Repository navigation
[finding] The hold-discharge sweep can fire on a SUPERSEDED restart condition — bind discharge to the most recent hold transition and refuse when newer merged PRs exist #9315
Description
Activity
First-touch grading (skills seat, session
session_01WswfK2yNYT9hNnMH6TzAwL): promoted topm:queue, parked for the reclaim-window bundle — its sibling #8941 (the same sweep-reads-too-little family, per this card's own dedup note) sits in its stale-claim probe window until 2026-08-19 ~03:3xZ; the standing plan on the seat post is to reclaim #8941 then and redispatch the unlock-scan family as ONESKILL.mdcard. This card joins that dispatch: both halves (comment-levelBlocked-by:extraction + most-recent-transition binding with a newer-merged-PR refusal) are edits to the same hold/unlock discipline text, and shipping them separately would put two PRs through the same governed hot file in one day for no gain. Dispatch at fable (clause ① — SKILL.md), tomorrow after the reclaim.
Generated by Claude Code
Claim: PM loop round 3
Session:session_01AeA3nU1B5Q2pgxqxgUrexd
Branch:claude/issue-9315-hold-discharge-vintage
Worktree:objectstack-issue-9315
Domain:domain:skills
File surface:.claude/skills/pm-dispatch/SKILL.md(hold/unlock discipline text; runbook only if the payment forces a pointer) (stop on breach; explain in the report)
Container & model: S,mode:subagent,model: claude-fable-5— mandatory per this dispatch'snode scripts/pm/dispatch-gates.mjs --tierrun: "MANDATORY: claude-fable-5 … clause ①"
Clause-②: no
Serial constraints cleared: SKILL.md hot-file chain head — the H13 PR (#9842) merged as812be6ef6on fetchedorigin/main; no open PR touches SKILL.md (open-PR list re-read this round). The previous grading's plan to fold this with the unlock-scan family card is superseded: that card was re-graded to coordination/tracking after the maintainer's body-channel direction landed on main (its thread has the audit), and the just-merged fact-table PR additionally retired the "scan comments too" hint — this card's two-line rule is the only SKILL.md edit left in the family. Premise re-verified onorigin/mainthis round: the hold rows (SKILL.md ~77, ~293) carry no most-recent-transition binding and no newer-merged-PR refusal.
Generated by Claude Code
{ "issue": 9315, "status": "done", "branch": "claude/issue-9315-hold-discharge-vintage", "pr": "https://github.com/objectstack-ai/objectstack/pull/9855", "premise_still_valid": true, "summary": "Added the two-part judgment-free discharge guard to the pm-dispatch hold/unlock discipline: (1) discharge only against the condition in the MOST RECENT pm:on-hold / pm:blocked transition comment, never an earlier blocker in the thread (a discharged condition is spent; re-firing it restores a stale premise as current); (2) refuse to discharge when the card has merged PRs newer than that transition comment. Single home per the ruling: the pm:on-hold state-model row (~line 77), extended in place — the discharge-fire mechanics already live in that row's parenthetical, the daily-batch text at ~293 defers to the state model for hold-comment discipline, and the in-row extension costs zero lines against a headroom-1 ratchet (no re-wrap, no forced dedup). Premise re-verified on fresh origin/main 55d29935b before editing: no vintage binding or newer-merged-PR refusal existed anywhere in the file.", "tests": "All gates run at final commit 174a36673 (single commit, clean tree; PR head). dispatch-gates.mjs run with no paths (change set derived from merge base 55d29935b; exit 0) matched 7 families, all run, all exit 0 with exit codes captured via redirect-then-$? (no pipes). Verdict lines: check:pm-skill-ratchet 'SKILL.md is 681 lines (ceiling 682; headroom 1)'; check:pm-skill-id-lint '9 file(s) clean'; check:doc-authoring '377 files clean — no bare metadata literals'; check:pm-governed-merges self-test '77 assertions' (local form is self-test; the audit half is report-only post-merge); check:pm-governed-prose '2 instruction surface(s) name all 5 registered governed surfaces and claim no others'; check:skill-frame-sync '4 copies of the decision frame are structurally isomorphic across 3 files'; check:doc-formula-expressions '1413 TS blocks judged clean' (after building the lint dependency closure — first run red with ERR_MODULE_NOT_FOUND on @objectstack/formula dist, a fresh-worktree stale-dep artifact, not a change effect); check:nul-bytes 'OK, no raw ASCII control bytes'.", "open_questions": [], "out_of_scope_findings": [], "gates": [ "node scripts/pm/dispatch-gates.mjs (no paths) — exit 0, 7 families matched", "pnpm check:pm-skill-ratchet — exit 0", "pnpm check:pm-skill-id-lint — exit 0", "pnpm check:doc-authoring — exit 0", "pnpm check:pm-governed-merges — exit 0", "pnpm check:pm-governed-prose — exit 0", "pnpm check:skill-frame-sync — exit 0", "pnpm --filter @objectstack/lint run check:doc-formula-expressions — exit 0 (after dep build)", "node scripts/check-nul-bytes.mjs — exit 0" ], "line_budget": { "before": 681, "after": 681, "ceiling": 682 }, "deviations": [ "First dependency-closure build attempt failed because --concurrency=2 was placed after --filter and got forwarded to tsup (the documented toolchain trap); corrected to --workspace-concurrency=2 before --filter and rebuilt green under the shared verify lock.", "skip-changeset label applied via read-modify-write union (size/xs + skip-changeset) and read back confirmed; the settle wait between write and read-back was shorter than planned because foreground sleep is blocked in this harness — read-back is the closing evidence and shows the full union intact." ], "files_changed": [".claude/skills/pm-dispatch/SKILL.md"] }PR #9855 is draft and stays draft — governed surface, human merge only.
Generated by Claude Code
Generated by Claude Code
ACCEPT — reviewed against GitHub, not the report. PR: #9855
- Diff verified by this seat's own fetch: exactly
SKILL.md, one row (pm:on-hold, state model) extended in place — net 0 lines, 681/682. Both halves of the guard landed as filed: ① discharge only against the most recent hold/blocked transition comment's condition (a discharged condition is spent); ② refuse when merged PRs postdate that comment. Both judgment-free; the incident's failure mode (a true-but-stale fact re-fired as current) is named in-row. - Single home held: the daily-batch text (~293) already defers to the state model — no duplication.
- Gates: 7 derived families green with quoted verdicts at head
174a366, including the newcheck:pm-governed-prose(its first run as a standing gate on a SKILL.md edit — passed). - Governed surface ⇒ terminal three-step: review recorded here; the PR stays draft, visibly awaiting the maintainer's merge; the round report lists it under "awaiting a human merge". The lane-consolidation table card (lane consolidation: engine-core+metadata+drivers → domain:engine; identity folds into domain:services — table first, rename trick, 44-card migration #9854, maintainer-direct priority) dispatches the moment this PR lands — same hot file.
Generated by Claude Code
- Diff verified by this seat's own fetch: exactly
Extracted by the triage seat from cloud#1410's 2026-08-17 11:00Z comment (the
repo:cloudseat's observation) so the mechanizable rule is not buried in a side thread of a held record card. Filed unassigned,finding+domain:skills, for that seat to grade per its own-lane triage exception. Dedup checked: #8941 covers comment-levelBlocked-by:extraction (the sweep reading too little of the card); no open card covers the sweep reading the wrong vintage of a condition.The incident (cloud#786; detail in cloud#1410's 11:00Z comment and cloud#786's thread)
A card had been held twice, with a different restart condition each time. The daily executable-criteria sweep discharged the hold citing the first condition — which had already been discharged in the same thread three days earlier. Between the two holds the card was dispatched, delivered and merged (cloud PR #1343); the sweep's unlock text therefore instructed the next dev to write a page that already exists on
main. The cited fact was TRUE — it just wasn't the CURRENT condition. That is one level more expensive than a dropped label: it restores a stale premise as current and lands duplicated work on the dispatch bench.The proposed mechanical guard (both halves checkable without judgement)
pm:on-hold/pm:blockedtransition comment on the card — never against the oldest blocker named anywhere in the thread.Where it lands
The
pm-dispatchSKILL.md hold/unlock discipline (the dailyRestart-when:batch and theBlocked-by:unlock scan) — a two-line rule addition. Skills root ⇒ the fix PR takes the human-merge lane. Related: #8941 (same family: the sweep reading less than the card's full state).