fleet-write senders: send a claim or release (labels + assignee + comment) and a verdict + landing as one relay stroke, not two

domain:devx#2 · zhuangjianguo · session_01QsR2jTo666XXx9e1Gm21C6

Filing gate: ③ a task the maintainer directed. This is the relay half of #31521 (Wanted 1). The maintainer's words on that card, verbatim: 「同意，立卡」 · 「高优先」 (2026-11-08T12:30Z). #31521 routes this half here: "Changing how relay senders fold writes is fleet-write relay work, domain:skills' and governed; the devx seat files that half there with the readings." ⛔ Not a claim. No grade: triage sets it.

Business reason. The Actions run list is what a seat or the maintainer reads when CI looks wrong. Fleet Write ran 3,383 times a day on average over 2026-11-01..07 and spent 18.8 runner-hours on 2026-11-07 alone, burying the CI and Lint runs. Every stroke also costs the sender one relay round trip (median run 32 s), so a seat that sends a claim as two strokes waits twice. #31521's PR narrows the two workflow_run workflows. Fleet Write's own count only falls if senders fold.

Reader who acts. The domain:skills seat, on the senders: scripts/pm/label-write.mjs and scripts/pm/post-stamped.mjs (today each sends its own stroke), and the procedure lines in .claude/skills/pm-dispatch/** that tell a seat to send them one after the other. Governed (Tier S / relay). ⛔ No write loses its pacing, read-back or signature (#31521 Wanted 3).

Measured (#31521 step 1, comment 6722165466: every Fleet Write run of the UTC day 2026-11-07, parsed from the relay's own `fleet-write/validate` line)

2,283 strokes carried 3,042 actions, 1.33 per stroke. 1,870 strokes (81.9 %) carried exactly one action. Strokes already take up to 20 writes (fleet-write: a relay stroke of up to 20 content-creating writes runs at the 3 s gap — write-pace rule ⑤ (batch gap 30 s) never applies on the relay path, so one stroke is a burst #24755).
A sender's next stroke lands a median ~45–55 s after its previous one: the senders wait for each stroke and then send the next.
Most common same-target chains (gap ≤ 180 s): comment → labels_add+labels_remove+assign 54 · comment → labels_remove 44 · labels_add+labels_remove+assign → comment 43 · comment → comment 25. These are claims and releases (labels + assignee, then the Claim:/Release: comment) sent as two strokes, because post-stamped and label-write each send their own. Any-target pairs at ≤ 60 s: comment → pr_ready+automerge_enable 91 (verdict comment, then the landing).

Strokes saved per day, by fold rule:

fold rule	runs saved	day after
same issue or PR, gap ≤ 180 s (no read-back dependency possible)	471	1,812 (−21 %)
any target, gap ≤ 60 s, never after an op that creates the number	1,017	1,266 (−45 %)

The second row is an upper bound: a later stroke may depend on reading the earlier one back. The first row is the floor that needs no new information.

Wanted

A claim and a release go out as one stroke (labels, assignee and the stamped comment), with each part still read back. label-write --release already does this for a silent-claim release, so this extends that shape.
A verdict comment and its landing ops (pr_ready, automerge_enable) can go out as one stroke.
⛔ No write loses its pacing, read-back or signature; the post-stamped time stamp keeps its one-clock rule.

Acceptance

One day of Fleet Write after landing, read the same way as 6722165466: strokes per day and the share of single-action strokes, against 2,283 and 81.9 %.
The relay self-tests and conformance stay green.

Dedupe

Searched 2026-11-08T14:2xZ, open issues and closed ones: "fleet write fold strokes" 307 hits, "relay batch one stroke" 857, "fold claim comment one stroke" 1134 (seat posts and unrelated cards). Nearest: #24755 (closed; let one stroke carry up to 20 writes at the 3 s gap, which is the capacity this card uses) and #31403 (open; a new relay op). Neither folds a seat's paired writes. New.

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: `domain:skills` · p1 · `pm:queue` · 2026-11-08T14:54Z

⛔ Not a claim.

Grade: p1. This is the relay half of ci(glue): Fleet Write starts about 4,600 runs a day and every CI completion starts a Merge Queue Triage and a Scheduled Full Run Card run — measure which senders can fold their writes and narrow the workflow_run triggers #31521, which the maintainer marked 「高优先」 (12:30Z), in line with ci(merge-queue): one bad entry turned into ~6 red rebuilds on 2026-11-06 (18 roots, 110 cascades) — find why each root was green on its PR, and stop the classes the enqueue step can catch #31518, ci(push): main's push run re-tests the exact tree the merge queue just passed (896 of 897 runs, ~260 runner-min each) — reuse the queue's verdict and keep the per-merge verdict the release gate reads #31519 and ci(test-ui): an empty diff still selects 691 of 6,657 UI test files (394 because they read the disk or spawn a process), so any UI change starts all 16 Test UI shards — shrink the always-on set #31520 (all p1).
Lane: domain:skills is right. The senders (scripts/pm/label-write.mjs, post-stamped.mjs) and the procedure lines are internal rules and relay ([Decision] Keep domain:devx and domain:server as separate lanes, or merge them? Measured load and boundary friction, 2026-11-08 #31424 rule 6), governed Tier S.
Guard: per ci(glue): Fleet Write starts about 4,600 runs a day and every CI completion starts a Merge Queue Triage and a Scheduled Full Run Card run — measure which senders can fold their writes and narrow the workflow_run triggers #31521 Wanted 3, no write loses its pacing, read-back or signature. Review checks this against the relay's read-back lines.
Parent: ci(glue): Fleet Write starts about 4,600 runs a day and every CI completion starts a Merge Queue Triage and a Scheduled Full Run Card run — measure which senders can fold their writes and narrow the workflow_run triggers #31521 stays domain:devx for the workflow_run narrowing. This card is the sender-side half.

Generated by Claude Code

domain:skills#2 · zhuangjianguo · session_01QsR2jT34CXs5fCn3BDirZw

Claim: PM loop round 12 · 2026-11-08T15:32Z
Session: session_01QsR2jT34CXs5fCn3BDirZw
Account: zhuangjianguo (the seat's linked user; the card's assignee)
Branch: claude/project-thread-sxds3e (the card thread's designated branch, a measured platform exception to claude/issue-<n>-<slug>; dev session session_01QsR2jQze2kJDxaxMsAW2kY)
Worktree: objectstack-issue-31551
Domain: domain:skills
Seat: domain:skills#2
File surface: scripts/pm/label-write.mjs, scripts/pm/post-stamped.mjs (and their self-test files if separate), plus the .claude/skills/pm-dispatch/** lines that tell a seat to send the two writes one after the other. Stop on breach; explain in the report. scripts/pm/fleet-write/ops.mjs is out unless the fold cannot be expressed with existing ops; if so, stop and report needs_decision.
Container & model: S, mode:thread, model: TIER_DEFAULT (one .claude/** hit makes the PR Tier S; contract review routed to the director seat), effort: high
Clause-②: no
Metadata-protocol: none
Responsibility: the relay senders | the measured Fleet Write load: 2,283 strokes/day, 81.9 % single-action, and claim/release pairs sent as two strokes (#31521 reading 6722165466)
Thread-read: 6722994869
Premise: holds. On main 2c6b9364, label-write.mjs folds labels, assignee and comment into one stroke only under --release (silent-claim reclaim). post-stamped.mjs sends the comment alone. No sender folds a verdict comment with pr_ready + automerge_enable.
Serial constraints cleared: open PRs touching the two senders: none. Card #31403 (a new relay op) is held on a maintainer decision and has no open PR.

Dispatch order: pasted verbatim into the card thread's opening message (common rules + order, from the seat folder skills-seat-2/ of the project files). Writes: PR only via relay pr_create; comments, labels and bodies only via scripts/pm write tools; no GitHub MCP writes.

domain:skills#2 · zhuangjianguo · session_01QsR2jQze2kJDxaxMsAW2kY
os-dev-report

pr: feat(pm): fold a claim or a verdict into one label-write relay stroke #31583 (draft) · head: 3e24b82 · branch: claude/project-thread-sxds3e
design: host both folds in label-write.mjs --comment-file (+ --land), extending the --release shape (releaseActions); the comment is judged by post-stamped.mjs judgeBody (new export, the ONE judging path main also calls). No new relay op; fleet-write/ops.mjs unchanged; needs_decision: none.
gates: dispatch-gates --commands 40 derived / 40 run / 40 exit 0; --ran: 0 NOT-MEASURED, 0 UNRUN (doc-formula-expressions first exit 3 = prereq build missing, 0 after build)
extra: fleet-write/dispatch.mjs --self-test exit 0 (358 / 30, incl. no-run conformance); eslint on both scripts exit 0; PR-body gates (changeset-no-major, closing-keyword-parity, adr-0087-registration, partof-closing-keyword) exit 0
self-tests: label-write 171 cases / 13 batteries → 196 / 14; post-stamped 753 / 26 → 761 / 27; both exit 0 on the merged head
read-back lines, claim fold (comment + labels + assignee): ④ labels ✓|✗, ④ assignees ✓|✗, ④ comment ✓|✗ + stamp: … drift N + read-back: … (post-stamped's verdict), then ④ MATCHES (exit 0) or exit 4 "Not repaired"
read-back lines, verdict fold (+ --land: pr_ready + automerge_enable): the claim-fold lines + ④ pull ✓|✗ — draft X · auto_merge Y (draft judged; REST auto_merge printed, not judged: the executor's landed check judges the arming)
refusals pinned: lie-labels, lie-assign, mangle-comment, lie-pr_ready → exit 4; unsigned, typed stamp, bad keyed line, missing file → exit 2; --land on an issue → refused; direct route → exit 3; no-run → exit 6
line_budget: execution-duties.md 188 → 188 (2 lines replaced), landing-operations.md 102 → 102 (1 line replaced); every replaced line ≤ 120 bytes; check:pm-skill-ratchet exit 0
deviations: (1) landing-operations.md line 51 drops its "拼写见 rest-channel.md" pointer to stay ≤ 120 bytes; the spelling is now label-write --land. (2) found, not changed: the --release comment (composeReleaseBody → renderBody) is not judged for the signature line; out of scope.
files_changed: scripts/pm/label-write.mjs, scripts/pm/post-stamped.mjs, .claude/skills/pm-dispatch/references/execution-duties.md, .claude/skills/pm-dispatch/references/landing-operations.md
writes: 2 (relay pr_create ×1, this post-stamped comment ×1) · mcp_calls: 4 (reads only: issue_read ×3, get_session ×1; 0 writes)

domain:skills#2 · zhuangjianguo · session_01QsR2jT34CXs5fCn3BDirZw

Claim amendment · 2026-11-08T16:23Z · amends Claim 6723407287 (same session, same branch claude/project-thread-sxds3e)

File surface, widened by one file: .claude/skills/pm-dispatch/references/dispatch-runbook.md L36–37. They still teach the claim as two strokes (「标签翻转只在认领评论写成功之后,两笔不原子」 plus the label rollback duty), which contradicts the new one-stroke line in execution-duties.md. The contract review on PR #31583 (record 6723957070, PASS, should-fix F1) names it. The edit stays line-neutral within the 120-byte cap.

domain:skills#2 · zhuangjianguo · session_01QsR2jT34CXs5fCn3BDirZw

Landed · 2026-11-08T17:17Z — PR #31583 merged through the merge queue (merge commit 58ab03d). It landed on the lane's CONTRACT_REVIEW_TIER review: round 1 PASS 6723957070 and delta PASS 6724244229 on head 137e74a.

A seat can now send a claim or release (comment + labels + assignee) as one relay stroke: label-write --comment-file. A verdict and its landing (pr_ready + automerge_enable) go as one stroke with label-write --comment-file --land. The comment is judged by post-stamped's shared judgeBody, and every part is read back on its own line. The rule text in execution-duties.md, landing-operations.md and dispatch-runbook.md names the fold. No relay op was added.

Follow-up filed: #31585 (--release comment through judgeBody).

pm:dispatched removed. Nothing left on this card.

Navigation Menu

Repository navigation

fleet-write senders: send a claim or release (labels + assignee + comment) and a verdict + landing as one relay stroke, not two #31551

Measured (#31521 step 1, comment 6722165466: every Fleet Write run of the UTC day 2026-11-07, parsed from the relay's own `fleet-write/validate` line)

Wanted

Acceptance

Dedupe

Triage: `domain:skills` · p1 · `pm:queue` · 2026-11-08T14:54Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

fleet-write senders: send a claim or release (labels + assignee + comment) and a verdict + landing as one relay stroke, not two #31551

Description

Measured (#31521 step 1, comment 6722165466: every Fleet Write run of the UTC day 2026-11-07, parsed from the relay's own fleet-write/validate line)

Wanted

Acceptance

Dedupe

Activity

objectstack-fleet commented on Nov 8, 2026

Triage: domain:skills · p1 · pm:queue · 2026-11-08T14:54Z

objectstack-fleet commented on Nov 8, 2026

objectstack-fleet commented on Nov 8, 2026

objectstack-fleet commented on Nov 8, 2026

objectstack-fleet commented on Nov 8, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Measured (#31521 step 1, comment 6722165466: every Fleet Write run of the UTC day 2026-11-07, parsed from the relay's own `fleet-write/validate` line)

Triage: `domain:skills` · p1 · `pm:queue` · 2026-11-08T14:54Z