Install the parameterised half-state patrol (sweeper + workflow pair from objectstack PR #11294)

[os-zhuang]

Blocked-by: objectstack-ai/objectstack#11217

Filed by the objectstack skills seat (session session_01RMTpSRF5CjMmQBFfPtPCwJ, 2026-08-23) at the ACCEPT of objectstack PR #11294, per the consumer-card duty on that landing. Measured context: 28 open pm:blocked cards in this repo have never been machine-swept, and a hand-run of the H19 predicate found 7 stale blocks (objectstack#11217's table).

What

Once the upstream landing merges, adopt the pair verbatim: scripts/pm/check-half-states.mjs + .github/workflows/half-state-patrol.yml from objectstack. Both are parameterised for exactly this: the swept repo resolves PM_SWEEP_REPO → GITHUB_REPOSITORY → default (a copy running here sweeps THIS repo with its own repo-scoped GITHUB_TOKEN — no cross-repo credential), and the anchor issue comes from the repository variable HALF_STATE_ANCHOR_ISSUE with a loud failure when unconfigured (it will NOT silently write into a wrong issue).

Executable acceptance criterion

A scheduled (or manually fired) run whose summary names THIS repo and writes its findings into this repo's own anchor issue; node scripts/pm/check-half-states.mjs --self-test green at the adopted revision.

⚠️ Manual maintainer step (cannot be done by an agent)

Create the anchor issue here and set the repository variable HALF_STATE_ANCHOR_ISSUE to its number — repository variables are settings-surface writes. The workflow fails loudly (after the sweep, findings preserved in the run summary) until it is set.

Note: cross-repo Blocked-by: targets (e.g. pointing at objectstack) stay UNJUDGED per install — accepted at grading (objectstack#11218 carries the loudness half).

Refs: objectstack-ai/objectstack#11217 · objectstack-ai/objectstack#11196 · PR objectstack-ai/objectstack#11294.

[os-zhuang]

Triage: added domain:devx (lane visibility for the eventual unlock). Blocker note: the wait target is objectstack PR #11294, not an issue — the cross-repo/PR-target shape H19 cannot resolve (recorded in objectstack#11218/#11220), so the unlock will not fire mechanically; the triage seat carries it in the manual scan until that PR merges.

---

Generated by Claude Code

[claude]

Unlock scan (triage daily round, session session_01Kktexqp6uVuFMztvvTMf3V, 2026-08-24): released to pm:queue — blocker objectstack#11217 closed (completed) at the ACCEPT of objectstack PR #11294, so the check-half-states.mjs + half-state-patrol.yml pair now exists on objectstack main to adopt verbatim. Adoption copies files across repos (no published-package gating). Re-verify the pair's current shape on objectstack origin/main before copying — it may have moved since the card was filed.

---

Generated by Claude Code

[os-warren]

Measurement for whoever installs this: pm:dispatched on closed cards is at 813, so one of the two conventions is wrong

Recorded by the domain:devx @ objectui seat (#5748), round R3, PM session session_01PQ3NihCHE9LUtHoGxo6A9f. ⛔ Not a claim on this card and no label changed by this comment — a reading the patrol's parameterisation should settle before it starts emitting rows.

What I hit. Landing three cards this round, I stripped pm:dispatched on close, following this lane's own recorded R2 practice (which cited a patrol row it called "H22"). Then I measured the corpus:

mcp: list_issues repo=objectui labels=[pm:dispatched] state=CLOSED -> totalCount: 813

813 closed cards carry pm:dispatched. So either the fleet rule is "strip on close" and essentially nobody does it — in which case installing the patrol here lights up 813 rows on day one, which is not a worklist, it is noise that trains people to ignore the patrol — or stripping is not the rule and one seat invented it, in which case the three I stripped this round are the deviation.

I could not settle it from inside this repo, which is the point. The patrol is not installed here (.github/workflows/half-state-patrol.yml does not exist — that is this card), so there is no H-row predicate in objectui to read. The "H22" reference in the R2 seat post therefore points at objectstack's patrol, and whether its predicate is scoped to open cards is exactly the thing this port must not guess at.

Concretely, for this card's implementer:

1. Read the objectstack predicate for the pm:dispatched row and check whether it filters on state: open. If it does, the 813 are irrelevant and stripping on close is unnecessary ceremony.
2. If it does not filter on open, decide before landing whether the port emits those 813 rows, backfills them, or scopes to open. ⛔ Do not land a patrol whose first run produces an 813-row report.
3. Whichever way it goes, the answer belongs in the seat-post protocol, because right now two seats can read the same corpus and land on opposite conventions — as happened here.

For the record, the three I stripped this round are #5367, #3840 and #5458 (all closed completed by merged PRs), plus pm:queue off the already-delivered #5313. I kept my own round internally consistent rather than switching conventions mid-round; ⛔ I am not proposing a backfill of the other 813.

---

Generated by Claude Code

[yinlianghui-tw]

Claim: PM loop round 7 (batch raised to 5 on maintainer instruction 「并发数默认好像是3，你加到5吧，加快进度」)
Session: session_019b5UBNMtTzKbVtZZGvFuxe
Branch: claude/issue-5791-install-half-state-patrol
Worktree: objectui-issue-5791
Domain: domain:devx (@ objectui, seat #5748)
File surface: .github/workflows/half-state-patrol.yml (new) + its sweeper script (new, expected under scripts/) + whatever workflow-inventory pin the repo requires
Container & model: M, mode:subagent, model: opus
Clause-②: no
Serial constraints cleared: co-dispatched this round — #5174 b3 (content/docs/guide/{plugin-development,schema-rendering,theming}.md + check-doc-snippet-types.mjs ledger), #4926 (tsconfig.scripts.json), #4850 (two source files, ⚠️ surface not yet known — the dev is locating them), #5410 (scripts/check-phantom-dependencies.mjs + its test). ⛔ #5754 deliberately NOT dispatched alongside this one: it also adds a workflow, and both would land on the same workflow-inventory pin — a guaranteed collision. It stays queued until this lands.

PM-verified premise on origin/main @ 2aff580b5

.github/workflows/half-state-patrol.yml is absent — 21 workflow files present (that count is the control that the listing works), none matching half-state, and no sweeper script under scripts/ matching half|patrol|sweep. The card's premise holds.

⚠️ This card fixes a gap that currently makes a standing seat rule vacuous. The domain:devx @ objectui seat's per-round first criterion is "read the half-state patrol anchor and action its rows". With no patrol installed here, that criterion has been silently no-op for every round this seat has run. That is the value of this card — ⛔ not tidiness.

⚠️ The instruction that must not be lost — read before porting anything

The prior seat recorded this on the card and it is the one thing that turns a clean port into a disaster:

pm:dispatched sits on 816 closed cards in this repo, while this lane's practice is to strip it on landing.

The objectstack patrol's predicate must be checked for a state: open filter BEFORE this lands. If it lacks one, its first run emits ~816 rows of pure noise on day one and the patrol is dead on arrival — nobody will read row 817. Verify the filter, and if it is absent, add it and say so in the PR. ⛔ Do not port verbatim and leave this to be discovered by its first run.

Source

Port the sweeper + workflow pair from objectstack PR #11294. That repo is readable from this session over git (git ls-remote / git fetch against objectstack-ai/objectstack succeed) though ⛔ not pushable — read it, port it, land it here.

⚠️ Port, ⛔ do not transplant. objectui differs from objectstack in ways that matter to this patrol: no scripts/pm/ directory exists here (dispatch-gates.mjs and friends are objectstack-resident), the label vocabulary may differ, and this repo's seats strip pm:dispatched on landing where objectstack's may not. Check each assumption the ported code makes about repository shape; report every place you had to adapt it, because those are exactly the places a later objectstack change will silently diverge.

Report-only, ⛔ not enforcing

The patrol is a report-only instrument. ⛔ Do not have it mutate labels, close cards, or fail CI on findings. It surfaces half-states for a human or a seat to action; a patrol that acts on its own findings is a different and much larger card.

Gates

No dispatch-gates.mjs here — derive from your actual diff. Adding a workflow very likely trips the repo's workflow-inventory pin (scripts/__tests__/ci-cd-pipeline-doc.test.ts and content/docs/guide/ci-cd-pipeline.md) — check and update it if so; that repo's own test records that an inventory which understates the gates is treated as a defect. Plus scripts/__tests__, Lint, and the new script's own test if you add one (you should). Run check-changeset-presence.mjs and follow its verdict; ⛔ no skip-changeset label — never a real mechanism here (#4912/#3724).

Do a dry run and paste its output. A patrol whose first real run is in production is exactly the failure this card is about.

---

Generated by Claude Code