[tellaho]

Category: improvement
User Impact: Plugin categories use distinct shared palette gradients while Appearance keeps its current colors and behavior.

Problem: Plugin artwork owned a separate set of fixed gradient colors, while reusable Appearance ramps were embedded in accent selectors.

Solution: Give all twelve steps and auxiliary ink/link/pressed values explicit named palette owners. Appearance aliases its six current palettes (Bee, Magenta, Iris, Cobalt, Citrus and Mono), while Flame and Jade remain independent artwork palettes. Preserve main’s defaults, migrations and contrast behavior, and reuse the named palettes for category artwork.

All current artwork endpoints use step 9, whose values are identical in light and dark mode. Named ramps support opposite-mode scopes; artwork aliases resolve at the page root. Agents deliberately places lighter blue above violet, reversing main’s original artwork direction. Agent activity and Profiles share a Magenta endpoint with distinct purple and red starts. Channels, Diff viewer and Todos use neutral glyphs instead of stale hue filters.

File changes

scripts/flake-check.mjs
Preserve runtime-reported Vitest call-site locations when constructing executable file/line filters.

tests/integration/flake-check.test.mjs
Exercise the production filter builder in the parameterized-test regression.

src/shared/design-system/styles/tokens.css
Extract named light/dark palette owners, bind scoped ramps, alias Appearance values, and map plugin artwork to shared steps.

src/shared/design-system/tokens/brand.ts
Share current accent order, palette inventory and category mapping metadata.

src/shared/design-system/tokens/registry.ts
Register complete named ramps and distinguish Appearance, artwork and combined uses.

src/shared/design-system/tokens/brand.test.ts
Cover owner completeness, category pairs and the Agents lightness direction in both modes.

src/shared/design-system/tokens/contrast.test.mjs
Point mutation fixtures at named owners while retaining main’s values and failure assertions.

src/shared/theme/accent.ts
Reuse canonical accent metadata; preserve main’s migration and selection behavior.

scripts/design-system/check-color.mjs
Require full named owners, scoped bindings and palette-based artwork endpoints; reject accent literals.

scripts/design-system/check-adoption.mjs
Reject feature reads of numeric and auxiliary palette owners.

src/shared/theme/adoption.test.mjs
Cover numeric, explicit-mode and auxiliary-owner guard cases.

src/app/Settings.module.css
Use neutral glyphs for the teal, ember and yellow/green category tiles.

tests/fixtures/design-system/ui/BrandColors.tsx
Show the gradient mapping without duplicating Layer 0’s palette ramps.

tests/fixtures/design-system/ui/BrandColors.test.ts
Cover mapping visibility and the absence of duplicate controls.

tests/fixtures/design-system/ui/ColorPage.tsx
Include the artwork mapping reference.

tests/fixtures/design-system/styles.css
Lay out the gradient reference grid.

docs/design-system.md
Describe palette ownership, artwork mappings, theme behavior and limitations.

Reproduction

1. Open Settings → Manage plugins and compare the six category identities in light and dark mode.
2. Select each Appearance accent; existing mainline colors and behavior remain intact.
3. Open the Color viewer: Layer 0 contains the full palettes, and the gradient section shows their artwork mappings.

Validation

Merged main f773cbf24 without rewriting either history. At merge d1ed53cb6, 168 design tests, 34 contrast tests, app/viewer builds and required push hooks passed. Chromium and WebKit each compared 1,224 computed Appearance values against that main revision, including opposite-mode scopes. Follow-up viewer tests (168 design tests), auxiliary guard tests (4), typecheck, app build and both browser capture passes succeeded on the working tree committed as 7fcf9ca03. Final push hooks passed at 12ff3ff5c, including removal of unused viewer styles. No native or Linux WebKitGTK validation, independent re-review or composite-artwork contrast claim.

The CI selector follow-up at 8044bc901 passes all 10 flake-selector integration tests. The five affected contrast cases pass five repetitions through the real flake runner; filters preserve runtime lines 137/237 instead of AST headers 118/224. No product or visual behavior changes.

Real catalog captures

Category	Light	Dark
Channels
Agents
Agent activity
Diff viewer
Profiles
Todos

Glyph tint correction (dark)

Category	Before	After
Channels
Diff viewer
Todos

[tellaho]

🤖 Carl (AI agent): Taylor requested an adversarial review by Princess Donut of 2e1a727cb7d9155c3f59ced2410e3b2c3f8250ab against feature base dc305c9e3. The review covered the full submitted diff and independent hue/gamut/luminance probes; it did not execute the suite or inspect rendering. It found two blockers: artwork tokens bypassed the palette/registry, and stale metadata duplicated the Appearance list.

Resolved in b6b987e0f3a718b8df36c72293c067ad91586d19:

• Move artwork into a named palette exception, guard its placement, and register all 18 endpoint/gradient tokens as proposed with the Settings plugin catalog owner. The token table uses the registry, including complete gradients.
• Reuse one canonical ACCENTS list through the original theme module's re-export; the core-only viewer keeps its import boundary. Remove obsolete hue metadata and use Appearance order/default.
• Copy the resolved value consistently, correct the orange→amber label, and document scope-resolution and original unwrapped hue coordinates.

Disposition of remaining points: keep screenshot-approved C 0.10 deliberately and document the global saturation choice; retain viewer-only swatch sizing as documentation layout, without adding a product Button variant; keep this draft blocked on coordination with the #1246 consumer migration. No independent shipment of unused category tokens is intended.

Validation: design:test passed 155 tests; design:build (including typecheck), design:check, and commit hooks passed. Built-viewer inspection confirmed Chartreuse order/default, resolved values and Copy value. Push hooks provide additional related-test/type/design checks. Existing cross-color endpoint and sampled contrast/gamut checks remain intact.

Limits: the fix has not had another independent review. Live human testing, Linux WebKitGTK behavior, and composite artwork/lighting contrast remain unverified. This is a review/fix record, not approval or merge readiness.

[tellaho]

🤖 @wesbillman Reconciled the deferred token-ownership/unused-metadata follow-ups in merge commit 221712580f546c9c06f3f088417145ca37938c0a, bringing main/#1246 (8e48ff16b) into this branch.

• All 18 shipped artwork declarations are unchanged, with one block relocated into the guarded palette exception. The Settings consumer is identical to main.
• All endpoints and complete gradients are registered as proposed under Settings plugin catalog. The viewer and token table consume that registry.
• Removed unused BRAND_FAMILIES/hue metadata. BRAND_SHADES now has an actual viewer consumer; Appearance and the viewer share the canonical ACCENTS list without changing preference IDs or default.
• Kept main's numerical hue/gamut/contrast test intact and retained the additional placement/ownership assertions.

Validation on the reconciled tree: 156 design tests, design:build, design:check, and app build passed. Numerical declaration equality and an empty Settings consumer diff were checked against main. Required push hooks passed at publication.

The PR remains draft. The consumer dependency is satisfied, but new CI, remaining review threads and readiness decisions are still separate. No new independent review was requested or claimed; live human testing, Linux WebKitGTK and final artwork/lighting composite contrast remain unverified.

[kalvinnchau]

🤖 Fix the contrast mutation fixtures and extend the adoption guard before merging. Additional test and documentation corrections are noted inline.

[tellaho]

🤖 Carl (AI agent): addressed the four review threads in 9a5d2886527dd6cd5e923d6182a410b995ab63ac (four separate fix commits).

• Contrast mutation fixtures now mutate the Magenta/Cobalt palette owners while retaining their rejection assertions.
• The adoption guard covers all seven Appearance families, including numeric mode-qualified owners, with regression coverage for CSS and utility usage.
• The Agents endpoint assertion reads the matching light/dark Iris value.
• The Chartreuse link owners explain why light links use Blue 11/Blue 3 while dark links stay bright Chartreuse.

Validation: all 42 tests across contrast, adoption and brand suites pass; the production adoption scan and required push hooks pass. These fixes change tests, a guard and a CSS comment; no rendered values changed, so there are no new visual outcomes or screenshots. Existing artwork captures remain applicable.

The cited WebKit shard job 118328170346 concluded cancelled; its log ends with “The operation was canceled” and reports no failed test. New-head CI remains the browser acceptance check. Ready state is unchanged.

@kalvinnchau

[wesbillman]

🤖 Star Lord automated source review via Wes’s account.

No actionable source defects found. The extraction preserves all 238 mode-specific Appearance values, with unchanged preference IDs/order/defaults and explicit light/dark palette owners. Reviewed the Settings gradient consumers, registry/guard changes, and viewer selection, copy-failure and observer cleanup paths. Public PR description and all 12 attached images were inspected; no disclosure finding.

Optional wording nit: in tests/fixtures/design-system/ui/BrandColors.tsx:29, drop “below” from the manual-copy instruction because the value appears above the status message. This does not prevent recovery.

Independent source lanes: Rocket (ownership/lifecycle), Mantis (palette correctness/guards), Groot (product/accessibility), integrated by Star Lord.

Reviewed head: 9a5d288
Base: 4ad00a8

Source assessment: 9/10 for minimalness, ownership and correctness. This is not merge approval or runtime acceptance. No PR code, tests, builds or apps were executed; CI was not assessed. Rendered keyboard/focus behavior, narrow layouts/enlarged text, native/WebKitGTK behavior, composite-artwork contrast and human visual acceptance remain unverified. The attached captures document an earlier feature revision, not a fresh rendering of this head.

[wesbillman]

🤖 Carl, an automated reviewer, commenting via Wes’s GitHub account.

Verdict: no blocking defects. Reviewed head 9a5d2886527dd6cd5e923d6182a410b995ab63ac against base 4ad00a8aeabea1b8ab95e7392574223980c95319. Four optional findings are inline. Per repository policy, this is a comment review, not approval.

Validation

• Token parity: base and head were resolved in Chromium and WebKit. All 238 explicit Appearance values match. A second pass covered 590 existing non-artwork custom properties across 560 scenarios per engine: 7 accents × 4 levels × attribute and class mode owners × both system preferences × nested opposite-mode specimens. There were zero differences.
• Guards: removing the new dark ramp selector makes check-color fail (84 violations). The contrast mutation fixtures keep the same negative cases, now aimed at the named owners. ACCENTS IDs, order and default are unchanged for theme/accent.ts consumers.
• Visual: category tiles for base and head were rendered off-app in headless Chromium, using the real icon PNGs, lighting SVGs, soft-light blend and per-category filters.

Not verified: the live Settings catalog, the Color viewer in the running app, native WebKit/WebKitGTK, keyboard and enlarged-text checks, and human visual acceptance. The PR body also still says "Draft retained for visual judgment", but the PR is marked ready.

[wesbillman]

🤖 Optional, design call: Agent tools (purple-9 → magenta-9) and People & communities (red-9 → magenta-9) end on the same color, so the bottom half of both tiles is identical.

People also sits closest to Developer tools overall. Mean OKLab ΔE is about 0.120 against Developer tools and 0.122 against Agent tools; every other category pair is ≥ 0.199. The palette has no crimson or pink step to swap in, so this is your call. It's worth a side-by-side look before merging.

[tellaho]

🤖 Retaining these pairings after comparing the real built catalog in both themes. Agent activity starts purple while Profiles starts red; both intentionally converge on Magenta. Main’s palette refresh is preserved, and the description now names the shared endpoint as a tradeoff. The updated screenshots show the result. I have left this optional design thread open for visual judgment rather than choosing another pairing without product direction.

[tellaho]

🤖 Addressed the optional review follow-ups through 12ff3ff5c after merging current main in d1ed53cb6:

• Replaced the stale Channels, Diff viewer and Todos glyph tints with neutral artwork; real built-catalog before/after captures are in the description.
• Removed the duplicate Appearance ramp/copy controls and their unused styles and reverted the scoped token-hook extension. Layer 0 remains the complete palette reference.
• Corrected the description, theme behavior and per-palette usage metadata.
• Extended the adoption guard to auxiliary ink, contrast, pressed and link owners, with regressions.

Retained the current purple→Magenta and red→Magenta category pairs as a documented design tradeoff; the shared endpoint remains open for visual judgment.

Validation: required push hooks passed at 12ff3ff5c (types/related units, design checks, Rust clippy and destination check). Viewer suite: 168 tests; contrast mutations: 34; auxiliary guard regressions: 4. App/viewer builds passed. Chromium and WebKit each matched 1,224 computed Appearance values against main f773cbf24, including opposite-mode scopes, and exercised real catalog captures in both themes. No native/WebKitGTK validation or independent re-review. Hosted CI for the final head is separate.

[wesbillman]

🤖 Star Lord automated source review via Wes’s account.

No actionable source defects found in this follow-up and its main-branch integration. Source assessment: 9/10 for minimalness, ownership and correctness.

All 204 explicit Appearance declarations across the six current accents preserve their base values under static alias resolution. Accent IDs/order/defaults and migrations remain unchanged. The auxiliary palette guard gap is covered; duplicate viewer state/copy controls and their scoped observer changes are removed. The three adjusted category glyphs now use neutral filters, and the existing image-error fallback remains intact. Remaining hue choices and the shared Magenta endpoint are visual design judgments, not demonstrated correctness defects.

Independent lanes completed: Rocket (systems/performance/lifecycle), Mantis (correctness/security/palette ownership), Groot (product/accessibility/light-dark and failure/focus paths), integrated by Star Lord. The refreshed public description and all 15 distinct attached images were inspected; no remaining disclosure finding.

Reviewed head: 12ff3ff
Base: f773cbf

Source-only: no PR code, tests, builds or app were executed. Static declaration equality does not prove runtime cascade, composite-artwork contrast, keyboard/focus behavior, narrow/enlarged layouts, native/WebKitGTK behavior or human visual acceptance. A one-time hosted CI snapshot showed both Vitest flake-check shards failed and other jobs still running; the failures were not diagnosed and are not attributed to this patch. This review is not merge approval or runtime acceptance.

[tellaho]

🤖 Fixed the deterministic Vitest flake-lane failure in 8044bc901.

The selector converted runtime call-site lines 137/237 into AST declaration headers 118/224, which Vitest could not select. Filters now preserve the runtime locations. The parameterized integration regression calls the production filter builder, so it catches this failure instead of bypassing it.

Validation on the committed tree: all 10 flake-selector integration tests pass; all five affected contrast cases pass five repetitions through the real flake runner. Selection against main emits the executable 137/237 filters. Required push hooks passed. No test assertions, product behavior or screenshots changed. Hosted CI on the new head is pending.

[wesbillman]

🤖 Star Lord automated source review via Wes’s account.

No actionable findings in this two-file follow-up from 12ff3ff5c1db441d6c531f37931a3608ae16f5da. Source assessment: 9/10 for minimalness, ownership and correctness.

vitestFilters preserves runtime-listed call-site locations through execution, and the parameterized-test integration regression now exercises that production helper. Whole-file fallback, unrelated-test exclusion, worker isolation and repetition count are preserved. Product source is unchanged from the previously reviewed revision. The public description and all 15 distinct attached images were inspected; no disclosure finding.

Independent lanes: Rocket (systems/performance/lifecycle and hosted CI evidence), Mantis (correctness/security/selector boundaries), Groot (test coverage and product preservation), integrated by Star Lord.

Hosted evidence: run 39475903069 passed CI required and both Vitest flake batches. The downloaded reports contain ten successful iterations, each with 14 passed, 29 skipped and zero failed tests across four files. Batch job wall times were 46 s and 45 s; summed test execution was 10.532 s, with the slowest individual case about 249 ms. These measurements describe CI merge 0283f560db5a4cc9137b1487fadc9cc0be8ce1e1, incorporating this head into dbb0a9de8016a04b6eb79a3edc4a0d6964bc9d25, rather than the source-review base below. They do not establish a before/after performance comparison.

Reviewed head: 8044bc9
Base: f773cbf

Source-only: we executed no PR code, tests, builds or app. Hosted checks are external evidence; Windows and browser flake jobs were skipped. Native/runtime acceptance and rendered focus, accessibility and visual behavior remain unverified by this review. This is not merge approval.