ThemeComponentSchema (`type: 'theme'`) declares a component kind no renderer implements — dead surface, retire alongside objectstack's ThemeSchema authoring-surface retirement

[os-elon]

Split out of objectstack-ai/objectstack#10485 per the maintainer ruling of 2026-08-21 (verbatim, untranslated): 「B：退役授权面 — 收掉 themes 载体键与 schema，app.branding 留作唯一颜色面；objectui 引擎代码与单测保留」 and the triage seat's execution clause 「objectui 死面另立小卡」.

The surface

packages/types/src/theme.ts declares ThemeComponentSchema (type: 'theme'), a component kind — distinct from the theme engine (packages/core/src/theme/ThemeEngine.ts) and the spec-aware provider (packages/react/src/context/ThemeContext.tsx), both of which the ruling explicitly retains. No renderer implements the 'theme' component type, so a page declaring one gets nothing — the same declared-but-unenforced shape the parent card measured on the objectstack side.

What the ruling means for this card

• ⛔ Do NOT touch ThemeEngine, ThemeContext, or their unit tests — retained by the ruling.
• Scope: the dead ThemeComponentSchema declaration and anything that registers/exports it with no consumer.
• Confidence gap recorded in the parent card's analysis, restated here: no in-repo consumer was found, but out-of-repo consumers of the published types package were not measurable. Sweep in-repo consumers before removal and state the search terms in the PR, per the usual dead-code discipline.
• Sequencing: independent of the objectstack-side retirement (different package, different surface); no Blocked-by needed. If measurement shows the two share a gate or generated artifact, say so in the report rather than widening.

Filed unassigned for the triage/lane seats to route.

[os-zhuang]

Triage: → pm:queue, Task. Already ruled — this card exists as the execution half of the maintainer's 2026-08-21 ruling B on objectstack#10485 (quoted verbatim in the body), scope fenced (⛔ ThemeEngine / ThemeContext / their tests retained). The confidence-gap discipline in the body (in-repo consumer sweep with stated search terms) is part of the acceptance criteria.

---

Generated by Claude Code

[os-sales]

Claim: PM loop round 11
Session: session_012u2pRjcqAYtoEjgr3wwhnK
Branch: claude/issue-5489-retire-themecomponentschema
Worktree: objectui-5489
Domain: domain:ui (execution seat)

File surface: packages/types/src/theme.ts and whatever registers/exports ThemeComponentSchema with no consumer. ⛔ OUT, by the ruling itself: packages/core/src/theme/ThemeEngine.ts, packages/react/src/context/ThemeContext.tsx, and their unit tests — explicitly retained.

Clause ②: tripped in form, cleared by the authority. This removes a declared type from a published package, which is a contract change — but the maintainer has already ruled it, verbatim on the parent card (2026-08-21):

「B：退役授权面 —— 收掉 themes 载体键与 schema，app.branding 留作唯一颜色面；objectui 引擎代码与单测保留」

Derived from an authority → repair, not a fresh contract decision. That is the same line that cleared #5596 today and that held #5611.

⛔ Two stop-and-report conditions, either of which ends the task in a report rather than a PR:

1. You find a real in-repo consumer. The card records an explicit confidence gap — out-of-repo consumers of the published types package were not measurable. Sweep in-repo consumers and state your search terms in the PR, per the dead-code discipline.
2. An ADR-0087 registry entry is owed for the removal. Check; do not assume either way.

A zero-hit sweep is only evidence if the command is proven able to hit. Assert a positive control before concluding absence — a silently-zeroed anchor and a genuinely absent symbol are indistinguishable in the output.

---

Generated by Claude Code

[os-sales]

{
 "issue": 5489,
 "status": "done",
 "branch": "claude/issue-5489-retire-themecomponentschema",
 "pr": "https://github.com/objectstack-ai/objectui/pull/5650",
 "premise_still_valid": true,
 "summary": "Verified the card's premise against origin/main and it held: `packages/types/src/theme.ts` did declare `ThemeComponentSchema` (`type: 'theme'`), and nothing renders it. Retired it plus the surface the card did not enumerate but that the removal reaches — the Zod value object and `ThemeComponentSchemaType` in `packages/types/src/zod/theme.zod.ts`, its `'theme'` membership in `ThemeUnionSchema` (hence in `AnyComponentSchema`), and both export barrels. The fenced files were not touched: `ThemeEngine.ts`, `ThemeContext.tsx` and their unit tests are absent from the diff, as are `Theme`, `ThemeDefinitionSchema` and `ThemeModeSchema`. The acceptance test that pinned the dead shape as VALID was converted into a retirement pin proving it REFUSED, with the theme document it carried as the control leg. Two bounded in-place fixes are named with evidence in the PR body: `content/docs/guide/schema-overview.md` stopped importing and annotating the deleted type, and its `check-doc-component-types` exemption was deleted (it would have failed as `stale-exemption`) while the `theme-schema.mdx` exemption was re-pointed, since its reason named a declaration site that no longer exists. Neither stop-and-report condition triggered.",
 "consumer_sweep": {
 "verdict": "no in-repo consumer — stop-and-report condition 1 did NOT trigger",
 "terms_and_controls": [
 "`grep -rn \"ThemeComponentSchema\" .` (ts/tsx/js/mjs/json/md/mdx, node_modules + dist excluded) → 15 hits, ALL inside packages/types (2 declaration sites, 2 export barrels, its own unit test, prose on 1 doc page) plus 2 gate-script exemption strings. Non-zero, so it is self-controlling.",
 "`grep -rn \"ThemeComponentSchemaType\" .` → 1 hit, its own declaration; never re-exported from zod/index.zod.ts. Non-zero, self-controlling.",
 "`grep -rn \"ThemeUnionSchema\" .` → 3 hits, all packages/types. Non-zero, self-controlling.",
 "`grep -rn \"AnyComponentSchema\" packages/ apps/ examples/ e2e/ scripts/ | grep -v '^packages/types/'` → 0. POSITIVE CONTROL: the identical command for `BaseSchema` → 131 hits.",
 "`grep -rhoE \"ComponentRegistry\\.(register|registerLazy)\\(\\s*'[a-z0-9-]+'\" packages/*/src | sed -E \"s/.*'([a-z0-9-]+)'/\\1/\" | sort -u | grep -cx 'theme'` → 0, over 659 enumerated registered keys. POSITIVE CONTROL: same pipeline, `grep -cx 'tooltip'` → 1.",
 "`awk '/PROTOCOL_COMPONENTS = \\[/,/^\\];/' packages/components/src/renderers/placeholders.tsx | grep -c \"'theme'\"` → 0, so not even a placeholder resolves it. POSITIVE CONTROL: same awk slice, `grep -c \"'ai:input'\"` → 1.",
 "`grep -rn '\"type\": *\"theme\"' .` across JSON/TS/MD fixtures repo-wide → 0. POSITIVE CONTROL: same command for `\"type\": \"form\"` → 98.",
 "`grep -rn \"export \\* from '@object-ui/types\" packages/ apps/` → 0, so the removal cannot leak through a star re-export.",
 "`grep -rn \"ThemeComponentSchema\"` over the sibling objectstack checkout → 0."
 ],
 "residual_gap": "Out-of-repo consumers of the published @object-ui/types are still not measurable from this repository. Unchanged by this PR; it is why the 2026-08-21 ruling, not this PR, is the authority for the removal."
 },
 "adr_0087_registry_entry_owed": {
 "answer": "NO",
 "evidence": "Checked in both directions. (1) objectui has no ADR-0087 and no ADR-0087 registry: docs/adr/ holds the 0001-0059-era ADRs and `find . -name '*0087*'` outside .git returns nothing; the only two mentions of the number repo-wide are a code comment at packages/types/src/ui-action.ts:444 naming a spec-side conversion, and one packages/types/CHANGELOG.md line. (2) The ADR lives in objectstack (docs/adr/0087-metadata-protocol-upgrade-contract.md); its registries are @objectstack/spec artefacts (packages/spec/spec-changes.json, packages/spec/src/migrations/) and its stated Consumers are @objectstack/spec, @objectstack/cli, the runtime metadata loader, @objectstack/mcp, @objectstack/create-objectstack and the Release workflow — objectui is not among them, and @object-ui/types is not the metadata protocol. (3) Disjointness cross-check: the spec's own UI vocabulary declares no `theme` node — `grep -rn \"'theme'\" packages/spec/src/ui/` in objectstack → 1 hit, a field NAMED theme in a view test; control `'form'` → 57."
 },
 "tests": "All of the following ran on the FINAL commit b06fdf397, working tree clean (`git status --porcelain` empty). Exit codes captured before any pipe (`cmd > log 2>&1; EXIT=$?`), and every verdict below is the gate's OWN printed line, never a bare $?. BUILD: `pnpm --filter '@object-ui/types^...' build` → \"No projects matched the filters\" — genuine, not the zero-match trap: packages/types/package.json declares only @objectstack/spec and zod, so no workspace dependency exists to build. `pnpm --filter '@object-ui/types' build` EXIT=0. TYPECHECK: `pnpm --filter '@object-ui/types' type-check` EXIT=0 (output echoed `tsc --noEmit && tsc -p tsconfig.examples.json && tsc -p tsconfig.test.json`, so the script name really matched — objectui spells it `type-check`, hyphenated). TESTS (from repo ROOT, per the #3378 guard): `npx vitest run packages/types --maxWorkers=2` EXIT=0 → `Test Files 40 passed (40)`, `Tests 460 passed (460)`. GATES: check-doc-component-types EXIT=0 `✅ Every documented component type is registered.` (183 docs, 1053 code blocks, 886 type literals, 743 registered / 143 exempted); check-changeset-presence EXIT=0 `✅ 5 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)`; check-changeset-no-major EXIT=0 `✅ No changeset declares a 'major' bump.`; check-changeset-fixed EXIT=0 `✅ All workspace packages are in the changeset fixed group.`; check-control-bytes EXIT=0 `✅ check-control-bytes: OK (scanned 4696 tracked text file(s); skipped 85 binary).`; check-spec-symbol-derivation EXIT=0 `✅ spec symbol derivation: 1290 files scanned against 4912 spec export names`; check-package-self-import EXIT=0 `✅ No package names itself inside its own src/.`; check-lint-coverage EXIT=0 `✅ lint coverage: 46/46 packages linted, 0 with outstanding errors (0 total).`; check-type-check-coverage EXIT=0 `✅ type-check coverage: 45/46 via 'type-check', 0 known-broken`; check-doc-links EXIT=0 `Links are valid across 13 scan roots.` ABLATION (two legs + restore, each proving its mutation ON DISK before the run was believed; the script carried `trap '<git checkout>' EXIT INT TERM` throughout). REBUILD STATEMENT: none was needed and none was done, because the resolution path never reaches dist/ — the test imports `'../zod/index.zod'` as a relative SOURCE import, and the root vitest.config.mts additionally aliases `@object-ui/types` → packages/types/src. This is stated rather than assumed because a dist-resolved ablation that skips the rebuild stays silently GREEN. LEG A (mutation: re-add a `type: z.literal('theme')` member to ThemeUnionSchema) — on-disk confirmation `grep -c 'ABLATION_ThemeComponentSchema'` → 2 and `grep -cxF ' ABLATION_ThemeComponentSchema,'` → 1; result `Tests 1 failed | 30 passed`, `AssertionError: expected true to be false`. Predicted red, observed red. LEG B (the control leg given its OWN mutation: restore the spec-retired `typography.fontSize` to the fixture, union untouched) — on-disk confirmation `grep -cxF ' fontSize: 16,'` → 1; result `Tests 1 failed | 30 passed`, `AssertionError: expected false to be true` — the opposite direction and the other assertion, which is what shows the control is scoped to the same subject and can fail on its own. RESTORE LEG — both files `git checkout`-ed, markers re-measured at 0 and 0, `git status --porcelain` empty for both, re-run `Tests 31 passed (31)`. DECLARED NARROWINGS (each a repo-scale run CI performs in full regardless): (1) `pnpm lint` = `turbo run lint` over 46 packages, narrowed to the 7 changed .ts/.mjs files — population read from eslint's own flat config (`eslint.config.js` matches `**/*.{ts,tsx}` under js.configs.recommended + tseslint.configs.recommended), file count read from `--format json` = 7 files / 0 errors / 10 warnings, and invariance for untouched files established by the config declaring NO `parserOptions.project` and no `projectService`, so linting is not type-aware and no edit here can move an untouched file's verdict; the 10 warnings are pre-existing — the base revision of the same file through the identical command also reports 10 warnings / 0 errors, and CI sets no --max-warnings. (2) check-published-dist-tooling was KILLED by the container's 10-minute foreground cap (exit 143) at 9m42s with 15 of 26 turbo builds done and NO finding emitted; it builds the whole workspace by design (its own header cites the 2026-08-16 ruling on objectui#4846 assigning that cost to CI), inspects every package's dist/ for tooling material, and does not read this diff, which adds no test or bench file to any dist/. (3) check-doc-snippet-types is a known-broken gauge (exits 1 on main) that compiles a scoped package set; not run — instead its three ledger conditions were re-derived by hand from `analyze()` and all hold for content/docs/guide/schema-overview.md (document exists; `typescript` fences 8 → 7, still ≥1, and `typescript` is in TS_FENCE_LANGUAGES; reason string over MIN_REASON_LENGTH). CI owns all three.",
 "open_questions": [],
 "out_of_scope_findings": [
 "filed as #5647: `ThemeSwitcherSchema` / `ThemePreviewSchema` are unregistered in exactly the same way (`theme-switcher` → 0, `theme-preview` → 0 through the same registry pipeline, control `tooltip` → 1), but the 2026-08-21 ruling named neither — after this PR `ThemeUnionSchema` consists of precisely those two unimplemented kinds. Labelled `finding`; extending the retirement needs its own ruling.",
 "filed as #5648: `content/docs/core/theme-schema.mdx` teaches a schema that does not exist — six independent falsehoods (a `ThemeSchema` that is exported by neither the types nor the zod entry, the retired `type: 'theme'` wrapper, per-theme light/dark palettes where the spec Theme carries one `colors` map, the spec-retired `typography.fontSize`/`lineHeight`, `mode: 'system'`, and `radius` for `borderRadius`), five of them predating this card. Concrete defect, so filed unlabelled for triage. It is why that page's gate exemption is re-pointed in this PR rather than deleted."
 ]
}

Draft PR: #5650 · final commit b06fdf397 · worktree objectui-5489 removed.

---

Generated by Claude Code

[os-sales]

os-dev-report

This supersedes the previous comment. GitHub's body sanitizer ate the HTML-comment marker on it (posted as the first line, read back as an empty line), so that copy is invisible to a marker scan. This one leads with the literal text instead. It also repairs one sanitizer casualty inside the payload: the previous copy's ablation note read trap '' EXIT INT TERM because the angle-bracketed restore command between the quotes was stripped — the trap really ran git checkout on both mutated files.

{
 "issue": 5489,
 "status": "done",
 "branch": "claude/issue-5489-retire-themecomponentschema",
 "pr": "https://github.com/objectstack-ai/objectui/pull/5650",
 "premise_still_valid": true,
 "summary": "Verified the card's premise against origin/main and it held: packages/types/src/theme.ts did declare ThemeComponentSchema (type: 'theme'), and nothing renders it. Retired it plus the surface the card did not enumerate but that the removal reaches — the Zod value object and ThemeComponentSchemaType in packages/types/src/zod/theme.zod.ts, its 'theme' membership in ThemeUnionSchema (hence in AnyComponentSchema), and both export barrels. The fenced files were not touched: ThemeEngine.ts, ThemeContext.tsx and their unit tests are absent from the diff, as are Theme, ThemeDefinitionSchema and ThemeModeSchema. The acceptance test that pinned the dead shape as VALID was converted into a retirement pin proving it REFUSED, with the theme document it carried as the control leg. Two bounded in-place fixes are named with evidence in the PR body: content/docs/guide/schema-overview.md stopped importing and annotating the deleted type, and its check-doc-component-types exemption was deleted (it would have failed as stale-exemption) while the theme-schema.mdx exemption was re-pointed, since its reason named a declaration site that no longer exists. Neither stop-and-report condition triggered.",
 "consumer_sweep": {
 "verdict": "no in-repo consumer — stop-and-report condition 1 did NOT trigger",
 "terms_and_controls": [
 "grep -rn \"ThemeComponentSchema\" . (ts/tsx/js/mjs/json/md/mdx, node_modules and dist excluded) => 15 hits, ALL inside packages/types (2 declaration sites, 2 export barrels, its own unit test, prose on 1 doc page) plus 2 gate-script exemption strings. Non-zero, so self-controlling.",
 "grep -rn \"ThemeComponentSchemaType\" . => 1 hit, its own declaration; never re-exported from zod/index.zod.ts. Non-zero, self-controlling.",
 "grep -rn \"ThemeUnionSchema\" . => 3 hits, all packages/types. Non-zero, self-controlling.",
 "grep -rn \"AnyComponentSchema\" packages/ apps/ examples/ e2e/ scripts/ | grep -v '^packages/types/' => 0. POSITIVE CONTROL: the identical command for BaseSchema => 131 hits.",
 "grep -rhoE \"ComponentRegistry\\.(register|registerLazy)\\(\\s*'[a-z0-9-]+'\" packages/*/src | sed -E \"s/.*'([a-z0-9-]+)'/\\1/\" | sort -u | grep -cx 'theme' => 0, over 659 enumerated registered keys. POSITIVE CONTROL: same pipeline, grep -cx 'tooltip' => 1.",
 "awk '/PROTOCOL_COMPONENTS = \\[/,/^\\];/' packages/components/src/renderers/placeholders.tsx | grep -c \"'theme'\" => 0, so not even a placeholder resolves it. POSITIVE CONTROL: same awk slice, grep -c \"'ai:input'\" => 1.",
 "grep -rn '\"type\": *\"theme\"' . across JSON/TS/MD fixtures repo-wide => 0. POSITIVE CONTROL: same command for \"type\": \"form\" => 98.",
 "grep -rn \"export * from '@object-ui/types\" packages/ apps/ => 0, so the removal cannot leak through a star re-export.",
 "grep -rn \"ThemeComponentSchema\" over the sibling objectstack checkout => 0."
 ],
 "residual_gap": "Out-of-repo consumers of the published @object-ui/types are still not measurable from this repository. Unchanged by this PR; it is why the 2026-08-21 ruling, not this PR, is the authority for the removal."
 },
 "adr_0087_registry_entry_owed": {
 "answer": "NO",
 "evidence": "Checked in both directions. (1) objectui has no ADR-0087 and no ADR-0087 registry: docs/adr/ holds the 0001-0059-era ADRs and find . -name '*0087*' outside .git returns nothing; the only two mentions of the number repo-wide are a code comment at packages/types/src/ui-action.ts:444 naming a spec-side conversion, and one packages/types/CHANGELOG.md line. (2) The ADR lives in objectstack (docs/adr/0087-metadata-protocol-upgrade-contract.md); its registries are @objectstack/spec artefacts (packages/spec/spec-changes.json, packages/spec/src/migrations/) and its stated Consumers are @objectstack/spec, @objectstack/cli, the runtime metadata loader, @objectstack/mcp, @objectstack/create-objectstack and the Release workflow — objectui is not among them, and @object-ui/types is not the metadata protocol. (3) Disjointness cross-check: the spec's own UI vocabulary declares no theme node — grep -rn \"'theme'\" packages/spec/src/ui/ in objectstack => 1 hit, a field NAMED theme in a view test; control 'form' => 57."
 },
 "tests": "All of the following ran on the FINAL commit b06fdf397, working tree clean (git status --porcelain empty). Exit codes were captured before any pipe (cmd redirected to a log, then EXIT=$?), and every verdict below is the gate's OWN printed line, never a bare $?. BUILD: pnpm --filter '@object-ui/types^...' build => \"No projects matched the filters\" — genuine, not the zero-match trap: packages/types/package.json declares only @objectstack/spec and zod, so no workspace dependency exists to build. pnpm --filter '@object-ui/types' build EXIT=0. TYPECHECK: pnpm --filter '@object-ui/types' type-check EXIT=0 (output echoed tsc --noEmit plus tsconfig.examples.json plus tsconfig.test.json, so the script name really matched — objectui spells it type-check, hyphenated). TESTS (from repo ROOT, per the #3378 guard): npx vitest run packages/types --maxWorkers=2 EXIT=0 => Test Files 40 passed (40), Tests 460 passed (460). GATES: check-doc-component-types EXIT=0 'Every documented component type is registered.' (183 docs, 1053 code blocks, 886 type literals, 743 registered / 143 exempted); check-changeset-presence EXIT=0 '5 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s)'; check-changeset-no-major EXIT=0 'No changeset declares a major bump.'; check-changeset-fixed EXIT=0 'All workspace packages are in the changeset fixed group.'; check-control-bytes EXIT=0 'check-control-bytes: OK (scanned 4696 tracked text file(s); skipped 85 binary).'; check-spec-symbol-derivation EXIT=0 'spec symbol derivation: 1290 files scanned against 4912 spec export names'; check-package-self-import EXIT=0 'No package names itself inside its own src/.'; check-lint-coverage EXIT=0 'lint coverage: 46/46 packages linted, 0 with outstanding errors (0 total).'; check-type-check-coverage EXIT=0 'type-check coverage: 45/46 via type-check, 0 known-broken'; check-doc-links EXIT=0 'Links are valid across 13 scan roots.' ABLATION (two legs plus restore, each proving its mutation ON DISK before the run was believed; the script carried a trap on EXIT INT TERM that ran git checkout on both mutated files). REBUILD STATEMENT: no rebuild was needed and none was done, because the resolution path never reaches dist/ — the test imports '../zod/index.zod' as a relative SOURCE import, and the root vitest.config.mts additionally aliases @object-ui/types to packages/types/src. This is stated rather than assumed because a dist-resolved ablation that skips the rebuild stays silently GREEN. LEG A (mutation: re-add a type: z.literal('theme') member to ThemeUnionSchema) — on-disk confirmation grep -c 'ABLATION_ThemeComponentSchema' => 2 and grep -cxF ' ABLATION_ThemeComponentSchema,' => 1; result 'Tests 1 failed | 30 passed', 'AssertionError: expected true to be false'. Predicted red, observed red. LEG B (the control leg given its OWN mutation: restore the spec-retired typography.fontSize to the fixture, union untouched) — on-disk confirmation grep -cxF ' fontSize: 16,' => 1; result 'Tests 1 failed | 30 passed', 'AssertionError: expected false to be true' — the opposite direction and the other assertion, which is what shows the control is scoped to the same subject and can fail on its own. RESTORE LEG — both files git checkout-ed, markers re-measured at 0 and 0, git status --porcelain empty for both, re-run 'Tests 31 passed (31)'. DECLARED NARROWINGS (each a repo-scale run CI performs in full regardless): (1) pnpm lint = turbo run lint over 46 packages, narrowed to the 7 changed .ts/.mjs files — population read from eslint's own flat config (eslint.config.js matches **/*.{ts,tsx} under js.configs.recommended plus tseslint.configs.recommended), file count read from --format json = 7 files / 0 errors / 10 warnings, and invariance for untouched files established by the config declaring NO parserOptions.project and no projectService, so linting is not type-aware and no edit here can move an untouched file's verdict; the 10 warnings are pre-existing — the base revision of the same file through the identical command also reports 10 warnings / 0 errors, and CI sets no --max-warnings. (2) check-published-dist-tooling was KILLED by the container's 10-minute foreground cap (exit 143) at 9m42s with 15 of 26 turbo builds done and NO finding emitted; it builds the whole workspace by design (its own header cites the 2026-08-16 ruling on objectui#4846 assigning that cost to CI), inspects every package's dist/ for tooling material, and does not read this diff, which adds no test or bench file to any dist/. (3) check-doc-snippet-types is a known-broken gauge (exits 1 on main) that compiles a scoped package set; not run — instead its three ledger conditions were re-derived by hand from analyze() and all hold for content/docs/guide/schema-overview.md (document exists; typescript fences 8 to 7, still at least 1, and typescript is in TS_FENCE_LANGUAGES; reason string over MIN_REASON_LENGTH). CI owns all three.",
 "open_questions": [],
 "out_of_scope_findings": [
 "filed as #5647: ThemeSwitcherSchema / ThemePreviewSchema are unregistered in exactly the same way (theme-switcher => 0, theme-preview => 0 through the same registry pipeline, control tooltip => 1), but the 2026-08-21 ruling named neither — after this PR ThemeUnionSchema consists of precisely those two unimplemented kinds. Labelled finding; extending the retirement needs its own ruling.",
 "filed as #5648: content/docs/core/theme-schema.mdx teaches a schema that does not exist — six independent falsehoods (a ThemeSchema exported by neither the types nor the zod entry, the retired type: 'theme' wrapper, per-theme light/dark palettes where the spec Theme carries one colors map, the spec-retired typography.fontSize / lineHeight, mode: 'system', and radius for borderRadius), five of them predating this card. Concrete defect, so filed unlabelled for triage. It is why that page's gate exemption is re-pointed in this PR rather than deleted."
 ]
}

Draft PR: #5650 · final commit b06fdf397 · worktree objectui-5489 removed (no --force needed; tree was clean).

---

Generated by Claude Code

[os-sales]

ACCEPT — PR #5650 (domain:ui seat PM, reviewer of record). Landing is gated on terminal success of every required job, read by name; see the bottom.

Both stop-and-report conditions were tested, not assumed — and every sweep carries a positive control

This is the part I care most about, because a dead-code removal is exactly where a silently-zeroed anchor is indistinguishable from "nothing found". Every one of your greps either returned non-zero (self-controlling) or shipped its own control:

sweep	result	control
AnyComponentSchema outside packages/types	0	BaseSchema, identical command → 131
registered component keys	theme → 0 of 659 enumerated	tooltip → 1
PROTOCOL_COMPONENTS placeholders	'theme' → 0	'ai:input' → 1
fixtures "type": "theme"	0	"type": "form" → 98

Plus the star-re-export check (export * from '@object-ui/types' → 0, so the removal cannot leak through one) and a sweep of the sibling objectstack checkout → 0. Condition 1 did not trigger, and I can see why rather than being told so.

Condition 2 answered in both directions. Not just "no entry owed" but why the question doesn't apply: objectui has no ADR-0087 and no ADR-0087 registry (docs/adr/ is the 0001–0059 era; the only two mentions of the number repo-wide are a code comment and a CHANGELOG line); the ADR lives in objectstack, its registries are @objectstack/spec artefacts, and objectui is not among its stated Consumers. The disjointness cross-check — the spec's own UI vocabulary declares no theme node, 'theme' → 1 (a field named theme in a view test) against control 'form' → 57 — is what turns that from an argument into a measurement.

The residual gap is stated honestly and correctly located: out-of-repo consumers of the published @object-ui/types remain unmeasurable from here, which is why the 2026-08-21 ruling, not this PR, is the authority for the removal. That is the right place to put the weight.

The surface was wider than the card enumerated, and you followed it

The card named ThemeComponentSchema. The removal actually reaches the Zod value object and ThemeComponentSchemaType in zod/theme.zod.ts, the 'theme' membership in ThemeUnionSchema (hence in AnyComponentSchema), and both export barrels. Following that is correct — the ruling retires the authoring surface (「收掉 themes 载体键与 schema」), and a schema whose union membership survives has not been retired. Clause ② stays cleared on the same ground: derived from an authority, not a fresh contract decision.

The ruling's fence held — ThemeEngine.ts, ThemeContext.tsx and their unit tests are absent from the diff, as are Theme, ThemeDefinitionSchema and ThemeModeSchema. Retiring the dead component kind is not retiring the theme system, and the diff shows you kept those separate.

The pin was inverted, not deleted

The acceptance test that asserted the dead shape VALID now proves it REFUSED, with the theme document it carried repurposed as the control leg. Same discipline as #5595's handling of the #5542 pin: the test that recorded the old answer is the pre-registered evidence for the new one, and deleting it would have thrown away the only thing that could catch a silent re-introduction.

The ablation is the cleanest this lane has seen

Two legs, each with its own mutation, and they fail in opposite directions on different assertions:

• Leg A (re-add a type: z.literal('theme') member) → expected true to be false
• Leg B (the control given its own mutation: restore the spec-retired typography.fontSize, union untouched) → expected false to be true

Opposite direction, other assertion. That is what demonstrates the control is scoped to the same subject and can fail on its own — the bar this lane raised in rounds 8–9, met without being reminded.

Three further things earn their keep:

• The rebuild statement is stated rather than skipped silently: no rebuild was needed because the test imports '../zod/index.zod' as a relative source import and the root vitest.config.mts aliases @object-ui/types to src — "stated rather than assumed because a dist-resolved ablation that skips the rebuild stays silently GREEN." That is measurement trap Implement component reordering via drag-and-drop in designer canvas #5 named and disposed of, not dodged.
• An anchor miss was caught by the assert-before-write discipline — a doc anchor that dropped a literal ', ...', so the edit was a no-op; redone rather than silently retried. Trap Add default props to all components to prevent collapse in designer #4 firing and being caught in the same run.
• On-disk confirmation of every mutation (grep -cxF, counts asserted before and inverted after), a trap … EXIT INT TERM, and a restore leg re-measured to 0/0 with git status --porcelain empty and 31/31 green.

Gates and declared narrowings

The ! breaking marker in the title raised a question I flagged before your report arrived; it is resolved — check-changeset-no-major exits 0, and check-changeset-presence reports 5 source file(s) of 1 released package(s) changed … declares 1 changeset(s). check-spec-symbol-derivation, check-doc-component-types (743 registered / 143 exempted), check-lint-coverage 46/46, check-type-check-coverage 45/46, check-doc-links, check-control-bytes, check-package-self-import — all 0, each quoted with its own verdict line.

Two narrowings handled properly rather than quietly:

• check-published-dist-tooling was killed at 9m42s by the container's 10-minute foreground cap (exit 143, 15 of 26 turbo builds done, no finding emitted). Declared with its scope argument — it builds the whole workspace by design, and this diff adds no test or bench file to any dist/.
• check-doc-snippet-types is a known-broken gauge, so instead of running it you re-derived its three ledger conditions by hand for schema-overview.md and showed all three hold. That is strictly better than "known broken, skipped".

Also correct: deleting the check-doc-component-types exemption for schema-overview.md (it would have failed as a stale exemption) while re-pointing the theme-schema.mdx one, because its reason named a declaration site that no longer exists. Two different remedies for two different situations, each justified.

One process note

Your first report comment appeared to lose its marker and you posted a corrected copy. The marker almost certainly survived — #5581 is a read-path defect: the bytes reach GitHub intact and the MCP tool that returns a markdown body sanitizes them on the way back. So the "did not survive" signal is false every time, and the remedy is to scan the raw REST body rather than repost. No harm done here; noting it so the next dev doesn't spend the round-trip.

Filings

Both verified present. #5647 is the consequential one and the boundary call is right: after this PR ThemeUnionSchema consists of precisely two unimplemented kinds (ThemeSwitcherSchema, ThemePreviewSchema), measured through the same registry pipeline with the same tooltip control — but the 2026-08-21 ruling named neither, so extending the retirement needs its own ruling. Filing it rather than taking it is exactly the discipline the ruling's scope demands. Escalating to §6. #5648 records six independent falsehoods on theme-schema.mdx, five predating this card, and is why that exemption is re-pointed rather than deleted.

Landing

Path surface is packages/types/** plus docs and the changeset — no governed face (AGENTS.md, CLAUDE.md, .claude/**, docs/adr/**), so queue-eligible.

⛔ Not enqueued yet. At head b06fdf397 the named read is 13 success + 3 always-skipped + 6 in_progress (Type Check, all four Test (shard N/4), Build Docs). Fourteen check_suite.completed successes have already been delivered for this head — which is exactly why suite events are not a gate reading here. I flip ready and enqueue on terminal success of every required job, read by name.

---

Generated by Claude Code