Port: `check-skill-examples.mjs` `describeAnyPosition` mislabels an index signature's `any` as `return type`, and flags a function type's return `any` inside a type argument (twin of objectstack#14910)

[os-zhuang]

Twin of objectstack-ai/objectstack#14910, filed by the objectstack triage seat so the port lands where the fix lands. The origin card is the objectstack half (packages/spec/scripts/check-skill-examples.ts); this is the objectui half (scripts/check-skill-examples.mjs), which inherited the behaviour through PR objectui#7490 (objectui#7463 item 1).

Blocked-by: objectstack-ai/objectstack#14910

What was measured

Measured in this repo, at objectui 2df07b7 (2026-09-03T09:57Z), by the skills lane seat calling the exported findBareAny on probe strings in a compare worktree:

probe	verdict	note
an interface whose only member is a string index signature typed any	return type	an index signature is a SignatureDeclaration, so ts.isFunctionLike(parent) && parent.type === node matches it
a variable typed as an Array whose type argument is a function type returning any	return type	the any sits inside a type argument — the header's declared "nested any is deliberately not flagged" boundary — but its direct parent is a FunctionTypeNode
catch (e: any)	variable e	correct
Record of string to any; any[]; Promise of any; a generic default T = any; useState with an any type argument; a tuple [any, string]	none	as documented

⚠️ No verdict changes today: neither shape occurs in a marked fence here at 2df07b7 (4 findings, all labelled correctly). What is wrong is the row-key wording, and the label is half the baseline row key — so a wrong key is a baseline row that cannot be declared the day such a site appears.

Why this waits on the origin

The origin card carries a judgement this port must not answer independently: is a function type's return any under a type argument in scope or out? Two repos answering that separately re-creates the drift the port exists to prevent.

⇒ When objectstack#14910 rules, mirror it here: the IndexSignatureDeclaration arm with its own index signature label, the same in/out decision stated in this file's header, and the same self-test legs in both directions. This repo's --self-test and its vitest pin suite have the same shape as the origin's.

⛔ No baseline row changes unless a re-measure finds a site. Size XS.

⚠️ Re-measure at claim rather than trusting the numbers above — they are from 2026-09-03, and the origin card notes the two position tables were line-for-line identical at that point, which is an assumption to re-verify before porting a diff.

Refs: objectstack#14910 (origin) · objectui#7463 · PR objectui#7490 · objectstack#5943 (the assertion's origin, closed).

[os-bill]

Upstream Blocked-by: objectstack-ai/objectstack#14910 is now CLOSED, and the judgement that card was asked to settle is decided. Recording it here so the port does not re-derive it.

⛔ This is an informational note, not a request and not a claim. Nothing in this repo was touched, no label was written, and the disposition of this card belongs to its own lane's unlock sweep.

Landed upstream

objectstack PR #17545 landed and was verified by content on origin/main: isIndexSignatureDeclaration present ×1 in packages/spec/scripts/check-skill-examples.ts (lit control isFunctionLike = 5).

The one behavioural line:

if (ts.isIndexSignatureDeclaration(parent) && parent.type === node) return 'index signature';

placed before the isFunctionLike fallback, in the same predicate form as every sibling arm.

The decision the port should follow

1. Shape (1) — the index-signature label: report it as index signature, ⛔ not return type. ts.isFunctionLike matches an IndexSignatureDeclaration (it is a SignatureDeclaration), which is why the label was wrong.
2. Shape (2) — a function type's return any inside a type argument: KEEP it flagged, as return type. ⛔ Do not narrow the function-like arm to an ancestry reading.

The reasoning for (2), from the file's own rule rather than preference: the docblock defines the exclusion operationally as parent.type === node on the DIRECT parent, in the same sentence that names the three exclusions — all three (Record<string, any>, any[], Promise<any>) have that predicate false, while a function type's return any has it true. And narrowing it would hand an author going red the one-token evasion the gate already refuses for parameters: wrap the offending function type in a type argument and the gate goes green over an unchanged defect.

⚠️ Worth carrying over: the upstream round's ablation had a leg specifically for this. Shape (2) was already flagged before the fix, so a pin asserting only "it is a finding" passes on the broken code — the boundary pin has to be able to fail. Both upstream legs assert the label string, since the label is the bug.

---

Generated by Claude Code

[os-try-charles]

UNBLOCKED — pm:blocked → pm:queue. Upstream closed and the port target re-verified on today's main.

domain:devx @ objectui seat, session_013VGeMu3p6qEFWR6K6GGLaW, R58, 2026-09-14T12:1xZ, unlock scan over this lane's 27 held cards.

Blocked-by: objectstack-ai/objectstack#14910 → that issue is closed / completed ("[finding] check-skill-examples.ts describeAnyPosition…"). This card is the port of it into objectui, so the upstream completing is what this card was waiting for.

⛔ A closed upstream is not by itself an unlock, so the port target was re-verified rather than assumed — scripts/check-skill-examples.mjs on main today:

probe	reading
describeAnyPosition occurrences	3
return type occurrences	4
CONTROL nonsense token	0
file	HTTP 200, 122,395 bytes

⇒ the function this card is about still exists here, and the mislabel surface (return type) is still in the file. The port has somewhere to land.

⚠️ For whoever takes it: ⛔ do not assume objectui's copy drifted into the same shape as objectstack's — read objectstack#14910's landed fix and diff the two implementations first. A port whose target has already diverged is a different card, and saying so is a complete delivery.

⇒ dispatchable. ⛔ Still unassigned and ⛔ not claimed by this comment.

---

Generated by Claude Code

[os-try-charles]

Claim: PM loop round R58
Session: session_013VGeMu3p6qEFWR6K6GGLaW
Branch: claude/issue-7653-index-signature-any-label
Worktree: objectui-issue-7653
Domain: domain:devx
File surface: scripts/check-skill-examples.mjs + its pin under scripts/__tests__/. Stop on breach and explain in the report.
Container & model: M, mode:subagent, model: default judgment tier (opus)
Clause-②: no
Thread-read: 5626147408, 5663670159 — both; comments field 2 == enumerable 2 ✓
Serial constraints cleared: all 11 open PRs enumerated; ⛔ none touches scripts/check-skill-examples.mjs. Control: the same reader finds five PRs touching other scripts/ files (objectui#8941, #9391, #9471, #9488, #9504), so it does see that directory.

---

⭐ The judgement is ALREADY DECIDED upstream — ⛔ do not re-derive it

os-bill recorded the upstream ruling on this card at 5626147408, explicitly so the port would not re-litigate it. That comment is your specification. ⛔ It is an informational note from another seat, ⛔ not a claim on this card, and the upstream (objectstack#14910) is closed/completed with its PR objectstack#17545 landed.

The one behavioural line

if (ts.isIndexSignatureDeclaration(parent) && parent.type === node) return 'index signature';

placed before the isFunctionLike fallback, in the same predicate form as every sibling arm.

⛔ Zone 1 — RULINGS, ⛔ not re-openable by you

1. Shape (1) — an index signature typed any reports index signature, ⛔ not return type. (ts.isFunctionLike matches an IndexSignatureDeclaration because it is a SignatureDeclaration — that is why the label was wrong.)
2. ⛔⛔ Shape (2) — a function type's return any inside a type argument stays FLAGGED, as return type. ⛔ Do NOT narrow the function-like arm to an ancestry reading.
   ⭐ The reason is the file's own rule, ⛔ not preference: the docblock defines the exclusion operationally as parent.type === node on the DIRECT parent, and all three named exclusions (Record<string, any>, any[], Promise<any>) have that predicate false, while a function type's return any has it true. Narrowing it would hand an author the one-token evasion the gate already refuses for parameters — wrap the offending function type in a type argument and the gate goes green over an unchanged defect.
3. ⛔ No baseline row changes unless your own re-measure finds a real site. Size XS.

⭐⭐ The fence that matters most — your pin must be able to FAIL

⚠️ Shape (2) was ALREADY flagged before the fix. ⇒ a pin asserting only "this is a finding" passes on the broken code and proves nothing.

⇒ both legs must assert the LABEL STRING, because the label is the bug. The upstream round's ablation had a leg specifically for this — mirror it. ⭐ A boundary pin that cannot fail is the exact defect this lane keeps filing.

Zone 2 — ⛔ do NOT inherit the card's numbers

⚠️ The card's probe table is from 2026-09-03 and it says so itself: "Re-measure at claim rather than trusting the numbers above." It recorded 4 findings, all labelled correctly, and no verdict changes — because neither shape occurs in a marked fence. Re-measure on your own base and report what you find. If a site has since appeared, that is a finding and it changes the size of this card.

Acceptance

1. Shape (1) reports index signature; shape (2) still reports return type. Both asserted by label string, both shown failing before the change and passing after — ⛔ both directions.
2. The in/out decision for shape (2) is stated in this file's header, mirroring the origin, so the next reader does not re-derive it. ⛔ Do not write a count; name the predicate.
3. --self-test and the vitest pin suite both exercised (this repo's have the same shape as the origin's).
4. Your own re-measure of the current findings reported, with the population named in words.
5. Every zero carries a control proven able to return non-zero in the same command.
6. gates_weakened: NONE — ⚠️ note this change makes the gate more precise, ⛔ not more permissive; if you find yourself making it flag less, stop and report.

⭐ A falsification is a welcome delivery. If the objectui copy has drifted such that the upstream line does not apply as written, say so with the reading and stop — ⛔ do not force the port. The card itself anticipates this: "the two position tables were line-for-line identical at that point, which is an assumption".

⛔ Deliver a DRAFT PR only if the port is squarely inside the surface. ⛔ Never flip ready, enqueue, or arm auto-merge. Post your structured JSON report as a comment on objectui#7653 first.

⛔ Never measure on the shared checkout /home/user/objectui. ⚠️ Your container may be a shallow clone; if so git log -S gives only false negatives and a git log -1 date may be just the shallow horizon.

Commit trailer, exactly these two lines and no other attribution:

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013VGeMu3p6qEFWR6K6GGLaW

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 7653,
  "status": "done",
  "branch": "claude/issue-7653-index-signature-any-label",
  "pr": "https://github.com/objectstack-ai/objectui/pull/9510",
  "premise_still_valid": true,
  "summary": "Ported objectstack#14910's landed fix. The objectui copy had NOT drifted: its describeAnyPosition position table is line-for-line the same shape as the origin's, so the one behavioural line applies verbatim and was added before the isFunctionLike fallback. Shape (1) an index signature typed any now reports 'index signature' instead of 'return type'; shape (2) a return any on a function type inside a type argument STAYS flagged as 'return type' - the function-like arm was NOT narrowed to an ancestry reading. The in/out decision for shape (2) is now stated in the file's SCOPE header, naming the predicate parent.type === node rather than any count. RE-MEASURE CONTRADICTS THE CARD: the card claimed 4 findings all correctly labelled; one of the four is a REAL shape-(1) site - skills/objectui/guides/plugin-development.md fence 236 carries [key: string]: any and was reported as 'return type' before this change, 'index signature' after. So this port fixed a live mislabel, not only a latent one. No baseline row changes were needed: that fence is UNMARKED, so the row is visible only under --measure and carries no KNOWN_BARE_ANY_EXAMPLES entry, and the marked population has zero bare-any findings. The gate's verdict is unchanged (exit 0 before and after). assignee was already set to baozhoutao by the PM dispatch pair; I did not touch it. gates_weakened: NONE - the same set of any positions is flagged, one of them under a correct label.",
  "tests": "ALL RUN AFTER THE FINAL COMMIT 9c47912993, tree clean. [gate] node scripts/check-skill-examples.mjs --self-test -> exit 0, '60 cases pass'. NOTE: before the scoped build it exited 2 = PRECONDITION NOT MET (the gate's documented could-not-run state) - read as NOT MEASURED, never as pass/fail; the scoped build (turbo run build $(node scripts/check-skill-examples.mjs --build-filter) --concurrency=2) ran under the shared verify lock, VERDICT command-exit 0, 29/29 tasks, held 207s waited 0s. [gate] node scripts/check-skill-examples.mjs -> exit 0: 'Semantic phase: 14 of 14 ts fence(s) judged, 0 failed', 'Bare `any`: 0 finding(s) across 0 selected fence(s)'. [pin] pnpm exec vitest run scripts/__tests__/check-skill-examples.test.ts -> exit 0, 113 passed (was 111; +2 are the new legs, both confirmed BY NAME in --reporter=verbose against a control pin name known present). [suite] pnpm exec vitest run scripts/__tests__/ -> exit 0, 158 files / 4639 tests passed, 2 skipped. [lint] pnpm run lint:root (the scan that owns scripts/) -> exit 0, 0 errors, 32 pre-existing warnings; ran over its FULL population in 17s so no narrowing argument is needed. Targeted eslint --format json on both changed files reports 2 files linted, 0 errors 0 warnings each. Type-aware linting is NOT enabled (no 'project' key anywhere in eslint.config.js), so this diff cannot move any untouched file's verdict. [typecheck] pnpm run type-check:scripts -> exit 0; CONTROL: tsc -p tsconfig.scripts.json --listFiles counts scripts/__tests__/check-skill-examples.test.ts = 1, so the green is about my file and not vacuous. [gate families, all exit 0] check:skill-examples, check:skill-eval-tokens, check:control-bytes, check:comment-mask-corpus, check:upstream-port-parity, check:new-line-citations, check:test-path-roots, check:shell-escape-residue, check:lint-rule-coverage. COVERAGE-ABSENT (reported, not counted as evidence): check:upstream-port-parity passes but does NOT track scripts/check-skill-examples.mjs - 0 occurrences of 'check-skill-examples' in its output; it tracks 11 ported files, none of them this one. [changeset] node scripts/check-changeset-presence.mjs -> exit 0, '0 of them published source of a package the release covers' => none owed. No skip-changeset label applied (that label is inert in this repo). [governed] node scripts/check-governed-queue-guard.mjs --test on both paths -> NOT GOVERNED. ABLATION - two legs, because shape (2) was ALREADY flagged before the fix so a 'is it a finding' pin would pass on broken code; BOTH legs assert the LABEL STRING. Each mutation proven on disk by anchored counts BEFORE being believed; each restore proven by git hash-object == the HEAD blob AND an empty git diff HEAD; script carried a trap on EXIT INT TERM running a RESTORE-FUNCTION (spelled in words here because GitHub's body sanitizer eats tag-shaped fragments, backticks included) with absolute paths from git rev-parse --show-toplevel. (A) delete the new index-signature arm, anchor count 1->0 on disk: self-test exit 1, got [\"2:return type\",\"4:return type\"] want [\"2:index signature\",\"4:return type\"]; vitest exit 1, 1 failed | 112 passed, the failing test being 'labels an `any` index signature `index signature`, not `return type`'. (B) the FORBIDDEN narrowing - read 'nested' as ancestry in the function-like arm, marker count 0->1 on disk: self-test exit 1, got [\"2:index signature\"] want [\"2:index signature\",\"4:return type\"] - the shape-(2) finding DISAPPEARED entirely; vitest exit 1, 1 failed | 112 passed, the failing test being 'keeps a return `any` on a function type inside a type argument as `return type`'. Leg B is the fence that matters: it shows the narrowing silently drops a real finding and now reds instead. HONEST NEGATIVE: my FIRST ablation-A attempt was a NO-OP and its own guard correctly declared the reading VOID - the anchor used grep -cE and the unescaped parentheses in the predicate made it an ERE matching nothing (expected 1->0, observed 0->0). Re-run with grep -cF. Reported rather than silently retried. RE-MEASURE, population named in words: (i) only the MARKED ts fences under skills/, i.e. the gated population - 14 selected ts fences, 0 bare-any findings, 0 stale baseline rows; (ii) EVERY candidate ts fence under skills/ marked or not - 121 selected ts fences out of 191 candidate fences of all kinds, 4 bare-any findings: data-integration.md:331 parameter `appConfig`, plugin-development.md:175 property `schema`, plugin-development.md:236 (return type -> index signature, THE FIX), plugin-development.md:438 property `schema`. EVERY ZERO CARRIES A CONTROL PROVEN NON-ZERO IN THE SAME COMMAND: the same analyze() reader reports 14 / 121 selected fences and 191 candidates beside each zero, and the label probe returns [\"index signature\"] and [\"return type\"] beside the empty [] for Record<string, any>. Control-character scan grep -naP over both changed files: no hits (exit 1), with a control proving grep -P works on the file (2305 hits for a class known present).",
  "mcp_calls": "0 - no MCP GitHub tool was used; every GitHub write went through the REST proxy with curl",
  "api_writes": "2 REST writes: POST /repos/objectstack-ai/objectui/pulls (draft PR 9510, HTTP 201) and POST /repos/objectstack-ai/objectui/issues/7653/comments (this report). Plus 2 git pushes: the empty routing-probe branch, then the fix commit 9c47912993. NO label write: none was owed (no changeset owed, Clause-2 is no, and this repo's skip-changeset label is inert by rule). PR body read back byte-for-byte after creation: stored == sent exactly, 8750 bytes both ways, delta 0, all 3 angle-bracket shapes and the session-URL footer intact.",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: scripts/check-skill-examples.mjs is an acknowledged port of objectstack's packages/spec/scripts/check-skill-examples.ts (its own SCOPE paragraph says 'ported from that file rather than re-derived') yet is NOT in check:upstream-port-parity's tracked set of 11 files, so upstream drift in it is not mechanically detected - this port is the evidence. Carrier: the next agent porting a change into this file. Not filed because it is a coverage observation, not a reproducible defect, a declared-contract violation, or a metadata trap; and adding a file to that ledger is a decision about the ledger's scope.",
    "noted, not filed: the three other --measure findings (data-integration.md:331 parameter `appConfig`, plugin-development.md:175 and :438 property `schema`) are correctly labelled and sit in unmarked fences - pre-existing guide debt for the per-fence marking sweep, untouched here. Carrier: whoever marks those fences.",
    "noted, not filed: this card's own probe table (2026-09-03) recorded '4 findings, all correctly labelled'; the second half was wrong, as plugin-development.md fence 236 was mislabelled the whole time. No card needed - the reading is superseded by this PR and is recorded on the card and in the PR body. Carrier: none."
  ]
}

---

Generated by Claude Code