@adia-ai/adia-ui-forge 0.8.45 → 0.8.46

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "adia-ui-kit-forge",
-  "version": "0.8.45",
+  "version": "0.8.46",
   "description": "Maintain the adia-ui (@adia-ai) framework itself \u2014 author primitives and shells, run the A2UI generation pipeline and its corpus, review gen-UI quality, sweep QA, cut releases, deploy. The maintainer counterpart to adia-ui-kit-factory (the consumer/app-author plugin).",
   "author": {
     "name": "Kim",

package/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog — adia-ui-kit-forge
 
+## [0.8.46] — 2026-08-22
+
+
+### Maintenance
+- **`.claude-plugin/plugin.json` version bump** — moves in lockstep with package.json (the `/plugin update` cache key).
+- **`agents/` touched in this release window** (4 file(s), e.g. `agents/demo-audit-agent.corpus.json`) — carried by the entries above.
+- **`skills/` touched in this release window** (2 file(s), e.g. `references/authoring-cycle.md`) — carried by the entries above.
+
 ## [0.8.45] — 2026-08-20
 
 ### Changed

package/agents/demo-audit-agent.corpus.json

@@ -25,11 +25,11 @@
     "update the dogfood findings tracker after fixing the leaks"
   ],
   "_measured": {
-    "as_of": "2026-08-20",
+    "as_of": "2026-08-21",
     "scorer": "routing_eval.py (nonoun-plugins/forge)",
-    "note": "measured-clear vs 0.70 floor (routing_eval.py, lexical-overlap legibility aid); split from framework-reviewer per gh#1733/PR #1747",
-    "precision": 0.818,
+    "note": "measured-clear vs 0.70 floor (routing_eval.py, lexical-overlap legibility aid); NOT-fence tightened gh#1819 — the two 2026-08-20 precision holes (\"fix the native-primitive leaks the sweep found\", \"which npm script runs the visual probe\") no longer stolen; split from framework-reviewer per gh#1733/PR #1747",
+    "precision": 1.0,
     "recall": 0.9,
-    "f1": 0.857
+    "f1": 0.947
   }
 }

package/agents/demo-audit-agent.md

@@ -9,7 +9,9 @@ description: |
   Reports findings only — applying a repair is a separate, later dispatch
   (generator ≠ critic).
   NOT for A2UI-engine gallery output scoring (gen-ui-review-agent); NOT for
-  applying a fix (primitive-authoring-agent) — this seat only reports.
+  applying a fix or patching what a sweep already found
+  (primitive-authoring-agent); NOT for looking up which npm script or
+  command a probe needs (see the skill docs) — this seat only reports.
 tools: Read, Grep, Glob, Bash
 skills:
   - demo-audit

package/agents/gen-ui-review-agent.corpus.json

@@ -20,11 +20,11 @@
     "visual probe the components in light and dark"
   ],
   "_measured": {
-    "as_of": "2026-08-20",
+    "as_of": "2026-08-21",
     "scorer": "routing_eval.py (nonoun-plugins/forge)",
-    "note": "measured-clear vs 0.70 floor (routing_eval.py, lexical-overlap legibility aid); split from framework-reviewer per gh#1733/PR #1747",
-    "precision": 0.714,
+    "note": "measured-clear vs 0.70 floor (routing_eval.py, lexical-overlap legibility aid); NOT-fence tightened gh#1819 — the two 2026-08-20 precision holes (\"explain the gen-review exit gate thresholds\", \"apply the corpus fixes for the failing prompts\") no longer stolen; split from framework-reviewer per gh#1733/PR #1747",
+    "precision": 1.0,
     "recall": 1.0,
-    "f1": 0.833
+    "f1": 1.0
   }
 }

package/agents/gen-ui-review-agent.md

@@ -7,10 +7,12 @@ description: |
   score the gallery before a merge or cut, or whenever someone asks "score
   the gen-ui gallery outputs", "review gen-ui outputs", "is the gallery
   clean".
-  Reports findings only — applying a repair (a corpus/pipeline fix) is a
-  separate, later dispatch (generator ≠ critic).
+  Reports findings only — a repair lands as a separate, later dispatch
+  (generator ≠ critic).
   NOT for demo/app-shell dogfood sweeps (demo-audit-agent); NOT for applying
-  a corpus fix (a2ui-maintenance-agent) — this seat only reports.
+  a corpus or pipeline fix (a2ui-maintenance-agent); NOT for explaining what
+  the exit gate thresholds mean — this seat scores against them, it does
+  not explain them.
 tools: Read, Grep, Glob, Bash
 skills:
   - gen-ui-review

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@adia-ai/adia-ui-forge",
-  "version": "0.8.45",
+  "version": "0.8.46",
   "description": "Maintain the adia-ui (@adia-ai) framework itself \u2014 author primitives and shells, run the A2UI generation pipeline and its corpus, review gen-UI quality, sweep QA, cut releases, deploy. The maintainer counterpart to adia-factory (the consumer/app-author plugin).",
   "keywords": [
     "adia-ui",

package/skills/primitive-authoring/references/authoring-cycle.md

@@ -210,6 +210,23 @@ not just one — `theme-provider.min.js` (a JS bundle) inlines the co-emitted
 bundle if only `build:bundle-css` runs (bit the v0.8.13 theme-panel work
 twice).
 
+**Rebuild JS bundles from an `npm ci` scratch checkout, never a pnpm-bootstrapped worktree.**
+`node scripts/dev/bootstrap-worktree.mjs`'s pnpm install produces `dist/*.min.js` output
+~30% larger than CI's npm-ci install (a lockfile/hoisting difference, not a code difference)
+— running `build:bundle-js` in such a worktree commits drift instead of fixing it, and a
+local `check:js-bundles-fresh` run there will falsely pass against its own inflated bundle.
+When a PR built in a pnpm worktree needs a JS-bundle rebuild: commit the CSS-side changes
+first if any (`build:bundle-css` has no install-shape risk — LightningCSS's bundling is
+deterministic from source — so the branch's committed `dist/web-components.sheet.js` is
+trustworthy going into the next step, satisfying the CSS-before-JS ordering above), then
+`git worktree add <scratch-path> <branch>`, `npm ci` there (real npm-shaped install), run
+`build:bundle-js`, copy only the changed `dist/*.min.js` file(s) back into the working
+worktree, commit there with an explicit pathspec, then remove the scratch worktree. This
+is the correction to gh#1825's own PR body, which deferred the rebuild to "CI's
+derived-artifact pipeline" — wrong per ADR-0069 (`dist/` is Class C, committed and
+PR-blocking, never derive-resync'd); the scratch-checkout procedure above is what actually
+landed the fix, 2026-08-21.
+
 **Never `@import` a remote URL from any file `build:bundle-css` bundles.**
 LightningCSS's `bundleAsync` (the engine behind it) resolves every `@import`
 as a local filesystem path, remote `http(s)` URLs included — a CSS-level

package/skills/primitive-authoring/references/svg-authoring.md

@@ -81,45 +81,45 @@ clips at `overflow: hidden` with a rounded `border-radius` (`card.css`, top
 of file), and `[bleed]` zeroes the section's own margin/padding
 (`card.css:342` onward) — so a chart's axis-label overhang, or gridlines
 extending to the plot edge, lands flush against that rounded corner and
-clips silently. This was gh#1095's real incident (PR #1105): a bar/line
-chart's Y-axis labels clipped under a bled card's corner.
-
-**The fix is the canonical convention** — `card.css:400-422` (comment block
-+ rule):
-
-```css
-/* Guide/value/legend safety inset (gh#1095) — [bleed]'s zero-inset is
-   only collision-free for the bare-marks case: a sparkline, or any other
-   chart-ui type with BOTH [no-grid] and [no-values] set. */
-& > section[bleed]:has(> chart-ui:not([type="sparkline"]):not([no-grid])),
-& > section[bleed]:has(> chart-ui:not([type="sparkline"]):not([no-values])),
-& > section[bleed]:has(> chart-legend-ui) {
-  margin: var(--card-inset);
-  padding: 0;
-}
-```
-
-This restores the card's own inset automatically the moment a bled chart
-still draws guides/values/a legend — a `:has()`-based fail-safe rather than
-relying on every author to remember. `#renderSparkline()` is the ONLY
-renderer that never emits axis ticks, gridlines, value text, or a legend
-(`chart.class.js`'s sparkline branch) — it's the sole type where genuine
-edge-to-edge bleed is collision-free with no attribute needed. Every other
-type needs BOTH `[no-grid]` and `[no-values]` set to opt into true
-bleed; short of that, the safety inset applies and the chart renders inset
-like a normal card section — expected, not a bug.
+clips silently. This was gh#1095's original incident (PR #1105): a
+bar/line chart's Y-axis labels clipped under a bled card's corner.
+
+**gh#1095's own mechanical fix — a `:has()`-based auto-restore of the
+card's inset — was itself unratified and removed (gh#1801, operator ruling
+2026-08-20)**: it silently defeated an author's own `[bleed]` the moment a
+chart drew any guide/value text or paired a `<chart-legend-ui>`, which
+collided with the card-chart design language's overlay-chip labels
+(rendered INSET within the plot box on purpose, but still enough to trip
+the old guard's `:not([no-grid])`/`:not([no-values])` test). `[bleed]` is
+now unconditionally author-controlled — card.css never re-inserts an inset
+the author explicitly zeroed.
+
+**The clipping hazard itself is real and unchanged** — only the mitigation
+moved from mechanical CSS to documented author responsibility. card.yaml's
+`bleed` prop docs and `chart-in-card.examples.html` both carry the
+resulting rule: a full-bleed chart-ui with visible guide/value text, or one
+paired with a `<chart-legend-ui>`, must keep that text clear of the card's
+rounded-corner clip — e.g. by rendering it as an overlay chip INSET within
+the plot area (never hanging outside it, and never relying on card margin
+for clearance) — or by putting the legend in its own non-bled section.
+`#renderSparkline()` is still the only renderer that never emits axis
+ticks, gridlines, value text, or a legend (`chart.class.js`'s sparkline
+branch), so it's still the one type where a bare `[bleed]` needs no such
+care; every other type is the author's call now, not a stylesheet fail-safe.
 
 **The generalized rule for any new SVG primitive placed inside a bleed
 section**: an SVG whose content can extend past its own nominal box
 (`overflow: visible`, or geometry computed with negative padding) needs
-either a "bare marks" mode (no overhanging content) that's safe to bleed, or
-a CSS `:has()` safety net analogous to `card.css:400-422` that restores
-inset automatically when the overhanging content is present. Don't assume
+either a "bare marks" mode (no overhanging content) that's safe to bleed,
+or a documented author-responsibility note at the point of use (the
+`bleed`-prop docs, the pattern's own examples) naming the clipping hazard
+and its mitigation — never a mechanical CSS guard that silently overrides
+an author's own explicit attribute (gh#1801's own lesson). Don't assume
 `overflow: hidden` on the ancestor container will clip cleanly — SVG content
 drawn PAST an ancestor's padding box (not its own) clips at whatever
 ancestor in the chain actually sets `overflow: hidden`, which for
 `card-ui` is the rounded-corner boundary itself, producing the specific
-silently-clipped-under-a-curve look #1095 reported.
+silently-clipped-under-a-curve look #1095 originally reported.
 
 ## 5. CSS custom properties don't resolve inside raw SVG attribute strings — only inside actual CSS declarations