[repo:hotcrm] Install the half-state patrol in hotcrm — workflow copy, anchor issue, repository variable (hotcrm half of the ruled 纳入)

[claude]

Seam card, filed by the domain:skills seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve) for the repo:hotcrm execution seat as its named reader. hotcrm's API is refused for this session (measured 403 on the repository and labels endpoints against 200 on objectstack), so the hotcrm half of the ruled install is handed over here rather than filed into hotcrm's own queue.

What was ruled

Maintainer ruling recorded on the parent card (director seat, summon 11, 2026-09-03, verbatim 「其他同意」 adopting 纳入): hotcrm runs the same half-state patrol the platform repos run — a workflow copy in hotcrm plus a pinned anchor issue whose rows the hotcrm seat reads at round open; report-only, no strip-on-close, no bulk relabel of the closed carriers.

What this card asks the hotcrm seat to do

Execute the install recipe the measurement flight delivered — the "Part B" section of the os-dev-report comment on the parent card (comment id 5529279872), measured against a read-only hotcrm clone at 18444c8. In one line each:

0. label prerequisite first: PR hotcrm's governed .github/labels.yml to declare tracking (its only anchor label and an H13 exemption; the fleet seeder does not create it — see the queued card on the seeder);
1. copy THREE files unchanged from objectstack main after the objectstack half lands: the sweeper, scripts/invoked-as.mjs (the sweeper imports it), and the workflow; verify with the sweeper's --self-test in the hotcrm tree (expect 2062 cases, exit 0);
2. open the install PR with skip-changeset;
3. open the anchor issue (title, single tracking label, pinned, first body line = the machine-findable marker, exactly as the recipe spells it);
4. set the repository variable HALF_STATE_ANCHOR_ISSUE to the anchor number;
5. prove it by one workflow_dispatch run: sweep exit 0, anchor body rewritten with a fresh Swept line.

The form is a verbatim copy: the upstream workflow declares no workflow_call, so uses: is structurally unavailable; the checkout variant would pin hotcrm to a floating upstream ref and is not taken.

Decisions this card carries for the hotcrm seat (not decided here)

• Priority-axis spelling. hotcrm spells it prio:p0/p1/p2; the shared sweeper reads priority:*. The patrol installs and runs either way (an absent label is a blind dimension, never an error — measured), but H10 can never fire there and H11 misses prio:* as a signal. The flight's recommendation, argued on the four axes in the report: A — hotcrm aligns to the fleet vocabulary, relabelling only the open prio:* cards (the hotcrm seat's own 11:19Z census counted six; not independently verified here) and adding priority:* to hotcrm's labels manifest; ⛔ no sweep over the closed archive, per the earlier ruling that refused a bulk relabel. B (a per-install prefix knob in the shared tool) is consumer-side tolerance for a producer-side divergence; C (document the gap) is the fallback if the census is stale. If the hotcrm seat reads this as human floor, it goes to the maintainer through that seat's own decision batch.
• PM_SWEEP_CLOSED_FLOOR. Decided by the skills seat as an execution detail: install with no floor, read one run's census line, add a floor only if the count is dominated by pre-convention closes (the row-per-carrier flood the floor was invented against has been folded to a count, and the closed-issue window is three days).
• Stale bot. hotcrm's stale workflow does not exempt tracking and closes after 7 days; the same hazard is queued for objectstack's own anchor. Decide whether the hotcrm anchor also needs pinned before the first 7 days elapse.

Boundaries

⛔ Nothing in the recipe writes a label or closes a card in hotcrm beyond the anchor's own tracking. ⛔ No local re-implementation of the sweeper. The objectstack half (the corrected adopt list and path filter) is PR 15000 on the parent card, Part of that card; the parent stays open until this card closes.

Refs: parent objectstack card 14881 (ruling comment 5523302416, recipe comment 5529279872) · seeder gap card 15001 · stale-bot card 15002 · objectstack anchor 9857.

---

Generated by Claude Code