Filed by the skills seat (session session_01DAcomhvR9kKizeYgg89Vo8) as a bare card: ⛔ not routed, not graded — triage grades (objectstack#17942 ruling ③, comment 5652544529; the filer attaches keywords). Derived from the dev report on #9418 (PR #9448, its Acceptance notes) — a class (b) finding: the guard's printed text contradicts the rule PR #9448 writes into AGENTS.md §9.
Dedupe keywords: check-governed-queue-guard, human merge IS the review record, leave the merge to the maintainer, ruling C, printed remedy, governed surface.
What
scripts/check-governed-queue-guard.mjs on origin/main efc1c9c narrates the pre-ruling-C landing shape: docblock at :48 / :86 / :318, the remedy strings at :772 (「…and leave the merge to the maintainer. A human merge…」) and :1163 (「a human merge IS the review record for a governed surface」, printed by --test — quoted verbatim at the top of PR #9448's body), and the self-test --test-prescribes-the-draft-remedy at :1707 that pins that text. After PR #9448, AGENTS.md §9 states ruling C (objectstack#17971, verbatim 「C. approve 后不管后续改动都由席位落地:」) — draft until an authorized APPROVED review (GOVERNED_APPROVERS, not dismissed), then the claiming seat lands — while the script AGENTS.md tells a seat to consult (「拿不准就直接问它」) still says draft-and-human-merge. The guard's decision already keys on that approval; only its narration is stale, and it fails safe (under-permits).
Shape (not ruled here)
The remedy strings, the docblock and the pinning assertion name the landed rule; no decision branch changes. Same file as #9404 (the references-tier port) ⇒ serial with it. The objectstack copy is filed as its own card there (same shape, scripts/pm/check-governed-queue-guard.mjs :52 / :1229).
Generated by Claude Code
Filed by the skills seat (session
session_01DAcomhvR9kKizeYgg89Vo8) as a bare card: ⛔ not routed, not graded — triage grades (objectstack#17942 ruling ③, comment 5652544529; the filer attaches keywords). Derived from the dev report on #9418 (PR #9448, its Acceptance notes) — a class (b) finding: the guard's printed text contradicts the rule PR #9448 writes into AGENTS.md §9.Dedupe keywords:
check-governed-queue-guard,human merge IS the review record,leave the merge to the maintainer,ruling C,printed remedy,governed surface.What
scripts/check-governed-queue-guard.mjsonorigin/mainefc1c9cnarrates the pre-ruling-C landing shape: docblock at :48 / :86 / :318, the remedy strings at :772 (「…and leave the merge to the maintainer. A human merge…」) and :1163 (「a human merge IS the review record for a governed surface」, printed by--test— quoted verbatim at the top of PR #9448's body), and the self-test--test-prescribes-the-draft-remedyat :1707 that pins that text. After PR #9448, AGENTS.md §9 states ruling C (objectstack#17971, verbatim 「C. approve 后不管后续改动都由席位落地:」) — draft until an authorized APPROVED review (GOVERNED_APPROVERS, not dismissed), then the claiming seat lands — while the script AGENTS.md tells a seat to consult (「拿不准就直接问它」) still says draft-and-human-merge. The guard's decision already keys on that approval; only its narration is stale, and it fails safe (under-permits).Shape (not ruled here)
The remedy strings, the docblock and the pinning assertion name the landed rule; no decision branch changes. Same file as #9404 (the references-tier port) ⇒ serial with it. The objectstack copy is filed as its own card there (same shape,
scripts/pm/check-governed-queue-guard.mjs:52 / :1229).Generated by Claude Code