Repository navigation
check-half-states: a governed PR with an ACCEPT verdict but no needs-user-decision or no 「维护者速读」 comment is a half-state — one report-only row #15895
Description
Filed by the domain:skills seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve) as the patrol half of the maintainer's 2026-09-05 instruction (the rule half is the sibling card filed in the same minute). Self-triaged priority:p3, pm:blocked, pure code (scripts/pm/check-half-states.mjs).
Blocked-by: #15884
What this card does
One report-only row in check-half-states.mjs, judged over the open PRs the sweep already lists: a PR on a governed surface (derive from GOVERNED_SURFACES, never a hand list) that carries an ACCEPT verdict from a seat (a comment whose first line begins **ACCEPT**) and either lacks needs-user-decision or has no comment whose first line begins ## 维护者速读 → finding, with the remedy (add the label; post the brief). A merged or closed PR is out of the population. Self-test cases both directions, ablation, summary clause, counters. No new fetch class if the PR comment threads are already read for H46; otherwise declare the class and its cost per PR, as #15702 did.
Serial
check-half-states.mjs is in flight under #15884 (the release-rule row); this card is hard-serial behind that PR and also waits for the rule half to land so the shapes it reads are on main.
Generated by Claude Code
Activity
claude commentedon Sep 5, 2026
Claim: PM loop round 7 — unblocked when #15884 closed (PR #15901 MERGED 14:29Z); the patrol half of the maintainer's 2026-09-05 instruction: one report-only row in
check-half-states.mjs(a governed PR carrying an ACCEPT verdict but noneeds-user-decisionlabel or no 「维护者速读」 comment), bandstate, self-test in both directions, ablation, summary clause, counters. Seat session_019RfFHiRCSs3JXLK4cwcfox (os-steve); branchclaude/issue-15895-half-states-governed-pr-brief-row; size S; pure code underscripts/pm/⇒ in-seat review, then ready + queue. #15899 (same file) follows serially behind this flight.Generated by Claude Code
claude commentedon Sep 5, 2026
os-dev-report
{ "issue": 15895, "status": "done", "branch": "claude/issue-15895-half-states-governed-pr-brief-row", "pr": "https://github.com/objectstack-ai/objectstack/pull/15956", "premise_still_valid": true, "summary": "H48 added to scripts/pm/check-half-states.mjs as one report-only row, band state, one file, nothing else moved. It reads three things off a governed OPEN PR: an ACCEPT verdict comment, the needs-user-decision label, and a 维护者速读 heading comment; it fires on verdict AND (no label OR no brief), names the missing half or halves with its remedy, and treats no-verdict as CLEAN. Population is the open /pulls listing narrowed by the IMPORTED governedPathsIn matcher over the changed-file pages H36 already holds (drafts IN, merged/closed OUT); the PR comment thread is a genuinely new fetch class, declared as H36/H43 declare theirs and bought one paged thread per governed open PR only, cached in prCommentCache. The rule half landed on main as 97afcb5ccc while this ran, so the premise is now on main and not merely in flight. Two of the four PRs the brief named (15908, 15641) merged mid-session at 15:46Z and 15:38Z and correctly left the population.", "tests": "ALL captured before any pipe. (1) node scripts/pm/check-half-states.mjs --self-test: 2341 on the base -> 2414 after, exit 0, \"✓ check-half-states self-test: 2414 cases pass.\"; the two case-NAME sets were diffed, 73 added and 0 dropped. (2) node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derives 32 families for this file list; all 32 run, 0 non-zero; --ran reconciles \"✓ dispatch-gates --ran: 32 derived famil(ies) accounted for — 32 run, 0 NOT-MEASURED.\" Re-derived and RE-RUN on the final head ea5b4f3912 after merging origin/main: identical 32-family set, 34 commands (the 32 plus the self-test and the governed --test) all exit 0. (3) node scripts/pm/check-governed-merges.mjs --test scripts/pm/check-half-states.mjs -> exit 0, \"✅ NOT governed\". (4) Whole-repo pnpm lint through bash scripts/pm/os-verify-lock.sh (slot issue-15895), twice, ending on the final head: \"VERDICT command-exit 0 · held the lock 82s · waited 0s\" (shared-box seconds, per the lock own disclaimer). (5) Self-tests of every script that imports or pins this file also green: check-clause2-carriers, ci-failure, check-governed-merges, dispatch-gates, check-test-completeness, check-partof-closing-keyword. ABLATION, three legs. There is NO build/dist leg for this artifact: the script is executed directly from source by node, no package exports and no dist on the resolution path, so nothing to rebuild — stated rather than skipped silently. Each leg asserted the working tree matched the HEAD blob BEFORE mutating, then proved the mutation ON DISK by counting the exact anchor text in both directions (anchor occurrences before/after plus injected-marker count, refusing to continue on a zero-hit edit) and by a byte-length change; the implementation was COMMITTED first (c97753f438) so HEAD was a real restore point. (A) predicate always declines -> 19 red, every one an H48 firing case, no clean-direction case moved. (B) the findings.push removed -> 3 red: the H48 emitted-set case and the registry invariant from BOTH sides (\"no band names a family the sweep never emits\"). (C) the summary clause deleted -> 11 red: anchor uniqueness, declared-order-is-rendered-order, the per-window PRESENT control and every H48 clause case. Restore proved on all three legs by blob hash equality (RESTORED_BLOB=ea5a25afd032d963db10b3c50d3285487ddbb2d1 = the HEAD blob) and by git diff HEAD --quiet exit 0, with a trap on EXIT/INT/TERM using an absolute repo-root path as the crash-path convenience only. LIVE SWEEPS, read-only, before and after: objectstack 297 -> 295 findings with 0 H48 rows and \"Maintainer briefs (H48): 11 of 11 governed open PRs judged\"; objectui (PM_SWEEP_REPO) 170 -> 166 findings with 0 H48 rows and \"1 of 1 governed open PRs judged\". Other families moved a little because the live board moved between runs, not because of this diff. THE ZERO IS MEASURED, NOT QUIET: reading the ten governed open PRs one by one with the shipped predicate over their real threads, 15924 and 15923 carry the COMPLETE handoff (verdict AND label AND brief) and the row read all three and passed them — a positive reading; six carry no verdict (15946, 15837, 15814, 15626, 15612, 15427) and adding one synthetic ACCEPT comment makes ALL SIX fire, so the row demonstrably fires on this board; 15885 and 15382 are the 代读 shape and stay clean either way; objectui 7749 likewise carries all three and reads clean. CI on the PR head ea5b4f391281bcb94e2f49f4c00a55b5ff13c31d polled in the foreground to completion: 35 check runs, ALL completed, 24 success and 11 skipped, ZERO failures. Required contexts: Lint & Repo Gates success, TypeScript Type Check success, Test Core success, Dogfood Regression Gate success, Build Core skipped, Temporal Conformance skipped. \"Live half-state sweep\" also ran green on the modified file, and \"Check Changeset\" skipped on the skip-changeset label.", "mcp_calls": "1 — one mcp__github__issue_read to read the card. Every other GitHub read and write (PR listings, changed-file pages, comment threads, PR creation, the additive label write and its read-back, the check-run polling) went through repo-scoped REST with curl, which answered 200 on a first probe in this container.", "open_questions": [], "out_of_scope_findings": [ "NOT FILED, by dispatch: a governed open PR carrying needs-user-decision with NO seat verdict behind it (the 代读 shape) is a different assertion from the one H48 patrols — the label is a claim on the maintainer attention, placed with no verdict standing behind it. Live today on PR 15382 and PR 15885. The dispatch ruled this out of scope for this card and asked for the judgement in the PR body rather than a card, so it is stated there and left for the PM to file or decline. I think it deserves a card." ] }Generated by Claude Code