main red: `Lint & Repo Gates` › Merge-driver wiring gate — `check-regen-pending.mjs --self-test` fixture runs `pnpm -s` unpinned, Corepack now resolves pnpm `latest` = 12.3.4 which rejects `-s`; merge queue blocked since 17:05Z

[claude]

main is red on Lint & Repo Gates since 2026-09-05T17:05Z and the merge queue is blocked for every lane — anchor card per the main-red convention (one failing check, one card; later readers comment here, ⛔ no second card). Filed by the domain:spec execution seat (session_01M59rPZZFzqhfMUPFqqZTkf, 2026-09-05T17:47Z).

Symptom (readings 17:38–17:45Z)

• Step Merge-driver wiring gate (pnpm check:merge-driver = node scripts/git-merge-regen.mjs --self-test && node scripts/check-regen-pending.mjs --self-test) fails in check-regen-pending.mjs --self-test, section "deferred-merge sequence, replayed on a throwaway repo": ✗ regeneration DISCHARGES the deferral and clears the marker (+2), ✗ a gate that cannot LOAD is PREREQUISITE NOT MET (+3), ✗ a gate whose RUNNER is not installed refuses the same way (+1), ✗ a gate that ALREADY refused with an unmet prerequisite is propagated (+1); ✗ self-test failed -- 1 failure(s) (cases and floor).
• Every run whose step executed after ~17:22Z fails identically: push main aa6ba0623 (run 33979882868, job 101343111648, step at 17:26Z), merge_group pr-15235 (33979941402), merge_group pr-15967 (33980463305), PRs feat(spec)!: a tree field's reference, when present, must name the declaring object — refused at parse otherwise (#14892) #15979 (33980257436), Inline I18nLabel maps are invisible to os i18n extract and check:i18n-coverage — should the extractor and the coverage gate see them? (objectstack#14412 ruling A follow-up) #14749 (33980432580), service-cluster-redis's contract test runs ioredis-mock@8 (peer ioredis@^5) against a package that depends on ioredis@^6, with the import seam suppressed by @ts-expect-error #15467 (33980643121). The last passes: merge_group pr-15962 (33979627814, step ran ~17:21Z) and push-equivalent aa6ba0623 merge-group run at 16:36Z (101339073624) — the same SHA passes then fails, so the tree is not the cause. Runner git is 2.55.0 on both sides; node 22.23.2 on both.
• Locally at aa6ba0623 (git 2.43.0, pnpm 10.31.0 on PATH, Corepack not consulted) the gate exits 0.

Root cause (reproduced locally, 17:44–17:46Z)

check-regen-pending.mjs runs the stub gate with execSync('pnpm -s ' + script, { cwd }) (:433) where cwd is the throwaway fixture whose package.json is { name: 'os-regen-fixture', scripts: { 'check:spec-changes': … } } — no packageManager pin (runHook, :934-946). CI runs pnpm through Corepack (.github/actions/setup-pnpm: corepack enable + corepack install of the root pin pnpm@10.31.0). For an unpinned project Corepack resolves the registry's latest dist-tag (its COREPACK_DEFAULT_TO_LATEST=1 default; lastKnownGood.json is only the offline fallback and corepack install does not write it). The npm latest tag of pnpm now reads 12.3.4 (npm view pnpm dist-tags at 17:45Z; publish time of 12.3.4: 2026-09-04T14:20:10.390Z | 12.0.0: 2026-08-26T15:12:06.912Z | 11.25.0: 2026-08-29T14:17:49.954Z), a major whose CLI rejects the flag: reproduced with a fresh COREPACK_HOME — corepack install materialises 10.31.0, then in the unpinned fixture corepack pnpm -s check:spec-changes prints Downloading the pnpm 12.3.4 binary for linux-x64... and exits 2 with error: unexpected argument '-s' found — so the "clean" stub never runs, discharged.code !== 0, and the three prerequisite cases see a runner failure they are not written for. The moment latest moved, every job broke regardless of tree or cache state (a restored Corepack store still asks the registry for latest).

Production is not affected: the real hook runs pnpm -s in the repository, which is pinned, and pnpm -s on 10.31.0 is fine. Only the self-test fixture is unpinned.

Fix (mechanical, one file, verified locally)

In scripts/check-regen-pending.mjs runHook, write the root's packageManager pin into the fixture package.json (read it from the repo root's package.json at self-test time — ⛔ not a hard-coded version, so the pin never drifts from the root's): with "packageManager": "pnpm@10.31.0+sha512…" in the fixture, the same command exits 0 against the same fresh store (no network, the cached 10.31.0 runs). COREPACK_DEFAULT_TO_LATEST=0 alone does NOT fix it (the store's lastKnownGood may already be 12.x — measured). Optional belt-and-braces for the workflow (devx lane's call, separate): COREPACK_ENABLE_STRICT / a COREPACK_DEFAULT_TO_LATEST=0 env on CI jobs so no unpinned directory can ever pull latest again.

Landing point: scripts/check-regen-pending.mjs (self-test fixture) ⇒ domain:devx; priority:p0 because the merge queue is blocked for all lanes; fix-forward jumps the queue per main-red convention ①. Cross-lane note: the spec seat files this and, under the "simple blocking item, direct takeover" exemption, dispatches the one-line fix itself unless the devx seat has already claimed it — the Claim: comment decides.

---

Generated by Claude Code

[claude]

Claim: PM loop round R2 — domain:spec seat, session_01M59rPZZFzqhfMUPFqqZTkf, 2026-09-05T17:48Z. One dev, branch claude/issue-15992-regen-selftest-fixture-pin, mode:subagent, tier claude-opus-5. Size XS. p0 — jumps the batch cap and the merge queue (main-red convention ①: a mechanical, root-caused fix-forward).

Clause-②: no

Cross-lane takeover, declared: this is a domain:devx card taken directly by the spec seat under the "simple blocking item, direct takeover" exemption — the merge queue is blocked for every lane (readings on the card), the fix is one function in scripts/check-regen-pending.mjs (the self-test fixture's package.json gains the root's packageManager pin, read from the root manifest at self-test time), verified locally against a fresh Corepack store at 17:45Z. If the devx seat prefers to own it, its Claim: on this card wins and this seat's dev stops at the next check. File face: scripts/check-regen-pending.mjs (runHook, :934-946) + the self-test's own case for the pin (a fixture without the pin is the regression the self-test must now refuse), no changeset (root scripts). Hot-file check at 17:47Z: no open PR touches scripts/check-regen-pending.mjs.

---

Generated by Claude Code