Rename the `type` input of `action:button` / `action:icon` (to `actionType`) so it stops colliding with the SDUI discriminator — objectstack#14490 ruling A, objectui half

[os-project-manager]

Filed by the director seat (objectstack #12708, summon #10, session session_01ShyhexkB2d1AeRZ85tgAAe) to carry the objectui half of a maintainer ruling. Filed unassigned; domain:ui, pm:queue. objectstack#14490 is pm:blocked on this card.

Ruling being implemented

Provenance: maintainer, live PM chat with the director seat, 2026-09-02, decision batch #13 item 4 on objectstack#14490, recommendation A, verbatim 「同意」.

Ruled: the collision between the component input named type on action:button and action:icon (the only two of 57 manifest components that declare one) and the SDUI envelope discriminator type is removed at the source. The input is renamed (working name actionType; the implementer may pick a better one if the objectui component config already has a convention) in the objectui component configuration; the pinned sdui.manifest.json in objectstack follows. No alias, no transition window (maintainer's standing 不渐进 rule); the changeset names the rename.

Why

• On the html tier the discriminator refusal landed by objectstack PR #14274 makes type= unauthorable, and its prescription ("write the tag you meant") is wrong for these two components.
• On the react tier the same collision is rescued by specType (objectui#2880); the 2026-09-01 ruling declined to bring that alias to the html tier.
• A declared input that one tier cannot author is a permanent special case in every doc, refusal message and future source format. Renaming removes the special case for every tier at once.

First step (measure, then rename)

Census the real usage of action:button.type and action:icon.type in this repo (component config, examples, schema catalog, console, docs) and in the shipped example apps. objectstack's own census covered only html-tier sources in that repo and found zero; the react-tier population lives here and is unmeasured. Whatever the count, the rename lands directly with the census in the PR body; the count decides the changeset wording, not the direction.

Scope

• objectui: the component config declaring the input; every read site of the prop in the two action renderers; specType rescue left as-is (it addresses other collisions); docs pages that show the prop; catalog entries.
• objectstack (follow-up, not this card): sdui.manifest.json pin bump; the html-tier refusal then needs no component-aware special case; content/docs/ui/react-pages.mdx sentence not needed.

Acceptance

• The manifest scan in objectstack#14490's re-check returns an empty list for inputs named type (control: the variant scan stays non-empty).
• An html-tier page setting the renamed input on action:button parses and renders.
• Clause-② yes (published component input renamed): CONTRACT_REVIEW_TIER; @object-ui/* changeset minor.

Related: objectstack#14490 (the finding and ruling), objectstack PR #14274 (the refusal), objectui#2880 (specType), objectstack #12708 (director seat ledger).

[os-project-manager]

Claim: domain:ui execution seat, session session_01EMrWaQw3XS5DxTHxp4yRyC, branch claude/issue-7415-action-type-input-rename.

Dispatched to one os-dev agent this round, implementing objectstack#14490 ruling A (maintainer, 2026-09-02, decision batch #13 item 4, verbatim 「同意」) as the card carries it. No transition window, no alias, per the standing 不渐进 rule.

File surface declared: the objectui component configuration declaring the input, the read sites in the two action renderers, the schema catalog entries and the docs pages that show the prop, plus a changeset. ⛔ Out of scope and stated so in the dispatch: the sdui.manifest.json pin bump and the html-tier refusal simplification are objectstack-side follow-ups, and specType (objectui#2880) is left alone — it addresses other collisions.

Serial constraints cleared. The lane's other live work is disjoint: #4895 is packages/types + content/docs/blocks, PR #7430 holds components/src/custom/view-states.tsx + plugin-list/src/ListView.tsx, and PR #7428 (enqueued) holds the four plugin views + RecordDetailView.tsx. None of them touches the action renderers or the component config.

Clause ② — yes, and routed. The card states it: a published component input is renamed, so CONTRACT_REVIEW_TIER applies. This seat runs opus and is not at tier, so it goes the same way as #7143 and #4895 — the lane implements against the shape the ruling already pins, the PR carries needs:contract-review, and the project director seat reviews before it lands, under item 3 of the 2026-09-01 ruling on #7143 (batch #27) whose item 4 extends that path to existing cards blocked by the same quota. ⛔ This seat will not self-review and will not enqueue it.

Two things the dispatch adds on top of the card:

1. The census is a first step, not a formality. objectstack's census covered html-tier sources only and found zero; the react-tier population lives here and is unmeasured. The implementer measures it before renaming and reports the figure with a control that fires — a bare zero from a pattern that cannot match the real spelling is the failure mode this lane has hit repeatedly today.
2. Byte reporting. The rename is expected to be roughly neutral, but actionType is longer than type at every occurrence. The measured before/after goes in the PR body rather than an assumption — the lane has two PRs parked on a 177-byte framework ceiling ([Decision] The framework per-chunk eager-closure ceiling leaves 177 bytes for the whole repo — two ruled user-facing fixes cannot land, and no wording of either fits #7399), and the relevant question is which chunk this lands in, not whether a rename "feels" free.

Landing this unblocks objectstack#14490, which is pm:blocked on this card.

---

Generated by Claude Code

[os-project-manager]

Re-dispatched — the first os-dev died with zero work surviving. Claim unchanged: same seat, same session session_01EMrWaQw3XS5DxTHxp4yRyC, same branch.

The container running the subagent restarted before it reported. Measured rather than assumed, so the empty branch does not read as a stalled claim:

• claude/issue-7415-action-type-input-rename exists on origin at 39af82fb4 — a plain (older) main commit, no commits of its own;
• no PR was opened (no subscription event arrived for one);
• main is unchanged at 6411def25, so nothing from this card reached it.

⇒ The branch was pushed at its branch point and nothing was committed — the ordinary shape of a subagent death, ⛔ not a maintainer abort and not a partial landing. A fresh agent is running against current origin/main on the same branch name, which fast-forwards cleanly since the remote tip is an ancestor of main.

Nothing about the card, the ruling or the routing changes: objectstack#14490 ruling A, clause ② yes → the PR will carry needs:contract-review for the project director seat.

---

Generated by Claude Code

[claude]

os-dev-report

{
 "issue": 7415,
 "status": "done",
 "branch": "claude/issue-7415-action-type-input-rename",
 "pr": "https://github.com/objectstack-ai/objectui/pull/7436",
 "premise_still_valid": true,
 "summary": "Renamed the declared input carrying the action execution type on `action:button` and `action:icon` from `type` to `actionType` (registry `inputs` + `defaultProps` on both), and removed the `|| schema.type` fallback at both read sites so there is one spelling and no alias, per objectstack#14490 ruling A. `actionType` is not new vocabulary: it is already this repo's spelling for this value (`action:bar` renames the declared type as it spreads a member; `ActionRunner.execute` resolves `action.type || action.actionType || action.name`; `ActionDef` declares it; both renderers already read it FIRST), and the same carrier collision was resolved upstream for `page:tabs.type` by a rename to `tabStyle`, not an alias - so the convention pointed at the ruling's working name rather than a better one. Behaviour change for consumers: a declaration omitting `actionType` now forwards `type: undefined` and the runner falls through to its own `action.name` leg instead of being handed the component id. `specType` (objectui#2880) untouched; `action:group` / `action:menu` untouched (their members carry the spec `ActionSchema.type` inside an `actions` array, a different surface). PR is draft, carries `needs:contract-review` (read back and confirmed), not self-reviewed, auto-merge not enabled, not enqueued.",
 "census": {
 "question": "authored sites setting the component input named `type` on action:button / action:icon",
 "scope": "4,797 files across examples/, apps/console/, content/docs/, packages/",
 "result_total": 0,
 "rows": [
 "JSON trees (517 files parsed, incl. all 433 schema-catalog entries): 0 nodes of either component. CONTROL on the same walk: 2,410 nodes carrying a `type` key, 127 of them `type: \"button\"`.",
 "md/mdx fenced JSON (234 blocks seen, 207 parsed): 5 `action:button` nodes, key sets `type,label,icon,action` x2 and `type,label,action` x3 - discriminator plus the `action` channel, 0 setting an execution type. CONTROL: 28 `type: \"button\"` nodes in the same blocks.",
 "INDEPENDENT CORROBORATION of that 5: this repo's own corpus census, recorded in packages/react/src/SchemaRenderer.tsx, already reads 'action:button (5 nodes)'. Two different scanners, same number.",
 "props/properties bag carrying `type` on either component, any surface: 0 - none of the 5 nodes has a bag at all.",
 "react-page / JSX tier, an ActionButton or ActionIcon tag with a `type=` attribute: 0. CONTROL: the tag pattern fires 5 times, every one verified individually as a local resolveIcon binding or prose.",
 "content/docs, a page documenting the `type` INPUT of either component: 0. CONTROL: content/docs/core/enhanced-actions.mdx documents the legacy `actionType` dialect, 30 hits, so the term fires.",
 "apps/console, an authored node of either component: 0. CONTROL: public-contract.test.ts lists both ids, so the term fires.",
 "packages non-test source: 0 authoring sites; 61 non-test hits on the component ids, all prose, registry ids, or action:bar's own composition.",
 "read sites of the prop: 2 (action-button.tsx, action-icon.tsx). declaration sites: 2 inputs + 2 defaultProps."
 ],
 "consequence": "The rename breaks no authored document in this repo. A re-check of objectstack#14490's manifest scan would now return an EMPTY list for inputs named `type` - these two were the only ones of the 57 manifest components declaring one - while the stated control, the `variant` scan, stays non-empty: both components still declare `variant`, untouched."
 },
 "name_chosen": "actionType (the ruling's working name). Evidence the objectui config convention points at it rather than something better: (1) it is already the live spelling for this exact value - action:bar spreads `type: componentType, actionType: action.type`, ActionRunner.execute resolves `action.type || action.actionType || action.name`, ActionDef declares `actionType`, and both renderers already read `schema.actionType` first; (2) the identical carrier collision on `page:tabs.type` was resolved upstream by a RENAME to `tabStyle` (objectstack#6776), recorded in apps/console/src/__tests__/registry-inputs-spec-parity.test.ts's HARVESTED table - so the precedent is rename-not-alias, and no competing naming convention exists for component inputs here.",
 "tests": "All readings at 09562aa26, the branch head and final commit; nothing landed after it. ACCEPTANCE (the card's own: an html-tier page setting the renamed input parses and renders) - new file packages/components/src/renderers/action/__tests__/action-type-input-html-tier.test.tsx, 4 rows through the real kind:'html' page path, 4 passed: (1) an action:button tag with `actionType=\"api\"` compiles with no error panel, renders a real button, and on click hands ActionRunner `type: 'api'` with the authored target; (2) NEGATIVE CONTROL - same page with the prop removed still compiles and renders and the api handler is NOT reached, which is what pins row 1's 'api' to the authored prop rather than to a default, the tag, or the handler being the only one registered; (3) no alias - the pre-rename `type=\"api\"` is still the discriminator, the node's type becomes the bare string `api`, no registry entry claims it, the page reports the compile error and renders no button; (4) THE DECLARATION, through the manifest built from the live registry (the same call getJsxManifest makes): `actionType` yields no unknown-prop diagnostic on either component while a bogus prop does, and the registry inputs contain `actionType` and no longer contain `type`. STATED PLAINLY: rows 1-3 are green on origin/main too (the renderer already read `actionType` first, and an undeclared prop is only a WARNING that does not stop a page compiling) - row 4 is the one that measures the rename, and it was added for exactly that reason. ABLATION, two legs, each mutated on disk, PROVEN there (HEAD blob hash vs worktree hash printed per file, injected-text and removed-text grep counts, files-differing-from-HEAD count, empty-hash treated as failure), measured IN THE SAME PROCESS, then restored via `git checkout HEAD -- ABSOLUTE_PATH` under an EXIT INT TERM trap with `git diff HEAD` empty afterwards. NO REBUILD IS INTERPOSED and that is a precondition, not an omission: the mutated files are imported by RELATIVE path from the tests, so they resolve to source, not to dist. Leg 1, revert the declaration on both files (both MUTATED): predicted row 4 red and rows 1-3 green; observed exactly that - `1 failed | 3 passed`, `AssertionError: expected [ Array(1) ] to deeply equal []`. Leg 2, restore the `|| schema.type` read on action-icon only (action-icon MUTATED, action-button UNCHANGED): predicted the new mirror row red; observed `1 failed | 5 passed`, `expected \"vi.fn()\" to not be called at all, but actually been called 1 times` - the component-id trap fires. DISCLOSED: the first ablation attempt was written as `ablate.sh && vitest`, so the EXIT trap restored the tree BEFORE vitest started and it vacuously reported 4/4 green on the unmutated tree. It was re-run with the measurement inside the same process; only the second run is reported. FIXTURE TRIAGE - 17 pre-existing failures in 5 files after the rename, each triaged rather than mass-edited: 6 sites in action-bodyExtra/bodyShape-forward changed SPELLING (they author the renamed input on a directly-rendered action component); action-forward-precedence's ACTION const changed spelling and its Renderer now stamps the component id as `type`, which is the honest post-rename node shape; action-onSuccess-forward got an explicit `asNode` helper because its `clone()` feeds BOTH the two leaf components (node shape, renamed input) and action:group/action:menu members (spec ActionSchema.type, unchanged) - a blanket rename there would have broken the group/menu rows; action-bar-member-type-resolution's standalone row was REPLACED, not respelled, because its stated purpose ('it refuses a fix written as schema.actionType alone') is precisely what the ruling now mandates, and a second mirror row was added that fails if the `|| schema.type` leg is ever re-added. SUITES, union re-run at 09562aa26: `pnpm exec vitest run packages/components/ packages/core/src/registry/ apps/console/src/__tests__/` -> `Test Files 258 passed (258) / Tests 2523 passed (2523)`. Consumer sweep, the 8 files naming either component outside packages/components (app-shell widget/DOM sweep, PageView, useConsoleActionRuntime x2, EnvironmentListToolbar x2, RecordDetailView, schema-catalog DOM-leak) -> `Test Files 8 passed (8) / Tests 275 passed (275)`. `pnpm --filter @object-ui/components run type-check` exit 0 - the script is spelled `type-check` with a hyphen in this repo and the echoed script name in the output confirms it ran rather than matching nothing. `pnpm --filter @object-ui/components run lint` exit 0, 0 errors, 950 pre-existing warnings, run as this repo spells it (`eslint .`, no --no-inline-config). GATES, re-derived from the actual diff paths rather than taken from the dispatch list (objectstack's scripts/pm/dispatch-gates.mjs deliberately NOT used - it only answers about its own tree, and this is a sibling repo): check:action-forward-parity OK, check:control-bytes OK (6,124 tracked text files), check:sdui-registration-pins OK, check:doc-types OK, check:doc-snippets OK (417/417 blocks) after building what it names - its first exit 2 was `PRECONDITION NOT MET`, read as not-a-measurement rather than as a red gate, check:designer-field-key-parity OK, check:vi-mock-specifiers OK, check:vi-mock-inherit OK, check:eager-closure OK. Every verdict quoted from the gate's own printed line; exit codes captured by redirect-then-read, never through a pipe. Repo-wide `pnpm lint` is CI's run and was not repeated locally - the declared narrowing is the single package the diff is confined to. CI not waited on, per the standing 'report at draft-PR time' rule; gate status at report time is in_progress.",
 "bytes": {
 "method": "two full console builds plus check:eager-closure on each; byte-exact figures read from apps/console/dist/eager-closure.json, not from the gate's rounded KB display (which is identical on both builds and would have hidden the delta). BEFORE = 6411def25, this branch's base, built in a dedicated comparison worktree that was removed afterwards; AFTER = 09562aa26.",
 "aggregate_eager_gzip": "3,254,604 -> 3,254,602 (-2 B)",
 "aggregate_eager_raw": "11,175,431 -> 11,175,439 (+8 B)",
 "eager_chunk_count": "48 of 516, unchanged",
 "chunk_framework_gzip": "523,823 -> 523,823 (0 B)",
 "chunk_ui_components_gzip": "396,598 -> 396,598 (0 B)",
 "chunk_ui_components_raw": "1,535,661 -> 1,535,669 (+8 B)",
 "chunk_vendor_objectstack_gzip": "948,329 -> 948,329 (0 B)",
 "verdict": "The chunk question is answered: this lands in `ui-components`, NOT in `framework`. framework is packages/core|react|types and had 177 B (0.2 KB) of headroom with two PRs already parked on it (#7399); this diff is entirely inside packages/components, so framework moves ZERO bytes and the +8 raw bytes land in ui-components, which absorbs them without moving its gzip figure at all. The residual +/-1..5 B wobble on eight other chunks is gzip entropy - every one of them has an UNCHANGED raw size. check:eager-closure passes on both builds. No ceiling raised, no baseline re-pinned, no gate weakened."
 },
 "mcp_calls": "5 - issue_read get, issue_read get_comments, and 3 search_issues (one targeted duplicate check plus two control terms). Everything else went through repo-scoped REST (PR creation, the additive label write and its read-back, the body read-back) or stayed local.",
 "open_questions": [],
 "declined": [
 "Did NOT touch specType (objectui#2880) - it addresses other collisions, and the card scopes it out.",
 "Did NOT do the objectstack-side follow-ups: the sdui.manifest.json pin bump, the html-tier refusal's component-aware special case, and the react-pages.mdx sentence.",
 "Did NOT rename the spec `ActionSchema.type` that action:group / action:menu members and the metadata-admin ActionDefaultInspector author - that is a different surface and out of scope.",
 "Did NOT keep the `|| schema.type` read as a compatibility leg. Reasoning stated so it can be overruled cheaply in review: the card's own scope line names 'every read site of the prop in the two action renderers', which only has content if the read changes; post-rename that leg is exactly 'the old spelling still works', which the ruling forbids; and on every authored path it now yields the component id, never an action type. The cost is real and is named in the changeset and PR body - a direct programmatic render passing a raw ActionSchema no longer resolves its `type`.",
 "Did NOT raise any ceiling, re-pin any baseline, or weaken any gate.",
 "Did NOT run repo-wide `pnpm lint` locally; declared as a narrowing to the one package the diff touches, with CI owning the full farm.",
 "Did NOT wait for CI before reporting, per the standing rule.",
 "Did NOT self-review, enable auto-merge, or enqueue - clause (2) routes this to the project director seat."
 ],
 "out_of_scope_findings": [
 "NOT FILED - handed to PM to file, because BOTH duplicate-check channels are structurally unavailable in this session and filing blind is the one thing worse than not filing. FINDING: the documented `action:button` shape for the page-mode navigation actions is inert. content/docs/guide/record-edit-modes.md (3 nodes) and packages/app-shell/README.md (2 nodes) - these are 3 of the 5 authored action:button nodes the census found, plus the 2 README ones - author it as a nested bag, `{ \"type\": \"action:button\", \"label\": ..., \"action\": { \"action\": \"navigate_create\", \"params\": {...} } }`. Nothing reads that nested key: `action` is not in either renderer's forward whitelist (verified in source and consistent with check:action-forward-parity reporting action:button as forwarding 22 of 25 owed keys), and ActionRunner has no `action.action` read anywhere. The handler IS registered - packages/app-shell/src/console/AppContent.tsx registers `navigate_create` / `navigate_edit` - but the runner resolves a handler from `action.type || action.actionType || action.name`, so the shape that would actually reach it is `actionType: 'navigate_create'` on the node, which is precisely the spelling this PR makes authorable. MY CHANGE DOES NOT REGRESS THIS and that was checked rather than assumed: before, those nodes forwarded `type: 'action:button'`, which binds no handler and no builtin; after, they forward `undefined`, which binds nothing either. Equivalent, both inert. WHY NOT FILED: the REST /search/ endpoint is refused in this session ('sessions are bound to their configured repositories, use repository-scoped endpoints'), and MCP search_issues returns a structural zero for this repo - a control term guaranteed present in issue #7415's own title and body ('actionType', and separately 'discriminator') also returned total_count 0, so the zero for 'navigate_create' is not a reading and cannot support a de-dup claim. Suggested labels if PM files it: docs, plus `finding` if PM reads it as observational rather than a concrete defect."
 ]
}

---

Generated by Claude Code

[os-project-manager]

Contract review — PR #7436 at head 09562aa26 — VERDICT: PASS

VERDICT: PASS
Implemented-by: session_01EMrWaQw3XS5DxTHxp4yRyC
Reviewed-by: session_01ShyhexkB2d1AeRZ85tgAAe

Director seat (objectstack #12708), summon #10, 2026-09-03. Reviewer served at claude-fable-5-1 (get_session re-read this stroke). The domain:ui seat is off tier and handed the review to this seat (5520215260); the maintainer's 「同意，然后执行契约复审」 on decision batch #16 dispatches it. Read against the diff at merge base 6411def25: both renderers, the changeset, the five touched test files and the new html-tier pin.

① Derived judgments — against objectstack#14490 ruling A (decision batch #13 item 4, 「同意」)

1. The rename. Registry inputs and defaultProps on action:button and action:icon declare actionType; type is no longer a declared input on either. actionType is the spelling this repo already used for the value (action:bar's member spread, ActionRunner.execute's resolution order, ActionDef), and the in-repo precedent for the same carrier collision (page:tabs.type → tabStyle) is a rename, not an alias. Judged correct.
2. No alias, no transition window. The || schema.type fallback is removed at both read sites. Judged correct and necessary: post-rename schema.type is the component id on every authored path (SchemaRenderer, the react-page wrapper, action:bar's spread), so the leg could only ever have meant "the old spelling still works", which the ruling forbids. The consumer-visible consequence — a node declaring neither forwards type: undefined and ActionRunner falls through to its action.name leg instead of being handed a component id — is named in the changeset and is strictly better than the objectui#6306 shape it replaces.
3. Scope held. specType (objectui#2880) untouched; action:group / action:menu untouched (their members carry the spec ActionSchema.type inside an actions array — a different surface, and the asNode helper in the onSuccess test keeps the two spellings apart rather than conflating them). The objectstack-side follow-ups (the sdui.manifest.json pin bump, the html-tier refusal's component-aware case, the react-pages.mdx sentence) stay with objectstack#14490, which this PR unblocks.
4. Census, with firing controls on every row: zero authored sites across 4,797 files; the md/mdx count of 5 nodes (none setting an execution type) is corroborated by the repo's own corpus census in SchemaRenderer.tsx. The rename breaks no authored document.
5. Pins. The html-tier file's row 4 (the declaration through the manifest built from the live registry: actionType declared, type and a bogus prop not) is the one that measures the rename, and the PR says so plainly rather than presenting rows 1–3 as proof. The mirror row in action-bar-member-type-resolution.test.tsx is the guard against a re-added fallback. The disclosed vacuous first ablation (trap fired before vitest) and its re-run in-process are the right shape.

② Semver

@object-ui/components: minor, naming the rename and "no alias" — this repo's fixed-group convention never declares major, and the body states the breaking shape. Correct.

③ Boundary flags

• Bytes measured on real builds: framework 0 B, ui-components gzip unchanged (+8 B raw), aggregate −2 B. [Decision] The framework per-chunk eager-closure ceiling leaves 177 bytes for the whole repo — two ruled user-facing fixes cannot land, and no wording of either fits #7399's parked PRs are untouched.
• Not on the governed register. CI: all 30 check runs on 09562aa26 are success or skipped.
• The dev's unfiled out-of-scope finding (the documented page-mode action:button shape with a nested action bag is inert on both sides of this change) is the ui seat's to file; not a landing condition.

Disposition: needs:contract-review cleared on this card and on PR #7436 in this stroke; PR flipped to ready with auto-merge (squash). pm:dispatched stays until the merge closes the card through Fixes #7415; objectstack#14490 leaves pm:blocked on the unlock sweep once this lands.

---

Generated by Claude Code