[finding] `ComponentMeta` has two published declarations (`base.ts` + `plugin-scope.ts`), already diverged by `tags`/`description`, and the `PluginComponentMeta` alias has zero consumers

[claude]

Noticed while retiring PluginComponentInput (#5674). Filed rather than fixed: it is out of that card's fence, and it is not the same shape as PluginComponentInput despite sitting on the adjacent line.

Measured

packages/types/src/index.ts publishes ComponentMeta twice, from two different declarations:

line	spelling	declaration
index.ts:99	ComponentMeta	packages/types/src/base.ts:511
index.ts:921	ComponentMeta as PluginComponentMeta	packages/types/src/plugin-scope.ts:149

These are structural copies, not an alias pair — which is exactly what makes this different from #5674. PluginComponentInput became a pure alias once #5671 converged its declaration; PluginComponentMeta still names a genuinely separate interface with a different key set.

The divergence is already live

Top-level keys, both declarations read off origin/main:

• base.ts (11): label icon category inputs defaultProps examples isContainer resizable resizeConstraints tags description
• plugin-scope.ts (9): the same nine, minus tags and description

resizeConstraints' six members are identical in both. So the delta is exactly two keys, and it is the same defect class #4972 recorded for ComponentInput — where the delta was min/max/step/placeholder — one release before it bit.

Why it is worth recording

#4580's ruling was written for this shape, and the wording is general: "a structural copy would reproduce the defect the moment either side moved." Either side has moved: tags and description exist on the published ComponentMeta and not on the plugin-facing twin, so a plugin author typing against the plugin-scope declaration cannot write two keys the main surface advertises.

There is already a test paying for the duplication. packages/types/src/__tests__/default-children-retired-contract-twins.test.ts asserts the same contract twice — once against ComponentMeta from base.ts (line 116) and once against ComponentMeta from plugin-scope.ts (line 131) — and its own header calls the second one "its plugin-facing twin". That file is the running cost of the copy.

Note this is not covered by #4972, which is closed and was scoped to ComponentInput's three copies plus WidgetInput. ComponentMeta is named nowhere in it.

Zero consumers at the published alias name

PluginComponentMeta has one occurrence repo-wide excluding node_modules//dist/: its own export line at index.ts:921. Nothing imports it.

Search scope: every root in the repo (packages/ apps/ content/ docs/ skills/ examples/ e2e/ scripts/ eslint-rules/ public/ .changeset/ and the root *.md), plus the sibling objectstack framework checkout. Control, searched identically so a broken search could not read as a clean one: AppMetadataPlugin, a neighbour in the same export type { ... } block, returns 12 hits including a real cross-package import at packages/core/src/registry/PluginSystem.ts:10; in the framework checkout the control SchemaRenderer returns 30 files and @object-ui/types 23. The apparatus finds names that are there.

The in-repo half is measurable; an importer on npm is not. That distinction is what #5674 was filed about and it applies here unchanged.

Possible directions

• (a) Converge, per finding(types): two competing SchemaNode declarations — core's interface vs types' union #4580: plugin-scope.ts's ComponentMeta becomes export type { ComponentMeta } from './base.js', and tags/description reach the plugin surface. Same move ComponentInput: re-export the one declaration instead of restating it three times #5671 made for ComponentInput, same package, and its PR measured the emitted .d.ts both ways.
• (b) Converge and retire PluginComponentMeta, which after (a) is a second published name for one type — i.e. the state [finding] PluginComponentInput is a published name with zero consumers, and after #4972 it is a pure alias of ComponentInput #5674 is retiring. Sequencing it as one job avoids deprecating a name that is about to change meaning.
• (c) Keep both and add a structural gate (same-named interfaces must have equal key sets), turning silent drift into a loud failure. This is [finding] ComponentInput 有三份结构副本(types 两份 + core 一份)外加一个窄孪生 WidgetInput,而 #4580 的裁定说的是「re-export 而不是 restate」—— 副本之间今天已经有一处分歧 #4972's option (c), which was not taken there.

Not urgent: no user hits this today, the same way #4972's divergence was dormant until it wasn't. Recording it while the measurement is fresh.

Related: #4580 / #4548 (the ruling and the 19/35 probe), #4972 (the ComponentInput census), #5671 (the convergence PR), #5674 (the alias retirement this was found under).

Filed unassigned.

---

Generated by Claude Code

---

Generated by Claude Code

[os-zhuang]

Triage first-touch (Routine seat, hourly round): promoted finding → pm:queue + domain:ui, type Task.

Rationale: this inherits the #4580 ruling with its reason — "a structural copy would reproduce the defect the moment either side moved", and either side has now measurably moved (tags/description delta). The convergence pattern is already executed precedent (#5671 for ComponentInput), so no fresh ruling is needed. Suggested scope is the card's direction (b) — converge plugin-scope.ts onto base.ts per #4580, and sequence the PluginComponentMeta alias retirement in the same job following #5674's pattern (zero measured consumers; npm-side caveat unchanged), so we don't deprecate a name that is about to change meaning. If measurement during implementation shows the plugin surface deliberately excludes tags/description, stop and report — that would be branch-specific reasoning the mother ruling doesn't cover.

---

Generated by Claude Code

[yinlianghui]

Claim: PM loop round 38 (refill slot)
Session: session_01CSoz9uGhaaSgiq3hshtN7L
Branch: claude/issue-5893-componentmeta-convergence
Worktree: objectui-issue-5893
Domain: domain:ui
File surface: packages/types/src/base.ts, packages/types/src/plugin-scope.ts, their tests + a changeset. Consumers are read-only unless the convergence forces a call-site change — say so if it does.
Container & model: M, mode:subagent, model: opus
Clause-②: yes — two published declarations converge and a published alias is retired. Declared; Opus under this seat's recorded Fable-quota exemption.
Serial constraints: ⚠️ packages/types has a sibling PR in flight — #5903's PR #6053 (objectql.ts + zod/objectql.zod.ts, awaiting gate read). Different files, so no write collision, but ⛔ if you need to touch objectql.ts for any reason, stop and report. Other siblings: #5725 (app-shell/src/views/metadata-admin/previews/), #5676 (data-objectstack), #5883 (apps/console/src/components/FormPage.tsx), #5874+#5875 (plugin-kanban, plugin-detail ×2, app-shell/src/utils/ ×2). ⚠️ #5874 is live in app-shell/src/utils/ — you are in packages/types; ⛔ stay out.

裁决 — this inherits an existing ruling; ⛔ no fresh ruling is needed or to be sought

Triage (Routine seat, hourly round, 2026-08-24) promoted finding → pm:queue / Task, verbatim rationale:

this inherits the #4580 ruling with its reason — "a structural copy would reproduce the defect the moment either side moved" — and either side has now measurably moved (tags/description delta). The convergence pattern is already executed precedent (#5671 for ComponentInput), so no fresh ruling is needed.

⭐ Note the shape of that inheritance: #4580's ruling was not "don't duplicate" as a style preference, it was a prediction — a structural copy will reproduce the defect the moment either side moves. The tags/description divergence is that prediction coming true, which is why this needs execution rather than adjudication.

Ruled scope — the card's direction (b), two parts in one job:

1. Converge plugin-scope.ts onto base.ts per finding(types): two competing SchemaNode declarations — core's interface vs types' union #4580, following ComponentInput: re-export the one declaration instead of restating it three times #5671's executed pattern for ComponentInput.
2. Sequence the PluginComponentMeta alias retirement in the same job, following [finding] PluginComponentInput is a published name with zero consumers, and after #4972 it is a pure alias of ComponentInput #5674's pattern (zero measured consumers; npm-side caveat unchanged) — "so we don't deprecate a name that is about to change meaning."

⚠️ Part 2's ordering is the point: retiring the alias after the convergence means it is retired at its final meaning. Doing it first would deprecate one thing and land another.

⛔ STOP condition, carried verbatim

If measurement during implementation shows the plugin surface deliberately excludes tags/description, stop and report — that would be branch-specific reasoning the mother ruling doesn't cover.

⚠️ Take this seriously and look for it before converging: a plugin-facing metadata shape has a defensible reason to omit free-text fields the host renders. If you find such a reason recorded anywhere (ADR, docblock, a ruling on #4580's thread, the plugin-scope module's own header), ⛔ stop and report a fork; do not converge and do not decide it yourself.

PM mechanism assumptions — measure, do not assume

1. The card is from 2026-08-23 and packages/types has moved five times today — feat(types)!: retire the ActionCondition branch shape from ActionSchema.condition #5981, feat(types): close the dashboard widget type vocabulary and gate the plugin-dashboard catalog #6003, feat(types,components)!: ui:icon names its glyph with icon, not the identity key name #6012, fix(types): widen seven zod mirrors the renderer already implements (#5927 group A) #6032 (17 zod mirrors refuse spellings their TypeScript declarations declare — each needs a widen-the-mirror / narrow-the-declaration ruling #5927 Group A), and feat(types,plugin-gantt): declare the ten gantt keys ObjectGantt reads through a cast #6053 is pending. ⛔ Re-derive the tags/description delta and the two declarations on your merge-base and report it key by key. If the delta has grown or closed since, say so.
2. ⚠️ "Zero consumers" is a claim about a published name and needs the same care [finding] PluginComponentInput is a published name with zero consumers, and after #4972 it is a pure alias of ComponentInput #5674 took. Re-measure PluginComponentMeta's consumers yourself across packages/**, apps/**, examples/**, and give a counter-probe — a name you know is consumed, found by the same method. ⛔ A zero without a control is not a measurement. ⚠️ And in-repo zero ≠ npm zero: the retirement's external caveat is unchanged and must be restated in the changeset, not quietly dropped.
3. ⚠️ base.ts carries BaseSchema and its index signature, which is the ceiling on what any declaration can reject (finding(types): BaseSchema's [key: string]: any leaves every component schema open, so a "declare the surface" fix can never reject a misspelled TOP-LEVEL key #5155) and the reason the zod-mirror-parity guard has a measured blind spot on this exact axis (finding: the #5684 zod-mirror-parity guard is BLIND to a key declared on the TS side but absent from the mirror — measured on ObjectGanttSchema, and it is the guard #5927 just used as authoritative #6058, filed today). ⛔ Do not claim the convergence buys rejection of anything unless you have measured it — and if ComponentMeta has a zod mirror, ⛔ do not assume the ratchet will catch a TS-side-only change; finding: the #5684 zod-mirror-parity guard is BLIND to a key declared on the TS side but absent from the mirror — measured on ObjectGanttSchema, and it is the guard #5927 just used as authoritative #6058 shows it does not.
4. ComponentInput: re-export the one declaration instead of restating it three times #5671 is the executed precedent — read its PR before writing. ⛔ Do not invent a second convergence shape for the sibling type; two convergences of one family that differ structurally are worse than one.

Verification

Pin the divergence itself, not the resulting shape. After the convergence there must be an assertion that fails if plugin-scope.ts ever re-declares the type structurally — an identity pin on the shared declaration, the way #5874's family uses a spy on EXPANDABLE_FIELD_TYPES.has. ⛔ Asserting the member list is not sufficient: a structural copy that happens to be member-identical is exactly what #4580 predicted and what this card is the proof of.

Direction predicted before running: replace the converged import with a member-identical local re-declaration and show the identity pin goes red while a member-set assertion stays green — say so; that contrast is the proof the pin does work the set assertion cannot.

Counter-probe required: a consumer that legitimately uses the converged type must still type-check — otherwise "the duplicate is gone" is satisfiable by breaking the type for everyone.

Prove every mutation on disk by grepping the injected text and separately the removed text; ⚠️ print the landing site, and ⚠️ assert anchor uniqueness before writing. Restore under trap … EXIT INT TERM, and ⚠️ make the restore command cwd-independent — a trap that fires is not a trap that restored (measured on #5926 today). Confirm git diff HEAD --stat empty afterwards. ⚠️ If any ablation crosses a package boundary, rebuild between mutation and reading and prove the mutation reached dist — #5903 did exactly that today and it is what rules out the stale-artifact false green.

Gates by name with exit codes: pnpm --filter @object-ui/types type-check (⚠️ hyphenated; echo the script name; distinguish the unbuilt-closure red from the missing-node_modules red — both hit today, different fixes), root-form vitest over packages/types/src (⚠️ vitest guard refuses --filter forms — use pnpm exec vitest run <paths> from the repo root), the downstream prefix-form sweep pnpm --filter '...@object-ui/types' with the contract direction stated (⚠️ known gotcha measured on #5927: that form fails at apps/site and packages/components until @object-ui/react-runtime is built separately, because it does not depend on @object-ui/types and so is never in the prefix closure; a root build includes it — ⛔ not a finding), and eslint on changed files with merge-base delta. ⚠️ Plain eslint . per package — ⛔ no --no-inline-config.

Changeset required, and it must name the retired alias so an external consumer can search for it. Draft PR, Fixes #5893. ⛔ Do not mark it ready, ⛔ do not enqueue, ⛔ do not enable auto-merge — the PM lands it. Post the terminal report here (plain-text first-line marker os-dev-report; ⛔ avoid short angle-bracket placeholders in prose — the sanitizer ate one on #5926 today) and return it.

---

Generated by Claude Code