smoke
+  skipped, plus npm package contents validation with 45 files and extension zip
+  contents validation with 13 files.
+- Completed: P6.4.2 learning memory / diagnostic artifacts secret-boundary
+  tightening. `mcp/redaction.js` now provides shared likely-sensitive text
+  sanitization for Bearer tokens, JWTs, private keys, long token-like values,
+  emails, phones, Luhn-valid payment-card-like values, and address-like text.
+  `yunti_remember_learning` sanitizes title, detail, tags, and source before
+  disk writes. Console diagnostics sanitize text, stackTrace, and args before
+  caching. Raw CDP events intentionally remain raw low-level diagnostics and
+  are documented as an explicit exception to clear after debugging.
+- Latest P6.4.2 validation: `node --check mcp/redaction.js`,
+  `node --check mcp/memory.js`, and `node --check mcp/bridge-hub.js` passed;
+  `node --test tests/bridge.test.js` passed 70 tests, including new memory and
+  console redaction coverage; `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 121
+  node:test cases total, 120 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment; `npm run release:check` passed, plus npm
+  package contents validation with 45 files and extension zip contents validation
+  with 13 files.
+- Completed: P6.4.3 raw CDP / screenshot non-redaction diagnostic guidance and
+  cleanup boundaries. Tool descriptions, `yunti_get_tool_usage_hints`, Tool
+  Guide, packaged skill, and Security docs now tell agents to prefer sanitized
+  diagnostics, use raw CDP only for low-level debugging, filter by `method` and
+  `limit`, clear with `yunti_clear_cdp_events`, treat screenshots as real
+  visible pixels outside DOM redaction, prefer viewport screenshots when enough,
+  and summarize safe findings instead of storing raw payloads or images.
+- Latest P6.4.3 validation: `node --check mcp/tools.js` passed;
+  `node --test tests/bridge.test.js` passed 71 tests, including new raw CDP and
+  screenshot safety-boundary usage hints coverage; `git diff --check` passed;
+  token residue grep returned no matches; `npm run check` passed; `npm test`
+  passed with 122 node:test cases total, 121 passing and 1 default real-browser
+  smoke skipped; `YUNTI_E2E=1 npm run test:e2e` skipped because
+  Playwright/Chromium is not installed in the current environment;
+  `npm run release:check` passed, plus npm package contents validation with 45
+  files and extension zip contents validation with 13 files.
+- Completed: P6.1.4 real-browser closure validation. The local
+  `/opt/miniconda3/envs/pytest-playwright` environment provides Python
+  Playwright 1.58.0 and existing browser caches; the Node E2E still needs the
+  Node `playwright` package, so validation used temporary `npm install --no-save
+  --package-lock=false playwright@1.58.0` with
+  `PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD=1` and without changing package metadata.
+  `YUNTI_E2E=1 npm run test:e2e` then passed the real browser extension bridge
+  smoke with `observe -> click uid -> verify`. Final validation for this status
+  update: `git diff --check` passed; token residue grep returned no matches;
+  `npm run check` passed; `npm test` passed with 122 node:test cases total,
+  121 passing and 1 default real-browser smoke skipped; `YUNTI_E2E=1 npm run
+  test:e2e` passed with 1 real-browser smoke test; `npm run release:check`
+  passed, plus npm package contents validation with 45 files and extension zip
+  contents validation with 13 files.
+- Completed: P6.5.1 optional local runtime console minimum loop. `mcp/http-server.js`
+  now serves an optional `/console` page, protected `/console/state` JSON, and
+  protected `/console/cancel-pending` action. `BridgeHub` tracks sanitized
+  recent activity, summarizes connected sessions without raw secrets, exposes
+  pending/queued request counts, and can cancel runtime requests that are still
+  queued or waiting for browser results. `bin/yunti-browser-runtime.js` and
+  `package.json` now expose a `console` command that starts the bridge and
+  prints the local console URL.
+- Latest P6.5.1 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 125
+  node:test cases total, 124 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke test;
+  `npm run release:check` passed, including 11 action-result coverage rows, npm
+  package contents validation with 45 files, and extension zip contents
+  validation with 13 files.
+- Completed: P6.5.2 local console diagnostic polish. Extension sessions now
+  report `client.extensionVersion`; console state reports runtime and expected
+  extension versions, session extension versions, and warning entries for no
+  connected pages, unknown extension version, or mismatched extension/runtime
+  versions. The `/console` route now supports `HEAD` for simple probes, the
+  console UI renders warning cards and version metadata, and `doctor` JSON plus
+  human summary now include the optional console URL.
+- Latest P6.5.2 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 126
+  node:test cases total, 125 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke test;
+  `npm run release:check` passed, including 11 action-result coverage rows, npm
+  package contents validation with 45 files, and extension zip contents
+  validation with 13 files.
+- Completed: P6.5.3 real-browser console validation. `tests/e2e.test.js` now
+  verifies the optional local console against a real Playwright-loaded
+  extension session: `HEAD /console` returns HTML, `/console` renders the page,
+  `/console/state` includes runtime and expected extension versions, the
+  connected session reports the current extension version, and no
+  `NO_CONNECTED_PAGES` or `EXTENSION_VERSION_MISMATCH` warning appears after a
+  real page registers.
+- Latest P6.5.3 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 126
+  node:test cases total, 125 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke test;
+  `npm run release:check` passed, including 11 action-result coverage rows, npm
+  package contents validation with 45 files, and extension zip contents
+  validation with 13 files.
+- Completed: P6.6.1 browser extension distribution readiness. Added
+  `docs/EXTENSION_DISTRIBUTION.md` with the current manifest permission audit,
+  Chrome Web Store / Edge Add-ons policy references, submission checklist,
+  privacy disclosure topics, and permission narrowing options. Linked it from
+  README, Security notes, and Roadmap.
+- Latest P6.6.1 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 126
+  node:test cases total, 125 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke in the local
+  `pytest-playwright` environment; `npm run release:check` passed, including
+  public documentation residue checks, public Markdown link checks for 15 files,
+  11 action-result coverage rows, npm package contents validation with 46
+  files, and extension zip contents validation with 13 files.
+- Completed: P6.6.2 store-facing permission and privacy copy. Added
+  `docs/EXTENSION_STORE_COPY.md` with draft short and long descriptions,
+  permission rationale, privacy-policy language, store data disclosure guidance,
+  reviewer notes, local bridge dependency notes, and final review checklist.
+- Latest P6.6.2 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 126
+  node:test cases total, 125 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke in the local
+  `pytest-playwright` environment; `npm run release:check` passed, including
+  public documentation residue checks, public Markdown link checks for 16 files,
+  11 action-result coverage rows, npm package contents validation with 47
+  files, and extension zip contents validation with 13 files.
+- Completed: P6.6.3 pre-store permission strategy decision. Added
+  `docs/EXTENSION_PERMISSION_STRATEGY.md`, keeping the current npm/unpacked
+  developer path unchanged, deciding that the current broad-permission manifest
+  should not be submitted to browser stores unchanged by default, and defining a
+  future store-candidate track for optional host access and optional network
+  diagnostics.
+- Latest P6.6.3 validation: `git diff --check` passed; token residue grep
+  returned no matches; `npm run check` passed; `npm test` passed with 126
+  node:test cases total, 125 passing and 1 default real-browser smoke skipped;
+  `YUNTI_E2E=1 npm run test:e2e` passed with 1 real-browser smoke in the local
+  `pytest-playwright` environment; `npm run release:check` passed, including
+  public documentation residue checks, public Markdown link checks for 17 files,
+  11 action-result coverage rows, npm package contents validation with 48
+  files, and extension zip contents validation with 13 files.
+- Next: complete P7.1 `0.2.0` npm/unpacked release closure. Version has moved
+  to `0.2.0`; run metadata, release, real-browser E2E, and dry-run gates before
+  asking for confirmation to publish. P6.6.4 store-candidate permission UX is
+  intentionally deferred and should not block this npm/unpacked release.
+- P7.1 current validation: package and extension versions are both `0.2.0`;
+  `git diff --check`, token residue grep, `npm run release:check`,
+  `YUNTI_E2E=1 npm run test:e2e`, and direct
+  `npm publish --dry-run --registry=https://registry.npmjs.org/` passed.
+  `npm run check:metadata` passed once for the public GitHub/npm metadata, then
+  later failed because this environment timed out connecting to GitHub; scripted
+  `release:prepublish` and `release:dry-run` need to be rerun when GitHub is
+  reachable again.
 
 ## Current State
 
@@ -43,9 +364,11 @@ shows token as a first-run field, and the published package is verified with
 - `npm run release:dry-run` and `npm run release:publish` pin the official npm
   registry at `https://registry.npmjs.org/`, avoiding accidental publication to
   a locally configured mirror registry.
-- `yunti-browser-runtime@0.1.2` is published on the official npm registry.
-- Extension popup keeps Bridge Token under optional advanced settings; default
-  local installs do not require saving popup settings.
+- `yunti-browser-runtime@0.1.3` is published on the official npm registry.
+- Current branch package and extension versions are being prepared as
+  `0.2.0` release candidate.
+- Extension popup is a zero-config status panel by default; Bridge URL, page
+  matching, and optional Bridge Token stay under advanced settings.
 - `npm run test:e2e` provides an opt-in real-browser Playwright smoke test for
   extension loading, page registration, MCP target listing, snapshot, click/fill,
   and CDP `Runtime.evaluate`.
@@ -89,6 +412,635 @@ shows token as a first-run field, and the published package is verified with
 - Extension registers all `http` and `https` pages by default.
 - Product-specific fixed conversation, workspace, and side-panel entry points
   have been removed from the standalone extension.
+- The next major execution plan is captured in `docs/NEXT_MAJOR_PLAN.md`.
+- P6.1 observe/action compatibility has started: `yunti_observe_page` now has an
+  MCP tool schema, observe-first usage hints, bridge routing contract tests, a
+  content-script DOM observer, focused DOM observer tests, and an opt-in
+  real-browser smoke assertion. Fresh observe uids now feed the existing
+  click/hover/fill uid resolver while preserving the `yunti_take_snapshot`
+  compatibility path.
+- Latest P6.1 observe/action validation: `git diff --check` passed,
+  `npm run release:check` passed with 69 passing node:test cases and 1 skipped
+  real-browser smoke by default, and `YUNTI_E2E=1 npm run test:e2e` skipped
+  because Playwright/Chromium is not installed in the current environment.
+- Deterministic observe fixtures now cover hidden/offscreen target handling and
+  redaction edge modes in addition to fresh uid, text tree, scroll metadata,
+  balanced redaction, and stale uid replacement.
+- `docs/EXECUTION_PLAN.md` now includes a P6.1 real-browser closure runbook for
+  the remaining `observe -> click uid -> observe/verify` validation gap.
+- P6.2 preparation has started with tool usage action recovery guidance for
+  click/hover/fill, without changing the runtime action implementation.
+- Latest P6.2 preparation validation: `git diff --check` passed,
+  `node --test tests/bridge.test.js` passed 63 tests, `npm run release:check`
+  passed with 70 passing node:test cases and 1 skipped real-browser smoke by
+  default, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is
+  not installed in the current environment, and the token residue grep returned
+  no matches.
+- P6.2 action result contract guidance now distinguishes current
+  compatibility-shaped action returns from the additive target contract for
+  `action`, `target`, `ok`, `code`, `recoverable`, `nextStepHint`, and
+  before/after summaries.
+- Latest P6.2 action result contract validation: `git diff --check` passed,
+  `node --test tests/bridge.test.js` passed 64 tests, `npm run release:check`
+  passed with 71 passing node:test cases and 1 skipped real-browser smoke by
+  default, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is
+  not installed in the current environment, and the token residue grep returned
+  no matches.
+- Dispatcher-level action result shape tests now lock the current compatible
+  uid click, uid hover, uid fill, and content-script scroll passthrough result
+  fields before P6.2 adds structured action result fields.
+- Latest P6.2 action result shape validation: `git diff --check` passed,
+  `node --test tests/tool-handlers.test.js` passed 4 tests,
+  `npm run release:check` passed with 72 passing node:test cases and 1 skipped
+  real-browser smoke by default, `YUNTI_E2E=1 npm run test:e2e` skipped because
+  Playwright/Chromium is not installed in the current environment, and the token
+  residue grep returned no matches.
+- uid-based `yunti_click` now returns additive structured action result fields
+  (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`) while preserving
+  the existing `clicked`, `uid`, `x/y`, and `browserSessionId` compatibility
+  fields.
+- Latest P6.2 uid-click structured result validation: `git diff --check`
+  passed, `node --test tests/tool-handlers.test.js` passed 4 tests,
+  `npm run release:check` passed with 72 passing node:test cases and 1 skipped
+  real-browser smoke by default, `YUNTI_E2E=1 npm run test:e2e` skipped because
+  Playwright/Chromium is not installed in the current environment, and the token
+  residue grep returned no matches.
+- uid-based `yunti_hover` now returns additive structured action result fields
+  (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`) while preserving
+  the existing `hovered`, `uid`, `x/y`, and `browserSessionId` compatibility
+  fields.
+- Latest P6.2 uid-hover structured result validation: `git diff --check`
+  passed, `node --test tests/tool-handlers.test.js` passed 4 tests,
+  `npm run release:check` passed with 72 passing node:test cases and 1 skipped
+  real-browser smoke by default, `YUNTI_E2E=1 npm run test:e2e` skipped because
+  Playwright/Chromium is not installed in the current environment, and the token
+  residue grep returned no matches.
+- uid-based `yunti_fill` keyboard path now returns additive structured action
+  result fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`)
+  while preserving the existing `filled`, `uid`, `method`, `value`, and
+  `browserSessionId` compatibility fields. Select/contenteditable/scroll
+  semantics are intentionally unchanged in this slice.
+- Latest P6.2 uid-fill keyboard structured result validation: `git diff
+  --check` passed, `node --test tests/tool-handlers.test.js` passed 4 tests,
+  `npm run release:check` passed with 72 passing node:test cases and 1 skipped
+  real-browser smoke by default, npm package contents validation passed with 42
+  files, extension zip contents validation passed with 13 files,
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment, and the token residue grep returned no
+  matches.
+- uid-based `yunti_fill` select path now returns additive structured action
+  result fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`)
+  while preserving the existing `filled`, `uid`, `method`, `value`, and
+  `browserSessionId` compatibility fields. Select value dispatch semantics are
+  intentionally unchanged in this slice.
+- Latest P6.2 uid-fill select structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 5 tests.
+- Latest P6.2 uid-fill select structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 73 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- Coordinate fallback `yunti_click` now returns additive structured action
+  result fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`)
+  while preserving the existing `clicked`, `x/y`, `method`, and
+  `browserSessionId` compatibility fields. Coordinate click dispatch semantics
+  are intentionally unchanged in this slice.
+- Latest P6.2 coordinate-click structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 6 tests.
+- Latest P6.2 coordinate-click structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 74 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- Coordinate fallback `yunti_hover` now returns additive structured action
+  result fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`)
+  while preserving the existing `hovered`, `x/y`, `method`, and
+  `browserSessionId` compatibility fields. Coordinate hover dispatch semantics
+  are intentionally unchanged in this slice.
+- Latest P6.2 coordinate-hover structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 7 tests.
+- Latest P6.2 coordinate-hover structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 75 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- Selector fallback `yunti_hover` now returns additive structured action result
+  fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`) while
+  preserving the existing `hovered`, `selector`, `x/y`, `method`, and
+  `browserSessionId` compatibility fields. Selector hover dispatch semantics
+  are intentionally unchanged in this slice.
+- Latest P6.2 selector-hover structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 8 tests.
+- Latest P6.2 selector-hover structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 76 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- Selector fallback `yunti_click` now returns additive structured action result
+  fields (`action`, `target`, `ok`, `recoverable`, and `nextStepHint`) around
+  the existing content-script passthrough result while preserving `clicked`,
+  `element`, `selector`, `method`, and `browserSessionId` compatibility fields.
+  Selector click dispatch semantics are intentionally unchanged in this slice.
+- Latest P6.2 selector-click structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 9 tests.
+- Latest P6.2 selector-click structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 77 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- `yunti_scroll` now returns additive structured action result fields
+  (`action`, `ok`, `recoverable`, `nextStepHint`, and `browserSessionId`) around
+  the existing content-script passthrough result while preserving `scrolled`,
+  `deltaX`, `deltaY`, `target`, `before`, and `after` compatibility fields.
+  Document/container scroll selection and `scrollBy` dispatch semantics are
+  intentionally unchanged in this slice.
+- Latest P6.2 scroll structured result targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js` passed 9 tests.
+- Latest P6.2 scroll structured result full validation: `git diff --check`
+  passed, `npm run release:check` passed with 77 passing node:test cases and 1
+  skipped real-browser smoke by default, npm package contents validation passed
+  with 42 files, extension zip contents validation passed with 13 files,
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment, and the token residue grep returned no
+  matches.
+- Selector fallback `yunti_fill` now returns additive structured action result
+  fields (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, `selector`,
+  `method`, and `browserSessionId`) around the existing content-script
+  passthrough result while preserving `filled`, `element`, and `valueLength`
+  compatibility fields. Selector fill dispatch semantics are intentionally
+  unchanged in this slice.
+- Latest P6.2 selector-fill structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 10 tests.
+- Latest P6.2 selector-fill structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 78 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- `yunti_type_text` now returns additive structured action result fields
+  (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, and
+  `browserSessionId`) for both uid-based CDP keyboard typing and content-script
+  passthrough typing, while preserving uid path compatibility fields (`typed`,
+  `uid`, `text`, and `method`) and content-script compatibility fields
+  (`typed`, `element`, `textLength`, and `mode`). Type dispatch semantics are
+  intentionally unchanged in this slice.
+- Latest P6.2 type-text structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 12 tests.
+- Latest P6.2 type-text structured result full validation: `git diff --check`
+  passed, `npm run release:check` passed with 80 passing node:test cases and 1
+  skipped real-browser smoke by default, npm package contents validation passed
+  with 42 files, extension zip contents validation passed with 13 files,
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment, and the token residue grep returned no
+  matches.
+- `yunti_press_key` now returns additive structured action result fields
+  (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, and
+  `browserSessionId`) for both uid-based CDP keyboard dispatch and
+  content-script passthrough key press, while preserving uid path compatibility
+  fields (`pressed`, `uid`, and `key`) and content-script compatibility fields
+  (`pressed`, `key`, `element`, and `valueChanged`). Key dispatch semantics are
+  intentionally unchanged in this slice.
+- Latest P6.2 press-key structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 14 tests.
+- Latest P6.2 press-key structured result full validation: `git diff --check`
+  passed, `npm run release:check` passed with 82 passing node:test cases and 1
+  skipped real-browser smoke by default, npm package contents validation passed
+  with 42 files, extension zip contents validation passed with 13 files,
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment, and the token residue grep returned no
+  matches.
+- `yunti_upload_file` now returns additive structured action result fields
+  (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, and
+  `browserSessionId`) for both uid-based and selector-based file input
+  resolution, while preserving compatibility fields (`uploaded`, `fileCount`,
+  and `filePaths`). File input resolution and `DOM.setFileInputFiles` dispatch
+  semantics are intentionally unchanged in this slice.
+- Latest P6.2 upload-file structured result targeted validation: `git diff
+  --check` passed and `node --test tests/tool-handlers.test.js` passed 16 tests.
+- Latest P6.2 upload-file structured result full validation: `git diff
+  --check` passed, `npm run release:check` passed with 84 passing node:test
+  cases and 1 skipped real-browser smoke by default, npm package contents
+  validation passed with 42 files, extension zip contents validation passed with
+  13 files, `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium
+  is not installed in the current environment, and the token residue grep
+  returned no matches.
+- `yunti_drag` now returns additive structured action result fields (`action`,
+  `target`, `ok`, `recoverable`, `nextStepHint`, and `browserSessionId`) while
+  preserving compatibility fields (`dragged`, `from`, `to`, and `steps`). CDP
+  mouse event dispatch semantics are intentionally unchanged in this slice.
+- Latest P6.2 drag structured result targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js` passed 17 tests.
+- Latest P6.2 drag structured result full validation: `git diff --check`
+  passed, `npm run release:check` passed with 85 passing node:test cases and 1
+  skipped real-browser smoke by default, npm package contents validation passed
+  with 42 files, extension zip contents validation passed with 13 files,
+  `YUNTI_E2E=1 npm run test:e2e` skipped because Playwright/Chromium is not
+  installed in the current environment, and the token residue grep returned no
+  matches.
+- P6.2 action result coverage has been audited after the drag slice. Structured
+  action result fields now cover the main single-action paths for click, hover,
+  fill, scroll, type text, press key, upload, and drag while preserving existing
+  compatibility fields. Remaining action result gaps are standalone
+  `yunti_select` content-script passthrough and aggregate `yunti_fill_form`.
+  These must be completed before moving into deeper fill/select/scroll
+  semantics.
+- Latest P6.2 action result coverage checklist validation: `git diff --check`
+  passed, `npm run release:check` passed with 85 passing node:test cases and 1
+  skipped real-browser smoke by default, npm package contents validation passed
+  with 42 files, extension zip contents validation passed with 13 files. This
+  slice only updates docs and skill guidance, so `YUNTI_E2E=1 npm run test:e2e`
+  was not rerun; the token residue grep returned no matches.
+- Standalone `yunti_select` now returns additive structured action result
+  fields (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, and
+  `browserSessionId`) around the existing content-script passthrough result
+  while preserving compatibility fields (`selected`, `element`, and `value`).
+  Select value assignment and input/change event dispatch semantics are
+  intentionally unchanged in this slice.
+- Latest P6.2 select structured result targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js` passed 18 tests.
+- Latest P6.2 select structured result full validation: `git diff --check`
+  passed; `npm run release:check` passed with 87 node:test cases total, 86
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- Aggregate `yunti_fill_form` now returns additive structured action result
+  fields (`action`, `target`, `ok`, `recoverable`, `nextStepHint`, and
+  `browserSessionId`) around the existing aggregate summary while preserving
+  compatibility fields (`filled`, `failed`, and `results`). Per-field
+  `fillByUid` and content-script selector fill semantics are intentionally
+  unchanged in this slice.
+- Latest P6.2 fill_form structured result targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js` passed 19 tests.
+- Latest P6.2 fill_form structured result full validation: `git diff --check`
+  passed; `npm run release:check` passed with 88 node:test cases total, 87
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- `yunti_select` now has a P6.2 uid/visible text planning contract in schema
+  and usage hints. The current runtime remains selector/value compatible; uid
+  and visible text selection semantics are intentionally left for the next
+  runtime slice.
+- Latest P6.2 select uid/text contract targeted validation: `git diff --check`
+  passed and `node --test tests/bridge.test.js` passed.
+- Latest P6.2 select uid/text contract full validation: `git diff --check`
+  passed; `npm run release:check` passed with 89 node:test cases total, 88
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files. This slice only updates schema, usage hints, docs, and skill
+  guidance, so `YUNTI_E2E=1 npm run test:e2e` was not rerun; the token residue
+  grep returned no matches.
+- `yunti_select` now supports the `uid + value` runtime path while preserving
+  selector/value compatibility. The uid path reuses the latest observe/snapshot
+  uid map, matches select options by option value, dispatches change/input
+  events, and returns structured action result fields. Visible option text
+  selection remains intentionally deferred to the next runtime slice.
+- Latest P6.2 select uid/value runtime targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 85 tests.
+- Latest P6.2 select uid/value runtime full validation: `git diff --check`
+  passed; `npm run release:check` passed with 90 node:test cases total, 89
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- `yunti_select` now supports the `uid + visible text` runtime path while
+  preserving selector/value and uid/value compatibility. The text path reuses
+  the latest observe/snapshot uid map, matches select options by visible label,
+  dispatches change/input events, and returns structured action result fields.
+- Latest P6.2 select uid/text runtime targeted validation: `git diff --check`
+  passed and `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 86 tests.
+- Latest P6.2 select uid/text runtime full validation: `git diff --check`
+  passed; `npm run release:check` passed with 91 node:test cases total, 90
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- `yunti_select` uid failure paths now return structured recovery diagnostics.
+  Option misses return `selected: false`, `code: "OPTION_NOT_FOUND"`,
+  `matchMode`, `targetOption`, available option values/texts when exposed by
+  the page, and `recoveryHint`; non-select uid targets return `code:
+  "NOT_SELECT"` plus `recoveryHint.decision` pointing agents toward the actual
+  select element or selector/value fallback.
+- Latest P6.2 select uid failure diagnostic targeted validation:
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 93
+  tests.
+- Latest P6.2 select uid failure diagnostic full validation: `git diff
+  --check` passed; `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 93 tests; `npm run release:check` passed with 98 node:test cases
+  total, 97 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- `yunti_select` selector/value failure paths now return structured recovery
+  diagnostics. Value mismatches return `selected: false`, `code:
+  "OPTION_NOT_FOUND"`, `actualValue`, `matchMode`, `targetOption`, and
+  `recoveryHint`; thrown selector failures return `code:
+  "SELECTOR_SELECT_FAILED"` plus `recoveryHint.decision` pointing agents toward
+  the actual select element or a fresh uid/value fallback.
+- Latest P6.2 select selector failure diagnostic targeted validation:
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 95
+  tests.
+- Latest P6.2 select selector failure diagnostic full validation: `git diff
+  --check` passed; `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 95 tests; `npm run release:check` passed with 100 node:test cases
+  total, 99 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 action path coverage was re-audited after select uid/text support.
+  Structured action result fields now cover the main action surfaces, and
+  `yunti_select` has selector/value, uid/value, and uid/text operation paths.
+  The next work should move into deeper fill/select/scroll semantics while
+  preserving selector/value compatibility, CDP fallback, diagnostics, and
+  existing action result fields.
+- Latest P6.2 action path coverage audit validation: `git diff --check`
+  passed; `npm run release:check` passed with 91 node:test cases total, 90
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; the token residue grep returned no matches. Real-browser E2E
+  was not rerun for this docs-only audit patch.
+- P6.2 deeper fill semantics have started with uid-targeted contenteditable
+  result reporting. The runtime keeps the existing CDP keyboard input flow, but
+  contenteditable targets now return `method: "contenteditable"` plus
+  before/after text length summaries and a verification-oriented next step
+  hint. Selector fill and normal keyboard fill compatibility fields are
+  unchanged.
+- Latest P6.2 contenteditable fill targeted validation:
+  `node --test tests/tool-handlers.test.js` passed 22 tests.
+- Latest P6.2 contenteditable fill full validation: `git diff --check`
+  passed; `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 87 tests; `npm run release:check` passed with 92 node:test cases
+  total, 91 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but
+  skipped because Playwright/Chromium is not installed locally; the token
+  residue grep returned no matches.
+- P6.2 fill failure diagnostics have started. Selector fill thrown failures and
+  uid select option misses now return structured `filled: false` diagnostics
+  with `ok: false`, `recoverable: true`, `code`, `recoveryHint`, and
+  `nextStepHint`, while preserving successful fill compatibility fields and
+  dispatch semantics. Uid select option misses include `availableValues` /
+  `availableTexts` when the page exposes them.
+- Latest P6.2 fill failure diagnostic targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --test
+  tests/tool-handlers.test.js` passed 32 tests.
+- Latest P6.2 fill failure diagnostic full validation: `git diff --check`
+  passed; the token residue grep returned no matches; `YUNTI_E2E=1 npm run
+  test:e2e` was executed but skipped because Playwright/Chromium is not
+  installed locally; `npm run release:check` passed with 102 node:test cases
+  total, 101 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files.
+- P6.2 fill_form aggregation now preserves per-field structured fill failure
+  diagnostics. Aggregate results keep existing `filled`, `failed`, `results`,
+  `action: "fill_form"`, `target`, `ok`, `recoverable`, and `nextStepHint`
+  fields, while failed `results[]` items can carry `code`, `recoveryHint`,
+  `availableValues` / `availableTexts`, and field-level `nextStepHint` from
+  `yunti_fill`.
+- Latest P6.2 fill_form diagnostic aggregation targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --test
+  tests/tool-handlers.test.js` passed 32 tests.
+- Latest P6.2 fill_form diagnostic aggregation full validation: `git diff
+  --check` passed; the token residue grep returned no matches; `YUNTI_E2E=1
+  npm run test:e2e` was executed but skipped because Playwright/Chromium is
+  not installed locally; `npm run release:check` passed with 102 node:test
+  cases total, 101 passed, and 1 real-browser smoke skipped by default; npm
+  package contents validation passed with 42 files; extension zip contents
+  validation passed with 13 files.
+- P6.2 non-editable fill diagnostics are complete. Uid fill now inspects the
+  resolved target before dispatching keyboard events and returns structured
+  `TARGET_NOT_EDITABLE` diagnostics for hidden/no-size, disabled, readonly, or
+  non-editable targets. Selector fill now checks editability in the content
+  script before mutating values. Failed results keep `filled: false`, `ok:
+  false`, `recoverable: true`, `code`, `recoveryHint`, and `nextStepHint`; uid
+  failures can also include a lightweight `element` diagnostic summary.
+- Latest P6.2 non-editable fill diagnostic targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --check
+  extension/content.js` passed; `node --test tests/tool-handlers.test.js`
+  passed 34 tests.
+- Latest P6.2 non-editable fill diagnostic full validation: `git diff
+  --check` passed; the token residue grep returned no matches; `YUNTI_E2E=1
+  npm run test:e2e` was executed but skipped because Playwright/Chromium is
+  not installed locally; `npm run release:check` passed with 104 node:test
+  cases total, 103 passed, and 1 real-browser smoke skipped by default; npm
+  package contents validation passed with 42 files; extension zip contents
+  validation passed with 13 files.
+- P6.2 observe field-state hints are complete. `yunti_observe_page` element
+  entries can now expose `editable`, `fillable`, `readOnly`,
+  `fillBlockReason`, and select `options[]` summaries so agents can inspect
+  field state before calling `yunti_fill` or `yunti_select`. This is additive
+  observe metadata only; uid lifecycle, default balanced redaction, and existing
+  page action tools remain unchanged.
+- Latest P6.2 observe field-state targeted validation:
+  `node --check extension/dom-observer.js` passed; `node --test
+  tests/dom-observer.test.js` passed 5 tests; `node --test tests/bridge.test.js`
+  passed 65 tests.
+- Latest P6.2 observe field-state full validation: `git diff --check` passed;
+  the token residue grep returned no matches; `YUNTI_E2E=1 npm run test:e2e`
+  was executed but skipped because Playwright/Chromium is not installed
+  locally; `npm run release:check` passed with 105 node:test cases total, 104
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files.
+- P6.2 observe select-state hints are complete. Select elements can now expose
+  `selectedIndex`, `selectedValue`, `selectedValueRedacted`, and `selectedText`
+  alongside `options[]`, so agents can inspect the current selection before
+  deciding whether to call `yunti_select`. This is additive observe metadata
+  only; `yunti_select` execution, uid lifecycle, and default balanced redaction
+  remain unchanged.
+- Latest P6.2 observe select-state targeted validation:
+  `node --check extension/dom-observer.js` passed; `node --test
+  tests/dom-observer.test.js` passed 5 tests.
+- P6.2 select disabled-option diagnostics are complete. Uid/value, uid/text,
+  and selector/value select paths now return structured `OPTION_DISABLED`
+  failures with `disabledValue`, `disabledText`, and recovery guidance to
+  choose an enabled option, wait for the field to unlock, or ask the user
+  before retrying. Existing successful select paths and compatibility fields
+  remain unchanged.
+- Latest P6.2 select disabled-option targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --check
+  extension/content.js` passed; `node --test tests/tool-handlers.test.js`
+  passed 36 tests.
+- Latest P6.2 select disabled-option full validation: `git diff --check`
+  passed; the token residue grep returned no matches; `YUNTI_E2E=1 npm run
+  test:e2e` was executed but skipped because Playwright/Chromium is not
+  installed locally; `npm run release:check` passed with 107 node:test cases
+  total, 106 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files.
+- P6.2 deeper scroll semantics have started with fresh observed scrollable
+  container uids. `yunti_observe_page` scrollableContainers now feed the current
+  browserSessionId uid map, and `yunti_scroll` can resolve a container uid to
+  its center while reusing the existing coordinate/container scroll path. The
+  existing `target` compatibility field is preserved; uid scroll adds `uid`,
+  `method: "uid"`, and `scrollTarget`.
+- Latest P6.2 uid scroll targeted validation:
+  `node --test tests/tool-handlers.test.js` passed 23 tests.
+- Latest P6.2 uid scroll full validation: `git diff --check` passed;
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 88
+  tests; `npm run release:check` passed with 93 node:test cases total, 92
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 scroll boundary diagnostics now report movement when before/after scroll
+  positions are comparable. A scroll that dispatches but does not move returns
+  `moved: false`, `code: "NO_SCROLL_MOVEMENT"`, `ok: false`, and
+  `recoverable: true`, while preserving existing `scrolled`, `target`, and
+  before/after compatibility fields.
+- Latest P6.2 scroll no-movement targeted validation:
+  `node --test tests/tool-handlers.test.js` passed 24 tests.
+- Latest P6.2 scroll no-movement full validation: `git diff --check` passed;
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 89
+  tests; `npm run release:check` passed with 94 node:test cases total, 93
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 scroll no-movement diagnostics now include directional `edgeHint` values
+  when delta direction can be inferred, such as `possible-bottom-edge`,
+  `possible-top-edge`, `possible-right-edge`, and `possible-left-edge`. This
+  helps agents stop repeating the same scroll and choose a different direction,
+  nearby scrollable container uid, or recovery path.
+- Latest P6.2 scroll edge-hint targeted validation:
+  `node --test tests/tool-handlers.test.js` passed 25 tests.
+- Latest P6.2 scroll edge-hint full validation: `git diff --check` passed;
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 90
+  tests; `npm run release:check` passed with 95 node:test cases total, 94
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 scroll no-movement diagnostics now include structured `recoveryHint`.
+  For `NO_SCROLL_MOVEMENT`, the result includes reason, nextAction,
+  recommendedTools, currentTarget, edgeHint, and, for fresh scrollable container
+  uids, lastObservedContainer scrollability and remaining-pixel metadata.
+- Latest P6.2 scroll recovery-hint targeted validation:
+  `node --test tests/tool-handlers.test.js` passed 26 tests.
+- Latest P6.2 scroll recovery-hint full validation: `git diff --check` passed;
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 91
+  tests; `npm run release:check` passed with 96 node:test cases total, 95
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 scroll recovery guidance now includes `recoveryHint.suggestedRetry`.
+  When an opposite direction can be derived from `edgeHint`, the hint carries a
+  one-shot opposite `deltaX` / `deltaY`; uid-targeted scroll keeps the same
+  fresh container uid so agents can retry once, observe again, then switch
+  container or stop repeating if there is still no movement.
+- Latest P6.2 scroll suggested-retry targeted validation:
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 91
+  tests.
+- Latest P6.2 scroll suggested-retry full validation: `git diff --check`
+  passed; `node --test tests/tool-handlers.test.js tests/bridge.test.js`
+  passed 91 tests; `npm run release:check` passed with 96 node:test cases
+  total, 95 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but
+  skipped because Playwright/Chromium is not installed locally; the token
+  residue grep returned no matches.
+- P6.2 scroll recovery guidance now includes `recoveryHint.decision`. For
+  `NO_SCROLL_MOVEMENT`, the decision is derived from `edgeHint` and whether the
+  scroll used a fresh container uid, producing machine-readable guidance such
+  as observing for a container, retrying the opposite direction once on the
+  same container, or providing a nonzero delta.
+- Latest P6.2 scroll decision-hint targeted validation:
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 91
+  tests.
+- Latest P6.2 scroll decision-hint full validation: `git diff --check` passed;
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js` passed 91
+  tests; `npm run release:check` passed with 96 node:test cases total, 95
+  passed, and 1 real-browser smoke skipped by default; npm package contents
+  validation passed with 42 files; extension zip contents validation passed
+  with 13 files; `YUNTI_E2E=1 npm run test:e2e` was executed but skipped
+  because Playwright/Chromium is not installed locally; the token residue grep
+  returned no matches.
+- P6.2 uid fill post-value verification is complete. Keyboard/contenteditable
+  fill paths now perform a lightweight post-dispatch value check when possible.
+  If the target value does not remain, `yunti_fill` returns structured
+  `filled: false`, `ok: false`, `code: "VALUE_NOT_APPLIED"`,
+  `expectedValueLength`, `actualValueLength`, `recoveryHint`, and
+  `nextStepHint` diagnostics without echoing the raw field value. Existing
+  successful fill/select/scroll compatibility fields are preserved.
+- Latest P6.2 uid fill post-value verification targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --test
+  tests/tool-handlers.test.js` passed 37 tests.
+- P6.2 scroll partial-movement diagnostics are complete. When comparable
+  `before` / `after` positions show movement that is smaller than the requested
+  `deltaX` / `deltaY`, `yunti_scroll` keeps the action successful but adds a
+  structured `partialMovement` hint with requested/actual deltas, affected axes,
+  `edgeHint`, `nextAction: "observe-again"`, and
+  `decision: "observe-before-continuing-scroll"`.
+- Latest P6.2 scroll partial-movement targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --test
+  tests/tool-handlers.test.js` passed 38 tests.
+- P6.2 uid scroll missing/stale target diagnostics are complete. Missing,
+  stale, hidden, or coordinate-unresolvable scroll uids now return structured
+  `scrolled: false`, `ok: false`, `recoverable: true`, `code`, `recoveryHint`,
+  and `nextStepHint` instead of only bubbling a generic dispatcher error.
+  Existing successful scroll, no-movement, and partial-movement paths are
+  preserved.
+- Latest P6.2 uid scroll missing-target targeted validation:
+  `node --check extension/tool-handlers.js` passed; `node --test
+  tests/tool-handlers.test.js` passed 39 tests.
+- P6.2 coordinate scroll hit/fallback diagnostics are complete. Coordinate
+  scrolls can now report `coordinateTarget`, `scrollContainerFound`, and
+  `coordinateScrollFallback: "document"` from the content script, so agents can
+  distinguish nested container hits from document fallback without changing the
+  existing scroll dispatch behavior.
+- P6.2 coordinate scroll document fallback recovery hints are complete. When
+  handler results include `coordinateScrollFallback: "document"`, successful
+  scrolls keep `ok: true` and `recoverable: false` while adding
+  `coordinateFallbackHint` plus a clearer `nextStepHint` to observe again and
+  choose a fresh `scrollableContainers[]` uid when a panel/sidebar was intended.
+- Latest P6.2 coordinate scroll fallback targeted validation:
+  `node --test tests/tool-handlers.test.js tests/bridge.test.js
+  tests/content-scroll.test.js` passed 107 tests.
+- Latest P6.2 coordinate scroll fallback full validation: `git diff --check`
+  passed; the token residue grep returned no matches; `YUNTI_E2E=1 npm run
+  test:e2e` was executed but skipped because Playwright/Chromium is not
+  installed locally; `npm run release:check` passed with 113 node:test cases
+  total, 112 passed, and 1 real-browser smoke skipped by default; npm package
+  contents validation passed with 42 files; extension zip contents validation
+  passed with 13 files.
+- P6.2 wait-observe recovery guidance is complete for tool guidance surfaces.
+  `yunti_get_tool_usage_hints`, `docs/TOOL_GUIDE.md`, and the packaged skill now
+  tell agents to recover from async rendering, validation, option population, or
+  newly loaded content by calling `yunti_wait_for`, observing again, and using a
+  fresh uid for fill/select/scroll continuation.
+- Latest P6.2 wait-observe guidance targeted validation: `node --test
+  tests/bridge.test.js` passed 65 tests.
+- Latest P6.2 wait-observe guidance full validation: `git diff --check` passed;
+  the token residue grep returned no matches; `YUNTI_E2E=1 npm run test:e2e`
+  was executed but skipped because Playwright/Chromium is not installed locally;
+  `npm run release:check` passed with 113 node:test cases total, 112 passed, and
+  1 real-browser smoke skipped by default; npm package contents validation
+  passed with 42 files; extension zip contents validation passed with 13 files.
 
 ## Decisions
 
@@ -102,10 +1054,60 @@ shows token as a first-run field, and the published package is verified with
   `yunti_list_browser_targets` to refresh the route inventory.
 - The content script must not call product-specific login APIs in the
   standalone runtime.
+- `0.2.0` is a Yunti-first enhancement cycle aimed at becoming the most useful
+  local browser automation operation layer for agents and developers.
+- Reference projects such as Playwright, Puppeteer, Selenium, CDP, browser-use,
+  Page Agent, BrowserGym, and extension runtimes are sources of practical ideas,
+  not product shapes to copy.
+- Page Agent has been reviewed as the concrete first reference for P6.1/P6.2:
+  absorb browser-state observation, indexed interactive elements, scrollable
+  metadata, action sequencing, and retry discipline; do not copy its built-in
+  LLM loop or make its UI model mandatory.
+- A multi-perspective 0.2.0 review has been completed across product,
+  architecture/API, Agent workflow, security/privacy, and test/release strategy.
+  The resulting plan changes emphasize product success metrics, field-level
+  observation contracts, uid lifecycle, P6.1 minimum redaction, deterministic
+  fixtures, and observe-first skill/tool guidance.
+- Implementation should proceed step by step: first use Page Agent/browser-use
+  as the concrete reference for P6.1 page observation and P6.2 indexed actions,
+  then absorb Playwright/CDP/BrowserGym-style reliability and diagnostics later.
+- Each `0.2.0` implementation slice should be documented before code changes:
+  user-facing capability, compatibility rule, non-goals, acceptance checks, and
+  why it strengthens Yunti's own direction.
+- Yunti should not require an LLM API key, mandatory hub tab, side panel, or
+  single black-box `execute_task` tool in the default path.
+- Yunti should preserve its own distinctive capabilities: fine-grained MCP
+  tools, real Chrome/Edge state, CDP, screenshots, network/console diagnostics,
+  tab control, local bridge routing, and zero-config local install.
+- P6.1 implementation is proceeding in small slices: schema/tool hints/bridge
+  routing are complete, content-script observation MVP is underway, and fresh
+  observe uids now feed existing uid-based actions without replacing the
+  snapshot path.
 
 ## Open Work
 
-- Follow `docs/EXECUTION_PLAN.md` when defining the next post-0.1.1 phase.
+- Follow `docs/NEXT_MAJOR_PLAN.md` for the next major cycle.
+- P6.1 real-browser closure is complete: `YUNTI_E2E=1 npm run test:e2e` passed
+  locally with temporary Node Playwright 1.58.0 and the existing Playwright
+  browser cache from `/opt/miniconda3/envs/pytest-playwright`.
+- Continue 0.2.0 in small compatibility-preserving slices. Current completed
+  subphases are P6.2.1 action result coverage, P6.2.2 select semantics, P6.2.3
+  fill/form diagnostics, P6.2.4 scroll diagnostics, P6.2.5 wait-observe
+  guidance, P6.2.6 structured `yunti_wait_for` results, and P6.2.7 automated
+  action result coverage audit, P6.3.1 agent workflow contract, P6.3.2 minimal
+  Tool Guide use cases, P6.4.1 DOM redaction policy deepening, P6.4.2 learning
+  memory / console diagnostic secret-boundary tightening, P6.4.3 raw CDP /
+  screenshot non-redaction diagnostic guidance, P6.5.1 optional local runtime
+  console minimum loop, P6.5.2 local console diagnostic polish, and P6.5.3
+  real-browser console validation. P6.6.1 browser extension distribution
+  readiness, P6.6.2 store-facing permission/privacy copy, and P6.6.3 pre-store
+  permission strategy decision are complete. P6.6.4 store-candidate permission
+  UX design is deferred to post-0.2. The next anchored slice is P7.1 `0.2.0`
+  npm/unpacked release closure.
+- For the next coding slice, update all affected guidance surfaces in one
+  commit: `mcp/tools.js`, `docs/TOOL_GUIDE.md`,
+  `skills/yunti-browser-runtime/SKILL.md`, `docs/EXECUTION_PLAN.md`, and this
+  status file.
 - P4.1 release-readiness docs and permission review is complete.
 - P4.2 Agent integration examples are complete.
 - P4.3 npm publishing URL confirmation is complete with the GitHub repository
@@ -258,6 +1260,15 @@ shows token as a first-run field, and the published package is verified with
 - Latest P5.2 validation: `npm run release:check` passed; `npm run
   release:publish` published `yunti-browser-runtime@0.1.2`; `npm run
   release:verify-published` returned 0.
+- P5.3 extension zero-config first-run is complete: popup first screen now only
+  shows connection status plus refresh, optional Bridge URL/page match/token
+  controls are hidden under advanced settings, and docs explain that the MCP
+  server starts the local bridge automatically.
+- Latest P5.3 validation: `npm run release:check` passed; `npm run
+  release:publish` published `yunti-browser-runtime@0.1.3`; a first
+  `npm run release:verify-published` hit npm registry lag, then a retry passed
+  and `npm view yunti-browser-runtime version dist-tags.latest` returned
+  `0.1.3`.
 
 ## Known Risks