Repository navigation
studio(actions): switching an action to "Run a script" writes a body the spec refuses until a language is picked by hand — commitOperation seeds no body, and patchBody writes no language #11921
Description
Filing gate: ① product defect, class (a), reach by a named producer, measured. Studio's own action inspector (ActionDefaultInspector) writes the refused shapes. Since objectui#11820 (PR objectui#11918), a new action starts as an update action, so "What it does: Run a script" is the path for every author who wants a script action. Reader: objectui triage first-touch (grade and route).
Dedupe (MCP search_issues, objectui, open + closed):
- "action switched to run a script refused body language missing commitOperation seeds no body" returned objectui#2896 and
@object-ui/corehas no default server dispatcher, so standalone consumers cannot runaction.bodyat all #2904 (closed; the runner's handling ofaction.body). Neither is this. - Positive control, same run: "studio risky no-op defaults new action sandboxed JS hook all objects validation rule" returned objectui#11820: yes.
Filed by the domain:ui execution seat 3 (session_01CGZy1BGCjdN5cXqL9cnvB8) from the objectui#11820 dev report (6053900490, out-of-scope finding 1, and its open question 2, answered "file it, let objectui#11820 land"). ⛔ Not graded or routed here; ⛔ not a claim.
What happens (measured through the spec's ActionSchema on the shapes the inspector writes)
In packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.tsx (objectui main 3683591):
commitOperationclearsoperation/patchwhen the author picks "Run a script", and seeds nobody.ActionSchemarefuses the result atbody: "A script action requires either an inline body … or a target".patchBodywrites{ source }as the author types, with nolanguage.ActionSchemarefuses it atbody.language: "Invalid discriminator value. Expected 'expression' | 'js'".- It parses only after the author explicitly picks a language (for example Sandboxed JS (L2)).
Done when
- Switching an action to a script writes a body
ActionSchemaparses, with a seeded language, so the first keystroke is not refused. - A pin parses each shape the inspector writes through
ActionSchema: after the switch, after the first keystroke, and after a language change.
Dedupe words: action what it does run a script body language refused · commitOperation seeds no body · update action switch to script
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 8, 2026
Triage: first grade,
bug·priority:p2·domain:ui·area:studio·pm:queue. Direction: every shape the inspector writes parses throughActionSchemaTriage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) ·
session_01AavokzJ5DndAwitDXvKy4U· 2026-10-08T08:02Z. ⛔ Not a claim, ⛔ not a dispatch.Triage: lands in
packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.tsx(commitOperation,patchBody) ⇒domain:ui; rationale: Studio's action inspector in objectui.objectstack-fleet commentedon Oct 8, 2026
Claim: PM loop round 4
Session:
session_01MgfduSkFrfM3eorB3UGfAUAccount:
os-billBranch:
claude/issue-11921-script-action-languageWorktree:
objectui-issue-11921Domain:
domain:uiSeat:
domain:ui#2File surface (on
e81c846):packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.tsx:commitOperationandpatchBody. Switching to Run a script seeds a body with a language, and every shape the inspector writes parses through the spec'sActionSchema..changeset/11921-*.md.⛔ Not on it:
ActionSchema, which is the reference.ObjectActionsPanel.tsx's create flow (objectui#11820, landed) beyond what the switch needs.StudioDesignSurface.tsx.packages/components/src/ui/**andpackages/i18n/**.Any file outside this list: the dev reports it before opening the PR (stop on breach; explain in the report).
Container & model:
S,mode:subagent,model: opus(dispatch-gates --tier --repo objectstack-ai/objectuiover these paths: no path-derived mandate; default tier)Clause-②: no
Responsibility:
objectui app-shell Studio action inspector: switching an action to "Run a script" writes a body the spec refuses until a language is picked by hand, because commitOperation seeds no language | the spec's ActionSchema states the required shape | every Studio author writing a script action since objectui#11820 made "Run a script" the path, measured by triageThread-read: 6055517030
Serial constraints cleared:
noneblocking.ActionDefaultInspector.tsx.Why
Clause-②: no: Studio writes a shape the spec already accepts. No export, prop, spec key orpackages/i18nkey changes.Generated by Claude Code
objectstack-fleet commentedon Oct 8, 2026
os-dev-report
{
"issue": 11921,
"status": "done",
"branch": "claude/issue-11921-script-action-language",
"pr": "#11941",
"session": "session_01MgfduSkFrfM3eorB3UGfAU (subagent mode; seat domain:ui#2, claim comment 6055828336)",
"premise_still_valid": true,
"summary": "Draft PR #11941 (head 9bbaad1, first line 'Fixes #11921', 'Clause-②: no' at a line start) puts every script-body write in ActionDefaultInspector through one writer, writeScriptBody: the 'What it does: Run a script' switch (commitOperation) seeds body { language: 'expression' }, typing (patchBody) keeps the language, and a language change writes the new language's shape (keys only the old language admits, read off HookBodySchema.options, are dropped; a key neither admits is kept for the spec to name). H1: the spec declares NO default (HookBodySchema's discriminator is a bare z.literal on each side); the seeded 'expression' is the inspector's own offer, first in BODY_LANG_OPTS and the value the picker already showed for an absent language (on main: value={(typeof body.language === 'string' ? body.language : undefined) || 'expression'}), now one constant SEEDED_BODY_LANGUAGE read by picker and writer. H2 confirmed: switch, keystroke, language change and away-and-back all go through the writer; on main the switch was refused at body, the first keystroke at body.language, and JS-with-capabilities to expression at body (unrecognized_keys). H3 falsified for one write: commitOperation('update') wrote no patch, which ActionSchema refuses ('nothing to write'); it now seeds patch {} when absent (bounded in-place fix, 4 conditions stated in the PR); the Type control's writes for api/flow/modal/form/url are unchanged (pinned). RULING CONFLICT, not silently resolved: the triage pin 'parse after the switch' is impossible without writing a source the author never typed: HookBodySchema requires source z.string().min(1) in BOTH languages (measured: body {language} refused at body.source:invalid_type, {language, source: ''} at body.source:too_small, for expression and js). The PR does not invent a source; the switch pin asserts the body names the picker's language and the ONLY refusal is body.source; see open_questions[0]. CI at report time (one read, not awaited): 42 check runs, 21 success, 3 skipped, 18 in_progress, 0 failed.",
"tests": "All on 9bbaad1 via os-verify-lock (slot issue-11921), from the worktree root. (1) pnpm exec vitest run --maxWorkers=2 packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.scriptBodyLanguage-11921.test.tsx: 'Tests 11 passed (11)'. (2) the 17 test files git grep finds naming ActionDefaultInspector or ObjectActionsPanel (objectui#11820's newActionPin among them) + packages/core/src/utils/tests/column-identity.ratchet.test.ts: 'Test Files 18 passed (18)', 'Tests 152 passed (152)'. (3) pnpm --filter @object-ui/app-shell type-check (echo: 'tsc --noEmit && tsc -p tsconfig.test.json'): exit 0, after turbo run build --filter='@object-ui/app-shell^...' (28/28 successful, 8 cached); tsc -p tsconfig.test.json --listFilesOnly lists the new pin file (1 hit). (4) pnpm exec eslint --format json on the 2 changed .ts/.tsx files: 2 judged, 0 errors, 0 warnings; narrowing proof: population = eslint.config.js files: ['**/*.{ts,tsx}'] blocks, file count from the JSON, and the config sets no parserOptions.project/projectService (not type-aware), so the diff cannot move an untouched file's verdict; pnpm lint is CI's. (5) Ablation, objectstack scripts/ablation-replace.mjs WRAP mode, no build (the pin imports the inspector from src by relative path): leg 1 writeScriptBody 'return next;' replaced by 'return { ...body, ...change };' (anchor 1 to 0, blob d9fe3748fba9 to 5720b777da8e): 'Tests 5 failed | 6 passed (11)', all AssertionError (switch, first keystroke, away-and-back, no-language keystroke, JS-to-expression); leg 2 deleted 'if (!hasPatch) cleared.patch = {};' (blob d9fe3748fba9 to 4430ef259b01): 'Tests 2 failed | 9 passed (11)' (away-and-back refused at patch:custom; update-switch control); restore proven by the tool for both legs (blob after restore d9fe3748fba9 == HEAD blob, git diff HEAD empty), re-read after: DIFF_LINES=0. Direction observed: red, as expected.",
"gates": [
"HEAD (git rev-parse --short HEAD after the last commit): 9bbaad1",
"pnpm check:spec-symbols exit=0: 'ledger anchors: 2 declared; none renders an instruction its anchor card can no longer receive.'",
"pnpm check:metadata-write-doors exit=0: 'OK 17 metadata write door(s) derived (3 raw PUT, 14 SDK) ... no writer list anywhere.'",
"node scripts/check-changeset-presence.mjs exit=0: '2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s): .changeset/11921-script-action-language'",
"node scripts/check-changeset-no-major.mjs exit=0: 'No changeset declares a
majorbump.'","pnpm check:new-line-citations exit=0: 'VERDICT new-cross-file-line-citations: 0 new citation(s), enforcement report-only -> exit 0'",
"pnpm check:control-bytes exit=0: 'check-control-bytes: OK (scanned 7993 tracked text file(s); skipped 85 binary).'",
"pnpm exec eslint --format json (2 changed files) exit=0: 2 judged, 0 errors, 0 warnings",
"pnpm --filter @object-ui/app-shell type-check exit=0",
"pnpm exec vitest run (18 files incl. column-identity.ratchet) exit=0: 'Test Files 18 passed (18)', 'Tests 152 passed (152)'",
"re-derived additions, each exit=0 with its own OK/verdict line: check-changeset-overwrite.mjs, check-changeset-fixed.mjs, check:changeset-claims, check:pending-changeset-literals, check:action-forward-parity, check:designer-field-key-parity, check:handler-key-reads, check:i18n-keys, check:phantom-deps ('Every in-scope import is declared by the package that publishes it.' — the diff adds a runtime import of @objectstack/spec/data to app-shell), check:self-import, check:esm-specifiers, check:unreferenced-sources, check:shell-escape-residue, check:test-path-roots, check:vi-mock-specifiers, check:vi-mock-inherit, check:vi-mock-override-shape, check-type-check-coverage.mjs"
],
"files_changed": [
"packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.tsx",
"packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.scriptBodyLanguage-11921.test.tsx",
".changeset/11921-script-action-language.md"
],
"deviations": [
"Triage pin 'parse after the switch' not met as written: measured impossible without seeding a non-empty source (HookBodySchema source z.string().min(1) in both languages). Pinned instead: the switch's body names the picker's language and the only refusal is body.source. Raised as open_questions[0]; 'Fixes #11921' kept as dispatched, consistent with recommendation A there. If the seat rules B or C, the first PR line should become 'Part of #11921' (seat edit).",
"Dispatch H3 ('nothing else changes') falsified for the switch INTO 'Update fields': it now also seeds patch {} when the draft holds none, under the bounded in-place exemption (same defect class, mechanical with a shape objectui#11820 pins, file held by no other claim, same gate family); declared in the PR body with ablation leg 2.",
"commitOperation's switch to Run a script seeds no body when the draft carries a target (the spec accepts target in place of a body); a control pins it.",
"Docs-driven rule: no doc sentence describes the script action's language (git grep over content/docs and the app-shell README: zero; the same pipeline matches other 'language' lines), so nothing to report there."
],
"mcp_calls": "0",
"api_writes": "3 REST writes, all through the fleet-write relay as objectstack-fleet[bot] (each = one POST /repos/objectstack-ai/objectstack/dispatches executing one op): (1) pr_create POST /repos/objectstack-ai/objectui/pulls, draft #11941, read-back 11887 bytes sent == stored; (2) assign POST /repos//issues/11941/assignees os-bill (label-write.mjs, read-back MATCHES); (3) comment POST /repos//issues/11921/comments (this os-dev-report, post-stamped.mjs). Not REST: git push x2 (empty-branch probe, then 9bbaad1). Everything else was GET reads.",
"open_questions": [
{
"question": "The triage pins 'parse after the switch', but the spec refuses any body without a non-empty source in both languages. What should the switch to 'Run a script' write?",
"options": [
"A (this PR): seed { language: 'expression' } only; the switched draft is refused at body.source alone until the author types, and the first keystroke parses. The same incomplete state the Type control's url/flow/modal/form/api switches already leave at target.",
"B: also seed a starter source (e.g. the editor placeholder's 'return { ok: true }' with language js, the pre-objectui#11820 skeleton), so the switch parses at once; Studio then writes a runnable no-op handler the author never wrote.",
"C: a held state for incomplete actions in the objectui#11786 family (objectHeldEdit in metadataError.ts learns actions; a designer string for the hint): the switch is held unsent with a 'needs a script body' line instead of a refusal. Off this claim's surface (metadataError.ts, StudioDesignSurface.tsx, metadata-admin/i18n.ts)."
],
"recommendation": "A now, C as a follow-up only if the pre-keystroke refusal is seen in practice. Business need: an author who picks 'Run a script' types next; the 1500 ms autosave window is the only time the refusal can show. Long-term: A writes only what the author chose plus the language the picker shows, and C is the one mechanism for every incomplete switch (url/flow/modal/form/api too), not a per-type workaround. AI-error axis: B writes metadata that runs and returns ok while doing nothing (the no-op default class objectui#11820 removed), whereas A's refusal names the missing input. Startup focus: A adds no surface; C adds a hold path and strings, so it should be wanted before it is built."
},
{
"question": "The body editor's placeholder ('// (input, ctx) => result' / 'return { ok: true }') and hint (engine.inspector.action.scriptBodyHint, 'Runs in the sandbox as (input, ctx) => Promise of output') are JS whatever language is picked, while the seeded and displayed language is Expression (L1). Seed js instead, or make the copy language-aware?",
"options": [
"A: keep 'expression' (the ruling's 'default the inspector already offers' and H1); a follow-up makes placeholder and hint language-aware the way HookDefaultInspector already does (needs a designer string in metadata-admin/i18n.ts, outside this claim).",
"B: seed 'js' to match the editor copy and the former Studio skeleton; the picker then shows Sandboxed JS (L2) for every new script body, against H1."
],
"recommendation": "A: it follows the ruling and H1 as dispatched, and the mismatch exists on main already (the picker showed Expression under the JS placeholder). Whether an expression body holding JS is refused by lint or only fails at run time was NOT MEASURED, so the follow-up's priority is the seat's call."
}
],
"out_of_scope_findings": [
"class: a · reach: named producer — Studio's hook inspector, HookDefaultInspector.tsx patchBody ({ ...body, ...p }) via its Language picker · evidence: a JS hook body carrying capabilities switched to Expression is written as { language: 'expression', source: 'return;', capabilities: ['api.read'] }, which @objectstack/spec 17.7.0 HookSchema.safeParse refuses at body (unrecognized_keys); the same writer defect this PR closes for actions (the same defect family as objectui#11921; routing is the seat's, the dev files nothing) · dedupe words: hook inspector language change capabilities refused · hook body expression unrecognized keys · HookDefaultInspector patchBody",
"class: a · reach: Studio action inspector renders it (pin DOM: Script language 'Expression (L1)' above the JS placeholder and the sandbox-signature hint) · evidence: placeholder '// (input, ctx) => result' / 'return { ok: true }' and hint engine.inspector.action.scriptBodyHint describe a JS body under an Expression default; runtime consequence NOT MEASURED · carrier: open_questions[1] above; noted in the PR's Acceptance notes · dedupe words: action script body placeholder JS expression language hint mismatch",
"carrier: none (承接者:无) · the Type control's switch to 'script' still writes no body (refused at body until the first keystroke, which now parses); url/flow/modal/form/api switches are refused at target until set, measured on objectui#11820 · noted in the PR's Acceptance notes, not filed (the objectui#11786 incomplete-state family; see open_questions[0] option C)"
],
"cleanup": "pending at report time: worktree /home/user/objectui-issue-11921 is removed after this comment is read back (node_modules first, no --force); no dev server or monitor started"
}
Generated by Claude Code
objectstack-fleet commentedon Oct 8, 2026
os-dev-report
{
"issue": 11921,
"round": "patch round 2 (REWORK, first-load bytes)",
"status": "done",
"branch": "claude/issue-11921-script-action-language",
"pr": "#11941",
"head": "b3c799644",
"session": "session_01MgfduSkFrfM3eorB3UGfAU (subagent mode; seat domain:ui#2, claim comment 6055828336)",
"premise_still_valid": true,
"summary": "The runtime import of HookBodySchema from @objectstack/spec/data is gone from ActionDefaultInspector.tsx. That file now names no @objectstack/spec/data specifier at all (grep count 0; the same grep finds 1 in the pin file). Its only @objectstack/spec import is the existing
import type { ActionLocation }from spec/ui. BODY_KEYS_BY_LANGUAGE is now a literal Map in the inspector: expression is language and source; js is language, source, capabilities, timeoutMs and memoryMb. SEEDED_BODY_LANGUAGE and writeScriptBody are unchanged. The pin file imports HookBodySchema test-side and adds a parity block with two tests. The first checks that the constant names the same languages as HookBodySchema.options. The second checks, per language, that the constant admits exactly the keys of that spec shape, with sorted arrays compared both ways. Commits: 539316f (constant + parity pin), then b3c7996, which merges main f3a0488 with no conflict. The merge brought no change to the three PR files, and the branch diff against main is still exactly those 3 files. The changeset needed no edit: it never said the keys were read from the spec. CI's Bundle Analysis on the merge ref is the byte measurement; no console build was run here.","tests": "Run on b3c7996 via os-verify-lock (slot issue-11921), from the worktree root. The lock held 442 s, waited 111 s, and printed 'VERDICT command-exit 0'. (1) Type-check prerequisite: the dependency closure, built with turbo run build --filter='@object-ui/app-shell^...' (not the console): 28 of 28 successful, 8 cached, CLOSURE_BUILD_EXIT=0. (2) pnpm --filter @object-ui/app-shell type-check (echo: 'tsc --noEmit && tsc -p tsconfig.test.json'): TYPECHECK_EXIT=0. (3) The same 17 test files naming ActionDefaultInspector or ObjectActionsPanel (the list is unchanged after the merge) plus packages/core/src/utils/tests/column-identity.ratchet.test.ts: 'Test Files 18 passed (18)', 'Tests 154 passed (154)', TESTS_EXIT=0. That is 152 + the 2 new parity tests. (4) Ablation with objectstack scripts/ablation-replace.mjs in WRAP mode, no build (the pin imports the inspector from src). Leg A, the asked one: the fake key 'fakeKey' added to the constant's expression entry (anchor 1 to 0, blob 7c2cfc2c03ae to 36a8d8d8d988). Result: 'Tests 1 failed | 12 passed (13)', the parity test failing with "AssertionError: expression: expected [ 'fakeKey', 'language', 'source' ] to deeply equal [ 'language', 'source' ]". Leg B, the other direction: spec key timeoutMs removed from the constant's js entry (blob 7c2cfc2c03ae to 1916f10fa686). Result: 'Tests 2 failed | 11 passed (13)': the parity test ('js: expected [ capabilities, language, …(2) ] to deeply equal [ …(3) ]') and the JS-to-expression behaviour pin. Restore was proven by the tool for both legs: blob after restore 7c2cfc2c03ae equals the HEAD blob, and git diff HEAD is empty. Re-read after the run: DIFF_LINES=0, and the HEAD blob equals the disk blob.",
"gates": [
"HEAD (git rev-parse --short HEAD after the last commit): b3c7996",
"pnpm exec vitest run (18 files) exit=0: 'Test Files 18 passed (18)', 'Tests 154 passed (154)'",
"pnpm --filter @object-ui/app-shell type-check exit=0",
"pnpm exec eslint --format json (2 touched .ts/.tsx files) exit=0: files judged 2, errors 0, warnings 0",
"node scripts/check-changeset-presence.mjs exit=0: '2 source file(s) of 1 released package(s) changed, and this change declares 1 changeset(s): .changeset/11921-script-action-language.md.'",
"node scripts/check-changeset-no-major.mjs exit=0: 'No changeset declares a
majorbump.'","node scripts/check-changeset-overwrite.mjs exit=0: 'No pre-existing changeset was modified or deleted.'",
"node scripts/check-changeset-fixed.mjs exit=0: 'All workspace packages are in the changeset fixed group.'",
"pnpm check:changeset-claims exit=0: 'No pending changeset names a file this change touches.'",
"pnpm check:pending-changeset-literals exit=0: 'No test source names a pending changeset.'",
"pnpm check:phantom-deps exit=0: 'Every in-scope import is declared by the package that publishes it.'",
"pnpm check:unreferenced-sources exit=0: 'OK Every shipped source file in every covered package is reachable.'",
"extra, cheap: pnpm check:control-bytes exit=0 ('OK (scanned 7998 tracked text file(s); skipped 85 binary)'); pnpm check:new-line-citations exit=0 ('0 new citation(s)')",
"NOT MEASURED here, by instruction: eager-closure gzip bytes. CI's Bundle Analysis on the merge ref is the measurement."
],
"files_changed": [
"packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.tsx",
"packages/app-shell/src/views/metadata-admin/inspectors/ActionDefaultInspector.scriptBodyLanguage-11921.test.tsx",
".changeset/11921-script-action-language.md (unchanged this round)"
],
"deviations": [
"'Module-private constant': BODY_KEYS_BY_LANGUAGE is a named export of ActionDefaultInspector.tsx, so that the parity pin can compare it directly. A test cannot read an unexported binding, and a behaviour-only probe would see only the key differences between the two languages, not a key added to or missing from both. It follows this file's precedent, PARAM_TYPE_OPTS (exported for objectui#6538's pin), with the same react-refresh/only-export-components directive. No barrel re-exports it: inspectors/index.ts exports only registerBuiltinInspectors, and the package entry does not export the inspector module. So it is not package surface, and Clause-② stays no.",
"Ablation: the asked leg (a fake key added) went red, and a second leg (a spec key removed) was also run to show the other direction. Both restores were proven.",
"The type-check prerequisite built the app-shell dependency closure (packages only, not the console) under the same lock call, because the fresh worktree had no dist.",
"The PR body was not edited (no PATCH by the dev). Seat edit needed. In '## What changes', the sentence starting 'The per-language key sets are read off the spec's own union (
HookBodySchema.optionsfrom@objectstack/spec/data)' is now false. Replace it with: 'The per-language key sets are a constant in the inspector,BODY_KEYS_BY_LANGUAGE; a parity pin in the test file compares it withHookBodySchema.optionsboth ways, importing the spec test-side only, so the schema stays out of the console's first load.' In '## Tests', the head becomesb3c799644: 'Tests 13 passed (13)' for the pin file, and 'Test Files 18 passed (18)', 'Tests 154 passed (154)' for the set. Add this round's ablation legs A and B.","The merge commit b3c7996 had its message amended before its first push to add the model-free trailer pair. No pushed history was rewritten."
],
"mcp_calls": "0",
"api_writes": "1 REST write this round: comment POST /repos//issues/11921/comments (this os-dev-report), through post-stamped.mjs and the fleet-write relay (one POST /repos/objectstack-ai/objectstack/dispatches). Not REST: git push x1 (9bbaad1..b3c7996). The PR stays draft and was not edited. Reads only otherwise.",
"open_questions": [],
"out_of_scope_findings": [],
"cleanup": "the worktree /home/user/objectui-issue-11921 and the private ref refs/os-dev-11921/main are removed after this comment is read back (node_modules first, no --force)"
}
Generated by Claude Code
objectstack-fleet commentedon Oct 8, 2026
Review: ACCEPT — PR objectui#11941 ·
domain:uiseat 2 · 2026-10-08T10:15ZImplemented-by:
os-devsubagent ofsession_01MgfduSkFrfM3eorB3UGfAU, branchclaude/issue-11921-script-action-language, headb3c7996(round 2, a merge ofmainf3a0488). Reports6056558111(round 1) and6056879954(round 2).Reviewed-by:
domain:uiseat 2 (session_01MgfduSkFrfM3eorB3UGfAU), against GitHub and the branch diff, not against the report's prose. The seat refreshed the PR body to matchb3c7996.Checklist:
main, first lineFixes #11921. No closing keyword sits next to another card.Clause-②: nois at the start of a line. The assignee isos-bill.6055828336:ActionDefaultInspector.tsx;writeScriptBody.{ language: 'expression' }.HookBodySchema's discriminator carries no.default.expressionis the inspector's own first offer, the value its picker already showed for an absent language. It is now one constant that the picker and the writer both read.source.HookBodySchemarequires a non-emptysourcein both languages. So triage's "parses after the switch" pin cannot hold without writing code the author never typed. Instead, the switch's only refusal isbody.source, and the first keystroke parses. That is the same incomplete state the other What it does switches leave attarget.expression. The JS-flavoured placeholder and hint under an Expression picker already exist onmain, and they are a follow-up,carrier: none.patch: {}when none is held, becauseActionSchemarefuses an update with nothing to write. It is the same defect class in the same writer, pinned with an ablation. The other What it does types write exactly what they wrote before, and that is pinned.HookBodySchemaimport kept 7 spec declarations in the eager closure, because the inspectors register eagerly throughregister-builtins.BODY_KEYS_BY_LANGUAGEconstant. A test-side parity pin checks it againstHookBodySchema.optionsin both directions.Clause-②: nostill holds.@object-ui/app-shell: patch. Every sentence was checked against the diff. It never claimed the keys are read from the spec at runtime.ActionDefaultInspectororObjectActionsPanel(objectui#11820's pins among them) and the column-identity ratchet.check:phantom-depsandcheck:unreferenced-sourcespass.b3c7996's merge ref at 3307.6 KB againstmainf3a0488's 3307.4 KB, about +0.2 KB. It is a Studio fix, so under ruling6056819248it lands against the +5,120 B allowance (PR objectui#11949), after objectui#11956 has mademaingreen and objectui#11949 has landed. ItsBundle Analysismust then re-run green on a fresh merge ofmain.Out-of-scope notes:
carrier: none: the action editor's placeholder and hint describe a JS body under the Expression default.carrier:to be filed by this seat when the write budget allows.HookDefaultInspector's language change spreads the old body, so a JS hook body withcapabilitiesswitched to Expression is refused atbody(unrecognized_keys). It is the same writer defect this PR closes for actions.Generated by Claude Code
fix(app-shell): the action editor writes a script body that names its…