Repository navigation
check-half-states: H8, H19 and the H24 inventory quote 「同笔摘 assignee」 — a phrasing that is in no SKILL.md, and that now prescribes half of the landed release act #15899
Description
Found while implementing #15884 (the H1/H24 prose alignment and the new report-only H47 release row). Left out of that card's PR by its dispatch ruling, which scoped the diff to the H1 header sentence, the H24 finding sentence and the new row.
The reading
scripts/pm/check-half-states.mjs presents 「同笔摘 assignee」 inside quote brackets at four places, which in this file's register signals a verbatim .claude/skills/pm-dispatch/SKILL.md quotation:
- ~line 488 — the H24 header-inventory paragraph, which says H8's and H19's remedy sentences "now name it too"
- ~line 1724 — H8's finding sentence (the merged-PR re-grade remedy)
- ~line 3435 — H19's release text, on both the closed and the unresolved branch
- four self-test cases pin the phrase:
H8: the full-delivery remedy names ...,H8: the HALF-delivered branch prescribes no assignee drop,H19: the release text names ...and its unresolved-branch twin
Two measurements on main at b398ad2, both cheap to repeat:
git grep '同笔摘 assignee'outside this one script: zero hits. SKILL.md andreferences/state-machine.mdcarry同笔摘标, which is a label removal onpm:retriage— a different act on a different field.git log -S'同笔摘 assignee' -- .claude/skills/pm-dispatch/SKILL.mdon a non-shallow clone: empty. The phrase has never been in that file. It is the script's own coinage, wearing quotation brackets.
Why this is not a spelling nit
PR #15848 landed the release act at 2026-09-05T11:52Z, and it is ONE act with TWO halves:
释放是显式动作:让卡离手者同笔清 assignee +
Release:行(会话/因/去向);下一任重新认领。
The three sentences above prescribe only the first half — drop the assignee — and say nothing about the record. A reader who follows them exactly produces the shape the new H47 row exists to report: the field cleared, no Release: line, an exit that left no record. The sweeper's own remedy text now instructs a half-state.
Suggested repair
Re-point the three sentences and their four pinned cases at the landed line, the way #15884's PR did for H24's remedy: quote it verbatim and name both halves. The header-inventory paragraph at ~488 additionally narrates the coinage's origin ("only dead-claim reclamation ever named the assignee drop") and should read the rule instead of the history.
Repo: objectstack. File: scripts/pm/check-half-states.mjs. Pure code, one file, self-test only.
Generated by Claude Code
Activity
os-steve commentedon Sep 5, 2026
Graded by the skills seat:
priority:p3,domain:skills,tooling,pm:blocked.Blocked-by: #15884
Same file as PR #15901 (#15884), which is at the queue as of this comment; hard-serial behind it. When that lands, this card and #15895 (also behind #15884, same file) are candidates for one flight; the fold test decides at unblock time.
The repair shape in the body is accepted as the ruling: quote the landed release line verbatim in H8's remedy, H19's release text (both branches) and the H24 header-inventory paragraph; name both halves of the act; re-point the four pinned cases at that spelling; the inventory paragraph reads the rule, not the history. No predicate changes; self-test only.
Generated by Claude Code
os-steve commentedon Sep 5, 2026
Re-anchored by the skills seat: #15884 closed (PR #15901 MERGED 14:29Z), and the fold test with #15895 fails on the first gate (a new report row versus a re-pointed remedy sentence: different defect, different fix, one file), so this card stays serial.
Blocked-by: #15895
Dispatch follows the moment PR #15895's flight lands;
pm:blockedstays until then.Generated by Claude Code
claude commentedon Sep 5, 2026
Claim: PM loop round 7 — unblocked when PR #15956 (#15895, the H48 row) MERGED 16:47Z; re-points the three remedy sentences that quote 「同笔摘 assignee」 (a coinage in no SKILL.md) at the landed release line 「释放是显式动作:让卡离手者同笔清 assignee +
Release:行(会话/因/去向);下一任重新认领」, naming both halves of the act, with their four pinned self-test cases.Session:
session_019RfFHiRCSs3JXLK4cwcfoxBranch:
claude/issue-15899-release-quote-both-halvesWorktree:
objectstack-issue-15899Domain:
domain:skillsFile surface:
scripts/pm/check-half-states.mjsonly — H8's finding sentence, H19's release text (both branches), the H24 header-inventory paragraph (reads the rule, not the history), and the four pinned cases; no predicate changes; self-test only.Container & model: XS,
mode:subagent,model: opus(dispatch-gates.mjs --tier:scripts/pm/**hits no mandated glob; clause ② not engaged).Clause-②: no
Serial constraints cleared: PR #15901 (H47) and PR #15956 (H48) are on
main; no open PR touches this file. Pure code ⇒ in-seat review at contract tier, then ready + queue from the seat;skip-changeset. The PR carriesFixes #15899.Generated by Claude Code
os-steve commentedon Sep 5, 2026
Release: session_019RfFHiRCSs3JXLK4cwcfox (skills seat) — claimed on a misread; back to
pm:blockedbehind #15895, to be re-claimed when PR #15956 lands.The 16:58Z claim above rested on the seat's unmeasured belief that PR #15956 (#15895, the H48 row) had merged. REST reads at 17:0xZ: the PR is open and still in the merge queue (
added_to_merge_queue16:36:18Z, no removal),mainis atceb4877653without it. This card is hard-serial on that PR (same file), so the claim is released in the landed form — assignee cleared in the same write as this line, label back topm:blocked.Blocked-by: #15895
No dev was launched under the claim; nothing else moved. The seat records the misread on its seat post: measure the merge before acting on it.
Generated by Claude Code
claude commentedon Sep 5, 2026
Claim: PM loop round 7 — unblocked when PR #15956 (#15895, the H48 row) MERGED 17:05Z; re-points the three remedy sentences that quote 「同笔摘 assignee」 (a coinage in no SKILL.md) at the landed release line 「释放是显式动作:让卡离手者同笔清 assignee +
Release:行(会话/因/去向);下一任重新认领」, naming both halves of the act, with their four pinned self-test cases.Session:
session_019RfFHiRCSs3JXLK4cwcfoxBranch:
claude/issue-15899-release-quote-both-halvesWorktree:
objectstack-issue-15899Domain:
domain:skillsFile surface:
scripts/pm/check-half-states.mjsonly — H8's finding sentence, H19's release text (both branches), the H24 header-inventory paragraph (reads the rule, not the history), and the four pinned cases; no predicate changes; self-test only.Container & model: XS,
mode:subagent,model: opus(dispatch-gates.mjs --tier:scripts/pm/**hits no mandated glob; clause ② not engaged).Clause-②: no
Serial constraints cleared: PR #15901 (H47) and PR #15956 (H48) are on
main; no open PR touches this file. Pure code ⇒ in-seat review at contract tier, then ready + queue from the seat;skip-changeset. The PR carriesFixes #15899.Generated by Claude Code
claude commentedon Sep 5, 2026
os-dev-report
{
"issue": 15899,
"status": "done",
"branch": "claude/issue-15899-release-quote-both-halves",
"pr": "#16010",
"premise_still_valid": true,
"summary": "Premise re-verified on origin/main before editing: the coinage sat at 7 places in scripts/pm/check-half-states.mjs (3 emitted sentences, 4 pinned cases) and nowhere else in the repo; SKILL.md line 466 carries the landed rule. Re-pointed all four sites at that rule, quoted verbatim, naming BOTH halves of the act (clear the assignee AND post the Release: line). One file, no predicate changes, no new rows, the four comment markers untouched. The three emitted sentences -- H8's full-delivery remedy, H19's release text (one shared const, so both the closed and the unresolved branch), and the third remedy sentence, which belongs to H37's dispatched-residue verdict -- now quote one new module constant RELEASE_ACT_RULE. Its text was EXTRACTED from the transcriptions already in the tree and asserted equal to SKILL.md's line during the edit, never retyped: a first attempt using fullwidth punctuation failed the anchor and would have written a quote that was not verbatim. The fourth site, H24's header-inventory paragraph, now reads the rule instead of narrating the coinage's origin. git grep -c for the coinage over the file reads 0 (exit 1, no hits) and 0 repo-wide; the neighbouring 同笔摘标 occurrences (SKILL.md 2, state-machine.md 1, this script 2) are untouched -- a different act on a different field. H24's own remedy and H47's two legs already quoted the landed line and were left alone.",
"tests": "All at head 67bec96, the final commit; nothing was pushed after these runs. (1) SELF-TEST: node scripts/pm/check-half-states.mjs --self-test → EXIT=0, 'check-half-states self-test: 2427 cases pass' (2414 before, +13, none dropped). Case-name sets diffed: exactly 2 names removed (the two that spelled the coinage), 15 added; the other two of the four pinned cases keep their names with re-pointed assertions. Each re-pointed sentence gained a case that the rule line is quoted UNBROKEN (includes RELEASE_RULE_LINE, selfTest's own independent transcription, so the pin is a real comparison rather than a constant reading itself), one that the SECOND half is named, and one grep-negative for the retired coinage. That negative needle is ASSEMBLED, not spelled (a two-element array joined on a space): a literal would be the single hit defeating the whole-file grep this repair is measured by. (2) ABLATION: the coinage was restored at ONE site (H8's full-delivery remedy) on a SCRATCH COPY, laid out beside a copy of scripts/invoked-as.mjs because the script imports it; an UNMUTATED CONTROL copy ran at the same path. No rebuild leg applies and none was faked: node loads this .mjs source directly, there is no dist/ anywhere in its resolution path. Mutation confirmed ON DISK by occurrence counts on both copies before running -- ablated: coinage 1 / landed phrase 0; control: coinage 0 / landed phrase 1. Control leg: 1 of 2427 failed -- 'H43 register', environmental (a copy outside scripts/pm cannot see the governed register), NOT MEASURED rather than red. Ablated leg: 4 of 2427 failed = that same environmental one plus exactly the three H8 pins: 'names the release act in the spelling that LANDED' (got false, want true), 'quoting the rule line unbroken' (got false, want true), and 'the retired coinage does not come back' (got true, want false). Direction was as predicted: red, and the grep-negative flipped. The 'names the SECOND half' case stayed green, correctly -- the mutation replaced only the first concatenated segment, so that case pins the second half and not the coinage. The tree file's blob hash was 3161d0b325e74bee1018ba0b0b5ea9620bcd399e before AND after, equal to its HEAD blob; git status clean throughout. A first ablation attempt exited 1 with ERR_MODULE_NOT_FOUND -- read as my own path mistake, not a red gate, and re-run properly rather than written up. (3) GATES: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 32 commands from the committed change set (1 path, three-dot vs merge base f7db8f4); all 32 ran, every exit code captured by redirecting BEFORE any pipe, all EXIT=0. Reconciliation: node scripts/pm/dispatch-gates.mjs --ran ran.list --repo objectstack-ai/objectstack → EXIT=0, '32 derived, 32 run, 0 NOT-MEASURED, 0 UNRUN'. The 7 roster gates whose baseline sits under scripts/ (where my path is) were also run: all EXIT=0, incl. check:pm-label-desc-cap and check-published-list-mirrors. node scripts/pm/check-governed-merges.mjs --test scripts/pm/check-half-states.mjs → EXIT=0, 'NOT governed'. (4) WHOLE-REPO LINT: OS_VERIFY_LOCK_SLOT=issue-15899 bash scripts/pm/os-verify-lock.sh -c 'pnpm lint' → its own VERDICT line reads 'command-exit 0 - held the lock 105s - waited 0s'. This is the FULL farm (eslint . --no-inline-config), not a narrowed run, so no narrowing is declared. (5) NOT MEASURED, named rather than skipped: check:dts-closure and check:published-readme-exports both exit 3 PREREQUISITE NOT MET here (no package dist/; they need a whole-repo build). Neither is reachable from a one-file scripts/pm/ diff, and CI builds before running them. (6) CI, polled in the foreground to completion on head 67bec96: 35 check runs, ALL completed. 34 success/skipped. The ONE failure is 'Lint & Repo Gates', and inside it exactly 1 of 153 steps failed: 'Merge-driver wiring gate' -- the inherited pnpm check:merge-driver red tracked as #15992, exactly as the dispatch caveat named it. Nothing was touched to work around it. For contrast, that same gate exits 0 locally on this branch under pnpm 10.31.0, which is consistent with the Corepack pnpm-12 cause on record. The job log download is blocked by this container's proxy (CONNECT 403), so the step name and the local contrast are the evidence I have; I did not re-derive #15992's cause. (7) NUL/control-byte scan over the edited file: grep -naP over the control ranges, exit captured before any pipe → exit 1, zero output, plus check:nul-bytes green in the gate set. (8) PR body read back in full after creation: byte-identical to what was sent apart from one stripped trailing newline. skip-changeset written through the additive labels endpoint with a comparison read-back: before {size/m}, after {size/m, skip-changeset}, union minus after = empty.",
"mcp_calls": "0 -- a repo-scoped REST probe (GET /issues/15899) returned HTTP 200 at the start of the run, so every read (card body, comments, check-runs, job steps, labels) and every write (PR create, labels, this comment) went through REST. No MCP GitHub call was made, and no channel switch needs declaring.",
"open_questions": [
{
"question": "The H24 header-inventory paragraph is a COMMENT, so its re-pointing has no mechanical pin: the self-test can only assert over EMITTED text, and the whole-file grep that measures the coinage's removal runs only in a reviewer's shell. Should this file gain a source-level self-check so a comment-level regression is caught too?",
"options": [
"A. Leave it. The dispatch ruling scoped the new cases to 'the emitted sentences', the three prescribing sentences are pinned, and a comment cannot instruct a seat to do anything.",
"B. Add one self-test case that reads this file's own source (SELF_PATH is already defined) and asserts zero occurrences of the assembled needle. Cheap, but it makes selfTest do file I/O it does not do today, and any future comment that must NAME the retired coinage to explain it would then be forbidden -- which is exactly why the new constant's docblock describes it without spelling it.",
"C. Move the check out of selfTest into a repo gate that scans for re-coined bracket quotations generally: bracketed text in this file matching no line in SKILL.md. Solves the class, not the instance, and is a card of its own."
],
"recommendation": "A for this PR -- it is what the ruling scoped and what shipped. C as a follow-up card if the PM wants the class closed: the real defect was 'quotation brackets promising a source that does not carry the text', and only C detects the NEXT coinage rather than this one. B is the worst of the three: instance-level coverage bought by forbidding the very explanation a future reader needs."
}
],
"out_of_scope_findings": [
"NOT FILED, reported for the PM to file -- reason for all three: they are observations about scripts/pm/check-half-states.mjs itself, the exact file the PM is running a hard-serial queue on, so filing them unassigned from here would drop cards into a lane the PM is sequencing and whose fold test only the PM can run; the dispatch brief's own routing ('Out-of-scope findings: report, do not fix') reads the same way. (a) H24's remedy sentence and H47's two legs still spell the release rule INLINE rather than through the new RELEASE_ACT_RULE constant. They are correct today and pinned by selfTest's RELEASE_RULE_LINE, so this is tidiness, not a defect: re-pointing them is a refactor the ruling did not scope and would have touched sentences nobody reported. Suggested label: finding.",
"NOT FILED (same reason) -- (b) two docblocks in this file cite 'SKILL.md line 465' and 'line 493' for lines that have since moved; the rule TEXT they carry is verbatim and correct, only the line numbers drifted. The file's own register already tells readers to grep rather than trust numbers, so this may be worth closing by deleting the numbers rather than correcting them. Suggested label: finding.",
"NOT FILED (same reason) -- (c) the observation behind the open question above: no gate detects a NEW bracketed pseudo-quotation in this file. The defect that produced this card was structural (brackets promising SKILL.md's own words over text SKILL.md never carried), and nothing prevents the next one. Suggested label: finding, domain:skills."
]
}
Generated by Claude Code