@reqvire-org/ads 0.3.17 → 0.3.18

package/dist/skills/ads-motion/references/captures.md

@@ -101,7 +101,12 @@ failed/cancelled capture does not replace existing image slots or apply source.
 If a scenario/target is newly introduced, `current` is null with an explicit
 `currentAbsent` explanation; only the real candidate image exists.
 
-Review the returned images and diagnostics, fix the draft if necessary, then apply
+The edit reply is a summary. Read `ads_results({resultId,section:"captures"})` for
+the requested image resource URIs and `section:"diagnostics"` for relevant
+messages; follow `nextCursor` when needed. Read image resources explicitly.
+Standalone `ads_capture` continues returning its selected capture directly.
+
+Review those images and diagnostics, fix the draft if necessary, then apply
 with `dryRun:false`. Apply reruns all gates. An image is promoted only when the
 accepted source revision matches the captured candidate; other proposals remain
 stale. Images do not approve a patch or certify interaction/accessibility/parity.

package/dist/skills/ads-motion/references/evidence.md

@@ -20,21 +20,67 @@ A complete failed run can contain independent passing cases. Suite interruption,
 missing/skipped cases, stale inputs and an unexplained process exit cannot certify
 a pass. Keep the reported case outcome separate from whole-run acceptance.
 
+## Retrieve only the evidence needed
+
+`ads_verify`, `ads_apply_patch` and `ads_apply_transaction` reply with counts,
+bounded acceptance, at most three failure previews and `resultId`, within 4096
+UTF-8 JSON bytes. Full validation still decides acceptance. Do not repeat a suite
+merely to recover its output, or restate the complete report in conversation.
+
+Read a relevant saved section:
+
+```js
+ads_results({resultId: "<returned ID>", section: "outcomes", status: "not_passed"})
+// Continue only when needed, with the same section and status:
+ads_results({resultId: "<returned ID>", section: "outcomes", status: "not_passed", cursor: "<nextCursor>"})
+```
+
+Pages default to 10 rows, allow at most 20 and stay within 8192 JSON bytes. The
+byte cap can return fewer rows; `nextCursor` is the continuation. Other sections
+are `checks` (pipeline steps), `diagnostics`, `advisories`, `warnings`, `changes`,
+`captures`, `acceptance`, `investigation` (mutation boundary) and `selection`.
+A row marked `detailTruncated` is a preview. To read that exact record, pass its
+`recordIndex`, the same result/section, and `offset`/`chars` (maximum 2000 UTF-16
+characters); follow `nextOffset`. Do not combine ranges with a cursor/status filter.
+Read only the relevant log range. Reading pages or ranges never executes checks.
+
+`ads://project/results` lists the latest IDs. `ads://results/<resultId>` reads a
+summary; matching query parameters read pages/ranges. Only the latest verification
+and latest mutation snapshots are retained, independently, up to 8 MiB each.
+A superseded ID fails explicitly. An oversized report retains only its summary
+with a warning; disk failure retains details in server memory until shutdown.
+Neither changes actual acceptance or applied state. These are logical JSON byte
+budgets, not token estimates; compatibility text can duplicate structured JSON.
+Saved outcomes describe completion-time evidence, not current source freshness;
+compare their revisions with `ads_behaviors` before claiming a current pass.
+
+`ads_behaviors` pages owner summaries, then an exact unit's `checks`, `guarantees`,
+`subjects`, `demonstrations` or `profiles`. Use `nextCursor` and identical filters;
+changes invalidate a discovery cursor. Read a truncated contract via its
+`sourceFile` and bounded `ads_read_file`, rather than requesting everything.
+
 ## Get a compact investigation
 
-Discover exact owner/check IDs with `ads_behaviors`, then:
+Take an outcome's exact owner/check/run IDs from `ads_results`, then:
 
 ```json
-{ "unit": "components/Example", "check": "keyboard-navigation" }
+{ "unit": "components/Example", "check": "keyboard-navigation", "runId": "<outcome runId>" }
 ```
 
-Send this to `ads_investigate`. It reads the latest bundle without running code.
+Send this to `ads_investigate`. It reads a summary within 4096 JSON bytes, without
+running code. The run ID rejects evidence superseded by another execution.
+`section: "details"`, `"requests"`, `"browserErrors"` or `"artifacts"` reads separate pages
+(default 10/max 20 rows, at most 8192 JSON bytes); artifacts themselves require
+explicit resource reads. Truncated investigation records also support
+`recordIndex`/`offset`/`chars`; keep the exact `runId` to avoid mixing evidence.
+Without a check, the tool pages available investigations.
 `ads://project/investigations` lists available entries; the returned `resourceUri`
 reads the same bundle. It records:
 
 - Project versus isolated candidate, configured framework/profile and scenario.
 - Exact input revision and the preview revision actually observed by the helper.
-- Failing step/source, observed startup viewport and relevant failed requests.
+- `details.failingStep`, `details.source`, startup viewport and failure boundary.
+- Counts for separately paged relevant requests, browser errors and artifacts.
 - Available registered trace, screenshot or context artifact resource URIs.
 - Explicit `acceptance.established` and `acceptance.unverified` scope.
 
@@ -58,7 +104,8 @@ Candidate evidence never becomes passing live-source evidence. Compare the
 recorded input revision with current `ads_behaviors` evidence; a bundle describes
 its run, and does not claim current source merely because it remains available.
 For a build/architecture rejection before behavior runs, inspect the mutation
-result's `investigation` (also retained in `ads_status.mutations.lastResult`).
+saved `investigation` section via `ads_results`. Status retains the compact
+execution summary and result ID, rather than embedding complete reports.
 
 ## Preserve the design while repairing the failure
 

package/dist/skills/ads-motion/references/guarded-editing.md

@@ -15,8 +15,10 @@ ads_apply_patch({"unit":"components/Example","patchFile":"<returned patchFile>",
 
 The dry run validates the full candidate, including affected registered behavior
 checks in an isolated project copy. Configured browser checks use that candidate's
-managed preview, not the live showcase. Inspect the returned `behavior` failures
-or the registered behavior-check results in `checks`, fix the draft and diff again.
+managed preview, not the live showcase. Read the summary, then page failing
+`outcomes`, `diagnostics` or pipeline `checks` through `ads_results` with its
+`resultId`. Inspect one exact outcome with `ads_investigate({unit,check,runId})`;
+fix the draft and diff again. See [evidence.md](evidence.md) for page/range limits.
 Keep behavior.json/specs with their owner and use registered checks; a scratchpad
 run against live source cannot verify unapplied changes. After validation, apply
 with `dryRun:false`; ADS revalidates before writing. Dry-run evidence is not saved
@@ -120,9 +122,10 @@ them through their actual public composition/customization path and list their r
 render references in covers. Do not count internal production renders as public API
 coverage. Behavioral APIs use exercises and registered guarantees instead.
 Scenario controls still use public UI. A native link/button can supply semantics
-as the single direct child of a public runtime slot when source tracing proves
-that owner clones the child with its props. An `asChild` flag alone is insufficient;
-standalone native controls and custom scenario styling remain rejected. See
+through a public runtime slot that demonstrably clones its child with owned
+props, or through matching runtime-owned selectors attached to the actual rendered
+DOM. Source ownership must be proven; an `asChild` flag or a lookalike marker is
+insufficient. Standalone controls and custom scenario styling remain rejected. See
 [scenarios.md](scenarios.md) for evidence and limits. Preserve the intended
 composition instead of substituting an unrelated styled control or adding a
 demo-only component prop. Review framing belongs in public layout patterns.

package/docs/ADS.md

@@ -259,6 +259,16 @@ custom styling in fixture files. The ADS frame pattern uses `--ux-*`. Foreign br
 CSS imports the approved token files in order; shared assignments belong in
 those token files, never in consumer styles or component state overrides.
 
+Typed custom-property registrations follow the same ownership. Runtime-variable
+`@property` metadata lives in component/pattern Linaria; shared and owner-local
+token registrations live beside real token definitions. Names are document-global:
+unrelated owners and conflicting descriptor sets cannot claim the same name.
+Descriptors have their own CSS grammar and independent literal defaults; normal
+visual styling still uses tokens. Registration does not create token/sample
+coverage, and base resets/font metadata cannot host it. Preserve whole rules with
+the ADS Linaria resolver in product and MCP preview builds. See the portable
+[atomic styling guide](../skills/ads-designer/references/atomic-styles.md).
+
 Foundation roles are enforced by `foundation-role`, `foundation-selector` and
 `foundation-import`, alongside the regular token checks. Shared definitions
 belong on the document root or explicit root themes; ordinary declarations belong

package/docs/ARCHITECTURE.md

@@ -155,6 +155,19 @@ values; source guards alone cannot establish extraction correctness. The
 skill-local [atomic styling guide](../skills/ads-designer/references/atomic-styles.md#preserve-custom-property-arithmetic-during-extraction)
 documents integration and source-workaround semantics.
 
+The resolver also preserves complete `@property` registrations and their
+conditional at-rule context for atomic `css`/`styled`. Their descriptors are schema
+metadata, not separate atomic declarations. CSS-tree validates CSS value grammar;
+PostCSS supplies source/at-rule structure. Registered names have one owner and a
+stable schema across its modules, because registration is document-global. Runtime
+variables are registered in owned Linaria; shared or namespaced design tokens are
+registered beside their real value definitions. Base resets and fonts cannot own
+registrations. Descriptor literals are narrowly permitted where CSS requires an
+independent initial value; ordinary visual assignments remain token-governed.
+MCP serves these styles to the sandboxed App document. Registration metadata
+neither creates new token definitions nor replaces token sample coverage.
+Compiled-CSS and browser tests verify interpolation in React and Preact Apps.
+
 ## Public APIs preserve the boundary
 
 The application and project-owned preview scenarios consume a project's public ADS entry. Components
@@ -445,6 +458,20 @@ Another run replaces that check's resource access; candidate evidence never
 certifies live source. Detailed usage lives in the portable skill-local
 [evidence guide](../skills/ads-designer/references/evidence.md).
 
+MCP delivery is separate from validation: verification and mutation controllers
+use complete results to decide acceptance. Their MCP replies contain counts,
+bounded acceptance, at most three failure previews and a result ID in at most
+4 KiB of UTF-8 JSON. `ads_results` reads saved outcomes, diagnostics and other
+sections without rerunning checks. Pages have at most 20 rows and 8 KiB of JSON;
+truncated records support bounded text ranges. Discovery and investigation
+sections use the same paging budget. Cursors bind the query and result/revision;
+superseded IDs fail explicitly instead of silently returning another run.
+Only the latest verification and mutation snapshots are retained, each at most
+8 MiB under `.ds/tool-results/`. Oversized reports retain a summary with an
+explicit unavailable-details warning. Storage/notification failure cannot change
+an accepted edit's applied state. These are payload byte budgets, not tokenizer
+estimates; MCP compatibility text may duplicate structured JSON on the wire.
+
 Edits stay focused without requiring entire file bodies in model context. Peer units and
 unrelated owners remain outside the selected unit's edit boundary. Shared foundation,
 branding and guides have their own units. Shared credits and text license notices

package/docs/BEHAVIORS.md

@@ -308,8 +308,20 @@ Guarantees report passing only when all their checks have current passing eviden
 Passing applies to those cases and profiles, not every application state.
 
 Results expose structured `failure` boundaries and explicit `acceptance` scope.
-Use `ads_investigate({unit,check})` for the latest compact investigation, including
-isolated dry runs; `ads://project/investigations` lists available entries. Exact
+MCP execution replies contain counts, bounded acceptance, at most three failure
+previews and a `resultId` (4 KiB UTF-8 JSON maximum). Use `ads_results` to read
+selected `outcomes`, `checks`, `diagnostics` or other sections without execution.
+Pages default to 10 rows, allow at most 20 and stay within 8 KiB; follow
+`nextCursor` with identical selection/status. Truncated records provide
+`recordIndex` for exact `offset`/`chars` ranges. Only the latest verification and
+mutation snapshots are retained, independently, at most 8 MiB each. Replaced IDs
+fail explicitly. Validation still uses complete internal outcomes.
+
+Use `ads_investigate({unit,check,runId})` with an outcome's exact IDs for its compact
+summary (4 KiB maximum); details, requests, browser errors and artifact metadata are separate
+paged sections (8 KiB maximum). Truncated records support exact text ranges with
+the recorded run ID. Saved outcomes describe completion time; compare current
+`ads_behaviors` revisions for freshness. It includes isolated dry runs; `ads://project/investigations` lists available entries. Exact
 input and observed preview revisions, failing step/source, startup viewport,
 failed startup requests and available project-recorded artifacts keep diagnosis
 separate from an inferred component defect. No artifact is invented or test

package/docs/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## 0.3.18
+
+- Preserve complete typed `@property` registrations through atomic css/styled
+  extraction, conditional rules and CSS minification. Validate descriptors with
+  CSS-tree, enforce runtime/token ownership and naming conflicts, and retain
+  ordinary style-token gates. Verify dummy interpolation in React/Preact MCP Apps.
+- Sync all skill references for registrations and selector-owned native
+  composition; keep implementation independent of any imported component library.
+
+- Deliver verification and guarded-edit replies as 4 KiB JSON summaries with
+  counts, acceptance and result IDs. Page outcomes, diagnostics, logs and captures
+  through `ads_results`; retain only the latest verification and mutation snapshots.
+  Keep complete internal validation independent of delivery.
+- Page behavior discovery and investigation details; bind cursors to their query
+  and recorded run, reject superseded evidence, and document selective retrieval
+  in every portable skill. Keep artifacts separate from tool JSON.
+
 ## 0.3.17
 
 - Recognize native semantic composition through source-verified runtime Linaria

package/docs/MCP.md

@@ -175,9 +175,10 @@ The server supplies agent guidance through its instructions, tool descriptions,
 | `ads_show_tokens` | Render a discovered token scenario; legacy compilers retain the definition inspector |
 | `ads_scenarios` | Bounded token/component/pattern/composition scenario discovery |
 | `ads_show_scenario` | Render an exact discovered scenario with its own fixtures |
-| `ads_behaviors` | Bounded guarantee/evidence discovery; pass an exact unit for contract details |
+| `ads_behaviors` | Paged owner discovery; select a unit and detail section |
 | `ads_verify` | Execute an owner's registered checks and record revision-bound evidence |
-| `ads_investigate` | Read one check's latest compact failure/evidence bundle without rerunning it |
+| `ads_results` | Page saved execution outcomes, diagnostics, logs and captures without rerunning |
+| `ads_investigate` | Read one check's summary, then paged requests/errors/artifacts |
 | `ads_capture` | On-demand PNG of a discovered scenario or named element; replace current and clear candidate |
 | `ads_capture_clear` | Remove generated captures or discard only the candidate |
 | `ads_show_<id>` | Optional explicitly configured grouped App |
@@ -193,8 +194,9 @@ public exports from `components/`, `patterns/` and `controllers/`; it does not e
 guess component props. Read the exported API and owning guide. An export name
 alone does not prove that a preview renders that particular element.
 
-`ads://project/status`, `/build`, `/tokens`, `/catalog`, `/scenarios`, `/behaviors`, and `/checks` carry complete JSON
-detail. Allowed project documentation/source and skill files are separate
+`ads://project/status`, `/build`, `/tokens`, `/catalog`, `/scenarios` and `/checks`
+carry project JSON detail. Behavior discovery, execution outcomes and investigation
+details are paged; their resource reads use the same bounds as their tools. Allowed project documentation/source and skill files are separate
 resources. No arbitrary filesystem reader is exposed. Prefer bounded tools for
 large catalogues; explicit resource reads can enter model context and a client
 may truncate large payloads.
@@ -215,6 +217,14 @@ informational limit nor a traced target proves a rendered failure or success.
 Bounded check results put warnings before informational limits; the checks
 resource retains complete results.
 
+`property-registration` is a blocking architecture diagnostic for invalid or
+conflicting typed CSS registrations. Keep registrations in their runtime owner
+or beside their real token definitions; `base.css` remains a closed reset contract.
+Use `@reqvire-org/ads/linaria` in both product and preview extraction so each
+registration survives intact in the App's CSS resources. Descriptor literals are
+metadata; normal styling still uses tokens. Read `ads://guides/atomic-styles` for
+ownership, valid initial values, browser checks and examples.
+
 ### Screenshots and candidate comparison
 
 `ads_capture({scenario:"<discovered ID>",target?,theme?,viewport?})` renders current
@@ -249,8 +259,10 @@ limits, target declarations and the guarded review workflow.
 
 Read `ads://guides/behaviors` or the [behavior guide](BEHAVIORS.md) for the contract
 and runner schema. `ads_behaviors({ "unit": "components/MessageScroller" })`
-returns subjects, guarantees, demonstration links and check IDs for a discovered
-owner. `ads_verify({ "unit": "components/MessageScroller", "checks": ["prepend-anchor"] })`
+pages an owner's checks by default; select `section: "subjects"`, `"guarantees"`,
+`"demonstrations"` or `"profiles"` for other details. Without a unit it lists owner
+summaries. Follow `nextCursor` with identical filters. Changed registrations or
+evidence revisions invalidate discovery cursors. `ads_verify({ "unit": "components/MessageScroller", "checks": ["prepend-anchor"] })`
 runs those registered cases. The names here are examples, never built-in cases.
 
 `covers` describes rendered UI; optional `exercises` describes public behavioral
@@ -276,14 +288,47 @@ single failed assertion does not turn the other checks into failures. Overall
 verification and guarded application still fail. Suite errors, unexplained exits,
 missing results and interrupted reports cannot certify passing checks.
 
-For diagnosis, call `ads_investigate({ "unit": "components/Example", "check": "keyboard-navigation" })`
-with discovered IDs. It reads the latest check bundle without rerunning anything,
-including rejected candidates. `ads://project/investigations` lists available
-bundles; returned resource URIs expose the bundle and available registered
+Verification and guarded-edit replies are summaries: counts, acceptance, at most
+three failure previews and `resultId`, within 4096 UTF-8 JSON bytes. Use:
+
+```js
+ads_results({resultId: "<returned ID>", section: "outcomes", status: "not_passed"})
+ads_results({resultId: "<returned ID>", section: "outcomes", status: "not_passed", cursor: "<nextCursor>"})
+```
+
+Each listing defaults to 10 rows, allows at most 20 and stays within 8192 JSON
+bytes. A byte-limited page can contain fewer than `limit` rows; follow
+`nextCursor` with the same result ID, section and status. Other sections are
+`checks`, `advisories`, `warnings`, `changes`, `captures`, `acceptance`,
+`investigation` and `selection`. No page reruns validation. Rows marked
+`detailTruncated` can be read exactly with `recordIndex`, `offset` and `chars`
+(maximum 2000 UTF-16 characters); follow `nextOffset`. Read full logs only for the
+relevant record. JSON also remains available through `ads://results/<resultId>`;
+use matching query parameters for pages or ranges. `ads://project/results` lists
+the latest IDs. Status embeds summaries, not complete runner reports. Saved
+outcomes are completion-time evidence; compare revisions with `ads_behaviors`
+before claiming current source passed.
+
+Only the latest verification snapshot and latest mutation snapshot are retained,
+independently, each up to 8 MiB in `.ds/tool-results/`. Replacing a snapshot makes
+its old ID unavailable; do not mix pages across runs. An oversized report retains
+only its summary and an explicit unavailable-details warning. A failed disk write
+keeps details in server memory until shutdown. Neither condition changes guard
+acceptance or whether source was applied. These are logical payload byte limits;
+compatibility text can duplicate structured JSON in the MCP response.
+
+For diagnosis, take one outcome's exact `unit`, `check` and `runId` and call
+`ads_investigate`. The run ID rejects superseded evidence. Without a check, it
+lists available investigations with paging. An exact check returns a summary
+within 4096 JSON bytes; `section: "details"`, `"requests"`, `"browserErrors"` or `"artifacts"`
+returns detail pages with the same 10/20-row and 8192-byte limits. Truncated records support
+`recordIndex`/`offset`/`chars` with the exact `runId`; artifact URIs are read
+separately. It never reruns checks, including rejected candidates.
+`ads://project/investigations` lists available bundles; returned resource URIs expose the bundle and available registered
 artifacts. Exact input/observed preview revisions, source/failing step, startup
 viewport and failed requests separate evidence from a suspected cause.
-`failure.category` distinguishes architecture, build, transport, runner and
-assertion boundaries; `acceptance` names established and unverified scope.
+The summary's `details.failure.category` distinguishes architecture, build,
+transport, runner and assertion boundaries; `acceptance` names established and unverified scope.
 Only the latest bundle per check remains resource-accessible. Read the portable
 [evidence guide](../skills/ads-designer/references/evidence.md) for collection
 limits and project-controlled trace/screenshot recording.
@@ -424,8 +469,9 @@ Pass that `patchFile` to `ads_apply_patch` with `dryRun:true` to validate before
 applying. This runs the normal candidate gates, including affected registered
 behavior checks against an isolated candidate and its configured browser preview.
 It writes no live source and does not persist passing evidence for live inputs.
-Read returned `behavior` failures or behavior-check results in `checks`; update the
-draft, diff again and repeat as needed. Apply with `dryRun:false` after validation;
+Read the summary, then use its `resultId` with `ads_results` for failing
+`outcomes`, `diagnostics` or named pipeline `checks`; update the draft, diff again
+and repeat as needed. Apply with `dryRun:false` after validation;
 the gates run again. `ads_verify({unit,checks?})` checks current source and accepts
 no draft argument; a selected-check draft runner is not currently exposed.
 Use `action:"discard"` to remove the draft. Remote clients can continue sending

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@reqvire-org/ads",
-  "version": "0.3.17",
+  "version": "0.3.18",
   "type": "module",
   "description": "A design system your agent can see and shape. Your components. Your look. Your code.",
   "repository": {
@@ -52,6 +52,7 @@
   },
   "dependencies": {
     "@modelcontextprotocol/server": "2.3.1",
+    "css-tree": "3.2.1",
     "postcss": "^8.5.30",
     "postcss-selector-parser": "^7.1.6",
     "postcss-value-parser": "^4.2.0",