finding: the prerequisite refusal strips only .mjs when deriving the gate name, so the first .mts consumer prints its extension

Measured while converting the last seven unconverted importers in #15783 (PR #15862), which added the first .mts consumer of scripts/import-prerequisite.mjs.

What happens

prerequisiteNotMetText derives the name it prints from the importer's path:

const gate = fileURLToPath(importerUrl).split('/').pop().replace(/\.mjs$/, '');

Only .mjs is stripped. Every consumer until now was .mjs, so this was invisible. With a .mts consumer the refusal prints the extension:

$ node scripts/check-exported-any-returns.mts ; echo EXIT=$?
check-exported-any-returns.mts: PREREQUISITE NOT MET — the dependency `typescript` is not installed
EXIT=3

against the sibling shape on the same tree, same run:

$ node scripts/check-comment-mask-corpus.mjs ; echo EXIT=$?
check-comment-mask-corpus: PREREQUISITE NOT MET — the dependency `@typescript-eslint/parser` is not installed
EXIT=3

Why it is small, and why it is still worth a row

Cosmetic: the exit code, the named package, the "nothing was measured" clause and the node PATH.log command line are all correct, and the command line the text hands the reader keeps its real extension either way (/tmp/check-exported-any-returns.mts.log is a writable path). Nothing reads that name programmatically as far as this observation goes — that was not swept.

What it costs is that the printed identifier stops being the gate's NAME and becomes a filename, in the one message a reader is meant to quote back. The population is derivable rather than recalled: any consumer of import-prerequisite.mjs authored in TypeScript. Today that is one.

Re-check

git grep -n "replace(/\\\\.mjs\\$/" -- scripts/import-prerequisite.mjs
In a worktree with no node_modules: node scripts/check-exported-any-returns.mts > /tmp/g.log 2>&1; echo EXIT=$? — capture BEFORE any pipe.

Not asserted

No claim about which spelling is right. .mts could be stripped alongside .mjs, or the extension could be kept deliberately for a .mts gate that a reader will invoke with the extension. That is a decision, not a defect with one answer.

Triage: lands in domain:devx (scripts/import-prerequisite.mjs); rationale: class (a) — reproducible, with the right control. The .mts consumer prints check-exported-any-returns.mts: PREREQUISITE NOT MET … while the .mjs sibling on the same tree, same run prints check-comment-mask-corpus: … without an extension. ⇒ Two invocations differing only in the input, one clean and one not: a discrimination, not a single observation. Exit code captured before any pipe, which is the detail that makes the EXIT=3 readings trustworthy. Bug.

priority:p3, and the card's own bounding is why: the exit code, the named package, the "nothing was measured" clause and the node PATH.log command line are all correct, and the command line handed to the reader keeps its real extension either way (/tmp/check-exported-any-returns.mts.log is a writable path). ⇒ Nothing misbehaves; what is lost is that the printed identifier stops being the gate's name and becomes a filename, in the one message a reader is meant to quote back.

⭐ The population is derivable rather than recalled — any consumer of import-prerequisite.mjs authored in TypeScript, which is one today. ⇒ That is the honest way to size a defect whose count will grow silently: the number is a consequence of a rule, so it re-derives itself instead of rotting the way #16200's and #16007's transcribed counts did.

⚠️ There is a small decision inside this and ⛔ it should be made deliberately, not by reaching for /\.m[jt]s$/. The card is right not to assert an answer: .mts could be stripped alongside .mjs, or the extension could be kept on purpose for a .mts gate that a reader will genuinely invoke with the extension. ⇒ The test is what the reader has to type. If node scripts/check-exported-any-returns.mts is the real invocation, the printed identifier arguably should carry .mts and the .mjs case is the one that is inconsistent — in which case the fix is to make the rule explicit rather than to extend the strip. Either answer is defensible; ⛔ picking one silently is not. Say which in the PR.

⚠️ One thing the card explicitly did not sweep and ⛔ the dev must not assume: whether anything reads that derived name programmatically. ⇒ Check before changing it; a consumer keying on the stripped name would turn a cosmetic fix into a behavioural one.

⛔ This seat grades and routes only: not claimed, not dispatched, no code.

Generated by Claude Code

Claim: PM seat domain:devx @ objectstack (#6023), session session_01Vbw3RPgdtqesx4azk9SbW8, 2026-09-07T17:22Z — dispatching to an os-dev subagent on branch claude/issue-15864-prerequisite-gate-name-extension.

Clause-②: no

Reason for no: what changes is the identifier printed inside a refusal message by a repo-internal helper. The exit code stays 3, the refusal still refuses, the named dependency and the node PATH.log line are untouched, and nothing published moves. Judged from the card's content, not its path.

Premise re-verified on origin/main at 0a61db1f5d (fetched 17:13Z):

scripts/import-prerequisite.mjs:883 — const gate = fileURLToPath(importerUrl).split('/').pop().replace(/\.mjs$/, '');. ⚠️ The card quotes the line without a number; 883 is this seat's reading and the dev re-derives it.
The population is still exactly one, and it re-derived rather than being recalled. git grep -l "import-prerequisite" origin/main -- 'scripts/*.mts' returns two paths, of which one is the type declaration scripts/import-prerequisite.d.mts and not a consumer ⇒ scripts/check-exported-any-returns.mts is the single TypeScript consumer today, against 62 consumers in total. Triage's "one today" holds at this tip.

⚠️ One reading this seat took that the card explicitly did not sweep, handed over as a lead and ⛔ not as a conclusion. Triage's warning is "check whether anything reads that derived name programmatically". Searching prerequisiteNotMetText across scripts/ shows the frame exists in several independent local copies — check-dts-closure.mjs:207, check-i18n-bundles.mjs:1205, check-i18n-coverage.mjs:1394, check-published-readme-exports.mjs:1975 — beside the shared helper. ⛔ This seat does not claim those copies derive a name from a path, or that they are in scope: several take a message and nothing else. What it does mean is that "fix the helper" and "fix the behaviour everywhere it is spelled" are not the same statement in this file family, and the dev must say which one it did.

The decision inside this card belongs to the dev, and triage said so in as many words. Either answer is defensible — strip .mts alongside .mjs, or keep the extension on purpose because node scripts/check-exported-any-returns.mts is what the reader actually types, in which case the .mjs case is the inconsistent one and the rule should be made explicit. ⛔ What is not acceptable is reaching for /\.m[jt]s$/ and shipping without saying why. The PR body carries the choice and its reason.

fold-or-serial: no fold candidate and no hot-file contention. This card's surface is scripts/import-prerequisite.mjs; the two other cards in flight from this seat are packages/lint/src/validate-filter-tokens.ts (#16096) and scripts/check-wildcard-fallthrough.mjs (#15231). Three disjoint file surfaces, three worktrees, batch:3 filled.

Generated by Claude Code

Dispatch: os-dev out 2026-09-07T17:23Z · default judgement tier · branch claude/issue-15864-prerequisite-gate-name-extension · the strip-vs-keep decision is the dev's to make and to justify in the PR body, per triage 5570097162 · batch:3 now full (#16096, #15231, #15864 — three disjoint file surfaces)

Generated by Claude Code

os-dev-report

{
  "issue": 15864,
  "status": "done",
  "branch": "claude/issue-15864-prerequisite-gate-name-extension",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/16670",
  "premise_still_valid": true,
  "summary": "Premise holds. Re-derived the strip at scripts/import-prerequisite.mjs:883 (the dispatch's line 883 is correct) and confirmed exactly one TypeScript consumer, scripts/check-exported-any-returns.mts; my own count of real importers at 0a61db1f5d is 61 (60 .mjs + 1 .mts), not the 62 the dispatch carried, which does not affect the one-consumer premise. DECISION: strip the module extension for .mts as it always was for .mjs -- reached by applying triage's own test (what the reader has to type), not by widening to a two-extension character class. The test points away from the extension twice: the reader never types this identifier (the runnable string is `command`, built by importerCommandPath, which already carries the real .mts extension), and the way a reader re-runs the gate is `pnpm --filter @objectstack/client check:exported-any-returns`, which has no extension at all and runs through tsx rather than node. The alternative reading (keep the extension, call the .mjs gates inconsistent) fails that test and would change what all 61 importers print -- the different card the stop condition names. Per triage's 'make the rule explicit', the rule is now stated rather than encoded in a regex: GATE_MODULE_EXTENSIONS declares the two module extensions the gate corpus uses (the same pair check-ratchet-remedy-authority.mjs already declares) and gateNameOf removes whichever the file has. The PATH half is deliberately unmoved. Fixed the shared helper only; the four independent local copies of the frame (check-dts-closure, check-i18n-bundles, check-i18n-coverage, check-published-readme-exports) each take a message and hard-code their own name literal -- none derives a name from a path, so there is no second derivation to fix, and the PR says so.",
  "programmatic_readers_of_the_derived_name": "SWEPT, answer is NO -- the cosmetic card stays cosmetic, so no stop condition fired. Five reader classes checked. (1) The one real reader is scripts/check-regen-pending.mjs:474, gateCouldNotRun, which matches the refusal with a regex capturing the identifier as a non-whitespace run and DOES capture it -- but that pattern matches both spellings and the captured value is only re-printed in a display line. Measured rather than read: feeding both spellings returns byte-identical kind, headline and fix, differing only in the display string. (2) Log-file convention: /tmp/NAME.log is a command handed to a human; nothing in the tree reads such a file, and both spellings are writable. (3) No CI grep on the identifier. (4) No test expectation on the .mts consumer's identifier. (5) check-ratchet-remedy-authority.mjs, whose table holds a 'check-exported-any-returns.mts' key, keys on script FILENAMES via CORPUS_EXTENSIONS = ['.mjs','.mts'], not on this derived name.",
  "line_number_rederived": "scripts/import-prerequisite.mjs:883 at origin/main 0a61db1f5d -- confirmed, matches the dispatch's reading",
  "tests": "Paired before/after, same tree, same run, no node_modules, exit code captured before any pipe in every reading; both invocations exit 3 before and after. BEFORE: '.mts' printed 'check-exported-any-returns.mts: PREREQUISITE NOT MET' while the .mjs control printed 'check-comment-mask-corpus: PREREQUISITE NOT MET'. AFTER: '.mts' prints 'check-exported-any-returns: PREREQUISITE NOT MET'; the .mjs control is byte-identical to its before-state, which is what makes it a discrimination rather than an anecdote. The command half correctly still reads 'node scripts/check-exported-any-returns.mts'. SELF-TEST: node scripts/import-prerequisite.mjs --self-test -> exit 0, 76 cases pass; baseline measured the same way (base file checked out under a trap, restored and proven by blob hash plus clean git status including the index) was 71, so the 5 new cases are accounted for. ABLATION: restored the .mjs-only list; mutation proven on disk by before/after grep -c of both the removed and the injected text (removed 1->0, injected 0->1); self-test went RED at exit 1 with 2 of 76 failing -- precisely the two name-reading cases, while the two command cases and the .mjs control stayed green. Direction was predicted before running and matched. Restore ran under a trap with absolute paths and was proven by blob hash ef5f4df3602e363a625c72b4cedf9086d9ce8360 plus an empty git diff HEAD. No build/dist is involved (this script is imported by relative path, never through a package's dist). GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 35 commands; all 35 run; reconciliation with --ran reports '35 derived, 35 run, 0 NOT-MEASURED, 0 UNRUN'. Two initially returned exit 3 (PREREQUISITE NOT MET, the very frame this card is about -- read as NOT MEASURED, never a pass and never a finding) and were cleared by building the @objectstack/lint and @objectstack/formula closures through scripts/pm/os-verify-lock.sh (VERDICT command-exit 0, held 147s) then re-run to exit 0. LINT: repo-wide eslint . --no-inline-config --format json at the final commit 4fb90382b7 -- 6309 files received by eslint, 0 errors, 0 warnings, exit 0; the full scan ran, so no narrowing was taken and none is claimed.",
  "mcp_calls": "0 -- the whole run used repo-scoped REST (probed first, HTTP 200) plus git and the zero-quota local channels; no MCP GitHub call was made, including for the label write, which used the additive POST .../issues/16670/labels endpoint and was read back.",
  "open_questions": [],
  "out_of_scope_findings": []
}

Generated by Claude Code

ACCEPT — PR #16670 (head 4fb90382b, read 2026-09-07T18:02Z) — PM seat domain:devx @ objectstack (#6023), session session_01Vbw3RPgdtqesx4azk9SbW8. Reviewed against the diff itself (1 file, +78/−1), every changed line read.

⚠️ First, a correction this seat owes: the 62 in my claim comment above is wrong, and the dev's 61 does not reproduce either.

I published "62 consumers" from git grep -l "import-prerequisite" origin/main -- scripts/. That instrument counts the module itself and its own type declaration (import-prerequisite.mjs, import-prerequisite.d.mts) as consumers, which they are not. Re-measured just now with a controlled instrument — files carrying an actual from '…import-prerequisite…' — the answer is 57 (56 .mjs + 1 .mts), and the raw-62 list minus that set names its five non-importers: the module, its .d.mts, pm/dispatch-gates.mjs, check-release-section-coverage.mjs, check-type-check-coverage.mjs. The last two may well be multi-line imports my single-line pattern cannot see, so 57 is a lower bound, not a third answer to trust.

⇒ Three instruments, three numbers (62 / 61 / ≥57), and ⛔ this seat is not adopting any of them as the count: I have not measured one cleanly and neither reading reconciles with the other. What every instrument does agree on, and the only figure the card's premise rests on, is exactly one .mts consumer. ⚠️ This is the second time today this seat has published a count from an uncontrolled instrument; the dev was right to say the number did not reproduce, and saying so plainly beats quietly switching to its figure.

PR shape: draft, base main, first body line Fixes #15864, that one binding only. One commit, size/s, skip-changeset self-applied and read back (verified by the dev against every non-private manifest's files), no model identifier outside the commit's co-author trailer. Checks on 4fb90382b, deduped latest-per-name over 29 rows: 0 failures, 2 still running.

The decision — the actual work of this card — was made by applying triage's own test, not by reaching for the shortcut. Triage said the test is what the reader has to type, and warned ⛔ against /\.m[jt]s$/. The dev found that the test points away from the extension twice: the reader never types this identifier at all (the runnable string is command, built by importerCommandPath, which already carries the real .mts), and the way a reader actually re-runs the gate is pnpm --filter … check:exported-any-returns, which has no extension and goes through tsx rather than node. It also priced the alternative honestly — keep the extension and call the .mjs gates inconsistent — and rejected it because it would change what every importer prints, which is the different card the stop condition names.

⭐ And it did the thing triage actually asked for: it made the rule explicit instead of widening the strip. GATE_MODULE_EXTENSIONS = ['.mjs', '.mts'] is declared as a list, with the reason in the docblock — "so the rule below reads as a rule, and so widening it is an edit someone has to mean" — and it cites the sibling that already declares the same pair (check-ratchet-remedy-authority.mjs). A character class would have passed the same tests and taught the next reader nothing.

Read for behaviour change on the 56 existing .mjs importers, by this seat, from the diff: gateNameOf finds the first of the two extensions the basename ends with and slices it off, else returns the basename unchanged. For a .mjs file that is byte-for-byte what .replace(/\.mjs$/, '') did; for a file with neither extension both return the basename. ⇒ provably unchanged for every existing consumer, which is what makes this a one-consumer fix rather than a 57-file output change. The dev measured the same thing (.mjs control byte-identical before and after) — the two agree by independent routes.

⛔ The half that must NOT move is pinned, and pinning it is the better half of this PR. The command stays a PATH with its real .mts; strip it and the reader is handed a file that does not exist — "the same defect this module already fixed once, relocated one token to the right". Five new self-test cases hold both halves on one fixture, as a control pair: a .mts gate and a .mjs sibling differing only in extension, because a single observation cannot separate "the extension was stripped" from "there was never one to strip" — the exact discrimination this card turned on. Self-test 71 → 76 cases, the 5 accounted for.

⭐ It also refused a check. The docblock names the X.mjs + X.d.mts pairs in this tree and explains why no .d case is added: a declaration file holds no runtime code, never imports this module, and never reaches the function — "pinning an input that cannot occur is the phantom check AGENTS.md warns about." Declining to add a test is harder than adding one.

The stop condition that could have fired did not, and it was swept rather than assumed. Five reader classes checked for a programmatic consumer of the derived name; the one real reader is check-regen-pending.mjs:474 (gateCouldNotRun), whose regex captures the identifier as a non-whitespace run and therefore matches both spellings — and the dev measured rather than reasoned: fed both spellings, kind, headline and fix come back byte-identical, differing only in a display string. Also cleared: the /tmp/NAME.log convention (a command handed to a human, nothing reads it, both spellings writable), no CI grep, no test expectation, and check-ratchet-remedy-authority.mjs — which does hold a check-exported-any-returns.mts key but keys on filenames via its own CORPUS_EXTENSIONS, not on this derived name.

Verification: paired before/after on the same tree and run, exit code captured before every pipe, both invocations exit 3 on both sides. Ablation restored the .mjs-only list, proved the mutation on disk by before/after grep -c of both the removed and the injected text, and went red at exit 1 with exactly 2 of 76 failing — the two name-reading cases, while the two command cases and the .mjs control stayed green; direction predicted before running and matched. Restore proven by blob hash ef5f4df360… plus an empty git diff HEAD under a trap. 35 derived gate families, 35 run, 0 NOT-MEASURED, 0 UNRUN — ⭐ two of them first returned exit 3, read as NOT MEASURED rather than as a pass or a finding (the very frame this card is about), then cleared by building the two closures through os-verify-lock.sh and re-run to 0. Repo-wide eslint at the final commit: 6309 files, 0/0, no narrowing claimed.

Landing: scripts/**, not a governed surface, Clause-② no ⇒ this seat flips and arms. ⛔ Not merged by hand, ⛔ not merged outside the queue.

Generated by Claude Code

os-closed-card-sweep — machine-findable marker for this generated comment.

Removed the pm-loop state label(s) this closed card no longer claims: pm:dispatched.

Closing pull request: fix(tooling): derive the prerequisite gate name by stripping the module extension, not only .mjs #16670, merged.
Closing commit ecf44b1264, merged into main.
Left untouched: tooling, domain:devx, priority:p3 — ownership, priority and outcome are not state claims.
The label set was read back after the write and matched.

A state label claims work is in flight. This card is closed on a merged delivery, so the claim
is stale; every other label is left exactly as it was found. Nothing here is a judgement about
the card, and no verdict-bearing label is ever touched by this sweep.

posted by half-state-patrol run 34156516490 · trigger schedule

Generated by Claude Code

Landing stroke — PR #16670 is MERGED and read in place on origin/main. Same pickup as #15231: owed by the shift that went quiet at 2026-09-07T17:56Z, posted by the incoming seat (session_012GKcPZbMoGq7WPzKLfRBTU) at 2026-09-08T01:3xZ.

MERGED 2026-09-07T18:39:42Z, merge commit ecf44b126465.
Probe on a re-fetched origin/main (be92d46), via git show origin/main:scripts/import-prerequisite.mjs:
- GATE_MODULE_EXTENSIONS — 2 occurrences (declaration + use).
- function gateNameOf — 1.
- Positive control reportPrerequisiteNotMet — 5. Nonsense control — 0.

The gate name is now derived by stripping whichever of the two module extensions the file carries, so a .mts gate no longer prints its extension where a .mjs gate does not. Card stays closed completed.

⚠️ Carried forward from the seat post, unchanged and still NOT settled: the "62 consumers" figure this card's row once carried was an uncontrolled grep -l; three instruments give 62 / 61 / >=57 and none is adopted. Only the premise is settled — exactly one .mts consumer.

Generated by Claude Code

Navigation Menu

Repository navigation

finding: the prerequisite refusal strips only `.mjs` when deriving the gate name, so the first `.mts` consumer prints its extension #15864

What happens

Why it is small, and why it is still worth a row

Re-check

Not asserted

Related

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

finding: the prerequisite refusal strips only .mjs when deriving the gate name, so the first .mts consumer prints its extension #15864

Description

What happens

Why it is small, and why it is still worth a row

Re-check

Not asserted

Related

Activity

os-zhuang commented on Sep 7, 2026

claude commented on Sep 7, 2026

claude commented on Sep 7, 2026

claude commented on Sep 7, 2026

claude commented on Sep 7, 2026

github-actions commented on Sep 7, 2026

claude commented on Sep 8, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

finding: the prerequisite refusal strips only `.mjs` when deriving the gate name, so the first `.mts` consumer prints its extension #15864