skills(pm-dispatch): a dispatched PR carries its card's assignee (the seat's linked user), set at pr_create and moved on takeover — maintainer 2026-09-25 「同意」

Filing-gate category: ③ — a maintainer-directed task. Filed by the domain:skills#1 seat (session_01LNRg6FUMXYWgkAbDsfZpC8) under the direct-dispatch channel; the audit comment on this card quotes the authorization verbatim.

Maintainer's words (verbatim, this seat's chat, 2026-09-24/25, ⛔ not translated)

The question: 「现在pr都是车队提交的，有点分不清是谁在处理，是否建议也写入 assignee，包括pr被接管时应该更新 assignee」

The seat's recommendation, given in chat: PR assignee = the card's assignee (the dispatching seat's linked user account), set by the dev right after pr_create with the label-write tool; a takeover moves every open PR's assignee of that card in the same stroke as the card's; a seat's own tool PR is assigned to the seat; the reader is the filter is:pr is:open assignee:LOGIN; no new patrol row unless the maintainer names one.

The ruling: 「同意」

What lands (rule text only; ⛔ no gate, no patrol row, no new tool)

.claude/skills/pm-dispatch/references/execution-duties.md, 认领 section — one rule: the dev's pr_create is followed by label-write --issue PR_NUMBER --assign with the card's assignee; the takeover four-piece (C9) moves the assignee of every open PR of the card in the same stroke (--unassign old --assign new on the PR number too); a seat's own PR is assigned to the seat. The file sits at its line ceiling (183 / 183): pay every added line with a deletion or an in-place rewrite, ⛔ never a re-wrap.
.claude/agents/os-dev.md, rule 2 (「你恒不写 assignee」) narrowed to the CARD's assignee, and the write budget line (「写预算四笔」) naming the PR-assignee write as the label-write slot it already has; the standard clause that objectui path labels belong to the labeler is unchanged.
Any other reader of the assignee rule found by a sweep of .claude/** and AGENTS.md for assignee — listed in the PR body with a per-hit decision (changed / already consistent / out of scope with reason).

Pilot already run: PRs objectui#10410, #10412, #10413 carry assignee os-elon-musk (= the cards' assignee), set by their devs through the relay; the measured spelling is node scripts/pm/label-write.mjs --repo OWNER/NAME --issue PR_NUMBER --assign LOGIN (the relay's assign op, POST /issues/{n}/assignees, works on a PR number).

Who acts

The domain:skills seat: Tier S (.claude/** only) ⇒ in-seat contract review at CONTRACT_REVIEW_TIER, then the queue.

Dedupe

search_issues on objectstack at filing: 「PR assignee equals card assignee dispatching seat linked user takeover updates assignee」 → 52 (open + closed); the nearest are #15670 (closed — who sets the ISSUE assignee, CLAUDE.md vs os-dev.md rule 2) and #17932 (closed — a re-routed card keeping the old lane's assignee). Neither concerns the PR's assignee. Dedupe words: PR assignee mirrors card · pr_create assign · takeover moves PR assignee · is:pr assignee filter · os-dev rule 2 assignee

Generated by Claude Code

Claim: PM loop round 1
Session: session_01LNRg6FUMXYWgkAbDsfZpC8
Branch: claude/issue-20063-pr-assignee-mirrors-card
Worktree: objectstack-issue-20063
Domain: domain:skills
Seat: domain:skills#1
File surface: .claude/skills/pm-dispatch/references/execution-duties.md (认领 section, the assignee rules — 183 / 183, every added line paid in place) · .claude/agents/os-dev.md (rule 2 and the write-budget line) · any other .claude/** / AGENTS.md reader of the assignee rule the sweep finds, decided per hit (stop on breach; explain in the report)
Container & model: S — rule text, a few sentences rewritten in place, mode:subagent, model: opus — dispatch-gates --tier --repo objectstack-ai/objectstack on these paths prints the clause-① mandate for .claude/agents/os-dev.md (CONTRACT_REVIEW_TIER) and its exit: "a one-line-class mechanical governed edit drops to opus execution … compensated by the skill-face review at claude-fable-5-1" — taken: opus build, this seat's in-seat review at CONTRACT_REVIEW_TIER
Clause-②: no
Thread-read: none
Serial constraints cleared: execution-duties.md and os-dev.md are FREE — PR #20009 (the last PR on execution-duties.md) MERGED on 2026-09-24 (the timeline on that PR carries the minute); no open objectstack PR touches either file (open PR head list read at {{NOW}}); no claude/issue-20063* head on origin; the sibling take of this round (objectui #10048) is in another repo

Take basis (written at 2026-09-25T01:02Z): category ③ — the maintainer's 「同意」 on this seat's recommendation, both quoted verbatim on the card; under 「并发恢复3」 (batch 3, 1 other in flight). Premise on origin/main 3557f85: os-dev.md rule 2 still reads 「你恒不写 assignee」 and execution-duties.md :56 「assignee 字段归 PM:原子对 step 1 设,dev 席恒不写它」 — both about the card; neither text mentions the PR's assignee. Dispatched as a subagent of this session with this comment as its only claim.

Generated by Claude Code

os-dev-report
Reported at 2026-09-25T01:18Z by session_01LNRg6FUMXYWgkAbDsfZpC8 for PR 20066.

{
  "issue": 20063,
  "status": "done",
  "branch": "claude/issue-20063-pr-assignee-mirrors-card",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20066",
  "session": "session_01LNRg6FUMXYWgkAbDsfZpC8",
  "premise_still_valid": true,
  "summary": "Rule text only, as the ruling says. execution-duties.md (the claim section): the card assignee stays the PM's and the dev never writes it; the PR assignee equals the card's; the dev sets it in the same round as pr_create with `label-write.mjs --issue PR_NUMBER --assign LOGIN`; a seat's own PR is assigned to the seat; the takeover's step two moves it in the same stroke; the reader is `is:pr is:open assignee:LOGIN`. os-dev.md: rule 2 is narrowed to the CARD's assignee and gives the PR spelling, and the write-budget line names the PR-assignee write inside its label-write slot. PR 20066 is a draft, its assignee is os-elon-musk (read back), and it carries skip-changeset. The newest Claim (5824909861) names this branch, and the card's assignee was os-elon-musk when this run picked it up.",
  "tests": "No code changed. The gates below ran at head 6364f11. dispatch-gates --ran: '✓ dispatch-gates --ran: 20 derived famil(ies) accounted for — 20 run, 0 NOT-MEASURED (a DERIVED zero — all 20 recorded an exit code and none of them is 3).'",
  "gates": [
    "node scripts/check-closing-keyword-parity.mjs :: exit 0",
    "node scripts/check-closing-keyword-parity.mjs --self-test :: exit 0 · '40 assertions, 5 mutations … each driven to red'",
    "node scripts/check-comment-mask-corpus.mjs :: exit 0 · '7124 files, 0 disagree'",
    "node scripts/pm/check-governed-queue-guard.mjs --self-test :: exit 0 · '279 cases pass'",
    "node scripts/pm/check-harness-current.mjs --self-test :: exit 0 · 'all 26 cases passed'",
    "pnpm --filter @objectstack/lint run check:doc-formula-expressions :: first exit 3 (PREREQUISITE NOT MET: formula+lint not built, nothing measured) → built under os-verify-lock (VERDICT command-exit 0) → exit 0",
    "pnpm check:agent-model-declared :: exit 0 · 'no definition leaves its tier to the dispatching session'",
    "pnpm check:agent-test-spelling :: exit 0",
    "pnpm check:commit-card-trailers :: exit 0 · '81 cases pass'",
    "pnpm check:cross-package-test-inputs :: exit 0",
    "pnpm check:doc-authoring :: exit 0 · 'no growth, no burn-down unrecorded'",
    "pnpm check:driver-memory-census :: exit 0",
    "pnpm check:gitlink-declared :: exit 0",
    "pnpm check:nul-bytes :: exit 0 · 'no raw ASCII control bytes'",
    "pnpm check:pm-governed-merges :: exit 0",
    "pnpm check:pm-skill-id-lint :: exit 0 · '✓ check-skill-id-lint: 34 file(s) clean'",
    "pnpm check:pm-skill-ratchet :: exit 0 · '✓ check-skill-line-ratchet: declared cross-file moves: 2, total ceilings down 477 lines.'",
    "pnpm check:refd-timer-probe :: exit 0",
    "pnpm check:skill-frame-sync :: exit 0",
    "pnpm check:watch-hint-literal :: exit 0",
    "beyond derivation: pnpm check:pm-settings-deny-roster :: exit 0 (roster under .claude); pnpm check:pm-governed-prose :: exit 0",
    "node scripts/pm/check-governed-merges.mjs --test (2 paths) :: 'GOVERNED — Tier S' (both .claude/**)"
  ],
  "line_budget": {
    ".claude/skills/pm-dispatch/references/execution-duties.md": "183 → 183 (ceiling 183). Diff +2/-2. No re-wrap; paid by deleting three things already stated elsewhere: the clause 「原子对 step 1 设」 (same file, atomic-pair line), the clause 「跨账号 assignee 不是你 ⇒ 永不碰」 (SKILL.md card-state row), and the line 「dev 每个可编译小步即 push …」 (os-dev.md 资源纪律 6 carries it). All changed lines ≤120 bytes (118, 107).",
    ".claude/agents/os-dev.md": "402 → 402 (ceiling 402). Diff +3/-3. Three lines rewritten in place (103, 116, 120 bytes)."
  },
  "files_changed": [
    ".claude/skills/pm-dispatch/references/execution-duties.md",
    ".claude/agents/os-dev.md"
  ],
  "deviations": [
    "The dispatch named worktree /home/user/objectstack-issue-20063 as cut from 3557f85, but it did not exist when I started. I created it from origin/main at 5b9402d, since main had moved on, and recorded BASE=5b9402d.",
    "PM route said the takeover-② line (execution-duties :78) should get a 「含该卡 open PR」 increment. That line is 118 bytes, and every spelling of the increment I measured came to 130+ bytes. So I stated the takeover half on the rewritten assignee line as 「接管 ② 同笔换」 and left :78 byte-identical.",
    "PM route hoped for a net 0 with no deletion. It came to +1 line in execution-duties, paid for by deleting a duplicate. My first candidate for the deletion was 「释放是显式动作 …」 (state-machine.md :41 repeats it). I rejected it before committing because scripts/pm/check-half-states.mjs quotes that exact line as RELEASE_ACT_RULE, the protocol's own words, and deleting it would leave that quotation with no source.",
    "Commit trailer: I used the model-free pair (Claude-Session + 'Co-authored-by: Claude') that AGENTS.md requires, not the model-named Co-Authored-By from the harness attribution reminder."
  ],
  "mcp_calls": "0",
  "api_writes": "4 relay strokes, each one repository_dispatch executed as objectstack-fleet[bot]: (1) pr_create → POST /repos/objectstack-ai/objectstack/pulls (draft); (2) label-write assign → POST /repos/objectstack-ai/objectstack/issues/20066/assignees [os-elon-musk]; (3) label-write add → POST /repos/objectstack-ai/objectstack/issues/20066/labels [skip-changeset]; (4) this report → POST /repos/objectstack-ai/objectstack/issues/20063/comments. git push ×2 (empty-branch probe, then commit 6364f11) is not REST.",
  "open_questions": [
    {
      "question": "AGENTS.md (Tier H, not touched) :405-406 says a dispatched executor 'never writes the assignee'. In its paragraph that means the card's, but read on its own it now contradicts os-dev.md rule 2's PR-assignee write. Should a separate maintainer-merged edit make it explicit?",
      "options": [
        "A: replace 'it ⛔ never writes the assignee' with 'it ⛔ never writes the card's assignee (the PR's assignee mirrors the card's, set by the executor at pr_create)'",
        "B: leave AGENTS.md as is; the paragraph is about claiming the card"
      ],
      "recommendation": "A. It is a single sentence, and it removes the one place where the Tier H text and the Tier S rule read differently. It needs its own PR with the maintainer's approval."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed (no class): the ruling does not cover the PR assignee when a card that still has an open PR is RELEASED without a takeover. The new rule moves it only on takeover. Recorded in the PR's Acceptance notes. carrier: 承接者:无"
  ]
}

Generated by Claude Code

ACCEPT — os-dev-report 5825053461 reviewed against references/review-checklist.md by the domain:skills#1 seat (session_01LNRg6FUMXYWgkAbDsfZpC8) at 2026-09-25T01:22Z. Tier S (.claude/** ×2, check-governed-merges --test GOVERNED — Tier S) ⇒ in-seat ## Contract review record (next comment) + every check green ⇒ this seat lands it through the fleet-write relay (pr_ready + automerge_enable), the landing route the maintainer chose today (objectstack#20048, 「B:走中继落地」).

PR form: draft PR #20066 on main, first line Fixes #20063, line 2 Clause-②: no — the only closing keyword; head 6364f11 (one commit, model-free trailers); assignee os-elon-musk (the rule this PR writes, applied to itself); labels documentation · size/s · skip-changeset (nothing publishes: .claude/**); 速读 draft present, session-URL footer intact.

Diff, read hunk by hunk (origin/main...6364f11, +5/−5 over 2 files):

.claude/agents/os-dev.md (402 / 402): rule 2's head 「assignee 归 PM」 → 「卡 assignee 归 PM」; its second sub-item → 「恒不写卡的 assignee(空也开工,报 summary);PR assignee = 卡的:--issue PR_NUMBER --assign LOGIN」; the write-budget line names the PR-assignee write inside the existing label-write slot. Three lines rewritten in place, 103 / 116 / 120 bytes.
references/execution-duties.md (183 / 183): the :56 rule becomes two lines — 「卡 assignee 归 PM,dev 恒不写;PR assignee = 卡的,接管 ② 同笔换;读者 is:pr is:open assignee:LOGIN」 and 「dev pr_create 同轮 label-write.mjs --issue PR_NUMBER --assign LOGIN;席位自有 PR 指派自己」 — the ruling's four parts (card assignee stays the PM's; PR assignee mirrors it, set at pr_create; takeover moves it; the seat's own PR; the reader) all present. Paid by deleting the line 「dev 每个可编译小步即 push …」 (os-dev.md 资源纪律 6 carries it verbatim: 「每个可编译小步即 commit + push」) and the two clauses 「原子对 step 1 设」 (the atomic-pair line ① four lines below) and 「跨账号 assignee 不是你 ⇒ 永不碰」 (SKILL.md's card-state row 「不是你的就永不碰」) — each checked by this seat to survive elsewhere; no re-wrap. The takeover line :78 byte-identical (the increment did not fit 120 bytes; the takeover half lives on the rewritten line — accepted).
The ruling's exclusions hold: no gate, no patrol row, no tool; AGENTS.md untouched.

Tests / gates (all at 6364f11, exit codes captured before any pipe): dispatch-gates --ran 20 / 20 derived families run, 0 NOT-MEASURED · pm-skill-ratchet 0 · pm-skill-id-lint 0 (34 clean) · skill-frame-sync 0 · doc-authoring 0 · agent-model-declared 0 · pm-governed-merges 0 · pm-governed-prose 0 · pm-settings-deny-roster 0 · nul-bytes 0 · commit-card-trailers 0 · governed-queue-guard self-test 279 · doc-formula-expressions 0 after the lock-guarded build (first attempt exit 3 = prerequisite, nothing measured). CI on the head at this ACCEPT: 15 success + 12 rostered skips, 2 in progress (Lint & Repo Gates, Type Check · workspace) — landing waits for them.

Deviations (four, all accepted): base 5b9402d rather than 3557f85 (main moved; both files' regions unchanged) · the takeover-② increment did not fit the 120-byte line — stated on the rewritten line instead, accepted · the deletion candidate 「释放是显式动作」 was correctly rejected (quoted as RELEASE_ACT_RULE by check-half-states.mjs) — the better reading · attribution per AGENTS.md.

Open question (the dev's), carried to the maintainer: AGENTS.md's claim paragraph (Tier H) says a dispatched executor 「⛔ never writes the assignee」 — in context the card's, but read alone it now contradicts os-dev.md rule 2's PR-assignee write. Recommendation A (one sentence: 「never writes the card's assignee (the PR's assignee mirrors the card's, set by the executor at pr_create)」) needs its own Tier H PR — put to the maintainer in the round report, ⛔ not filed by this seat without their word.

Finding (the dev's, judged): a card released WITHOUT a takeover while its PR is still open — the rule moves the PR assignee only on takeover; Acceptance notes only, not filed (the release line already clears the card's assignee; the PR follows on the next claim).

Landing from here: flip 定点 after the two jobs finish; relay pr_ready → automerge_enable (one actions file) → followed to MERGED; on MERGED the card closes by Fixes, then pm:dispatched and the assignee come off.

Four axes, one line each: business need — measured (the maintainer could not tell who handles a fleet-authored PR; the assignee filter answers it) · long-term — the same identity record the card already carries, no second field · AI-error — one line each in the two files every dev and seat reads, no fallback path · startup scope — five lines rewritten, zero added, no gate.

Generated by Claude Code

Contract review

Served-tier: CONTRACT_REVIEW_TIER
Head-sha: 6364f11c08248fa8a4d30902a108c4bfd05010db

① Derived judgments

The diff rewrites rule text in .claude/agents/os-dev.md (a MANDATORY-tier protocol file, built at the default tier under the one-line-class exit and reviewed here) and .claude/skills/pm-dispatch/references/execution-duties.md; no accepted set, schema or export changes: +5/−5, two governed Tier S files, skip-changeset correct (nothing publishes). Correct.
The rule as written matches the ruling 「同意」 on the seat's recommendation, part by part: card assignee stays the PM's and the dev never writes it (os-dev rule 2; execution-duties :56); PR assignee = the card's, set by the dev at pr_create with label-write.mjs --issue PR_NUMBER --assign LOGIN (both files); takeover ② moves it in the same stroke (execution-duties :56); a seat's own PR is assigned to the seat (:57); the reader is:pr is:open assignee:LOGIN (:56). The spelling is the one measured on objectui#10410 / docs(qa): rewrite the cli.migrate-meta-codemod P1 to what os migrate meta actually does #10412 / /api/v1/analytics/query still drops per-measure and dataset-level filter on the ObjectQL path — engine.aggregate receives no filter at all #10413 / objectstack#20066 through the relay's assign op. Correct.
Nothing the ruling excluded was added: no gate, no patrol row, no tool, AGENTS.md untouched. Correct.
Line-ratchet payment: execution-duties.md 183 → 183 by deleting one line and two clauses whose content survives elsewhere (os-dev.md 资源纪律 6; execution-duties ①; SKILL.md card-state row) — verified by this seat; no re-wrap. os-dev.md 402 → 402, three lines in place ≤ 120 bytes. Correct.
Reader check (skills-lane): every changed line has a named reader — the dev at pr_create (os-dev.md), the seat at claim and takeover (execution-duties.md), the maintainer's PR filter. No token tax.

② Semver level

None owed: nothing published changes; consistent with Clause-②: no and skip-changeset.

③ Boundary flags

open_questions: one — AGENTS.md's claim paragraph 「never writes the assignee」 now reads narrower than the rule; Tier H, carried to the maintainer in the round report (ACCEPT names it). Not a blocker for this PR: AGENTS.md's paragraph is about the card, and AGENTS.md outranks the skill text where they conflict, so a dev reading both still writes only the card's assignee never — and the PR assignee write is what os-dev.md now instructs.
Not a widening: no accepted set touched.

Implemented-by: claude/issue-20063-pr-assignee-mirrors-card
Reviewed-by: session_01LNRg6FUMXYWgkAbDsfZpC8

VERDICT: PASS

Generated by Claude Code

docs(pm-dispatch): a dispatched PR carries its card's assignee — set …

docs(agents): narrow the executor never-writes-assignee clause to the…

Navigation Menu

Repository navigation

skills(pm-dispatch): a dispatched PR carries its card's assignee (the seat's linked user), set at pr_create and moved on takeover — maintainer 2026-09-25 「同意」 #20063

Maintainer's words (verbatim, this seat's chat, 2026-09-24/25, ⛔ not translated)

What lands (rule text only; ⛔ no gate, no patrol row, no new tool)

Who acts

Dedupe

Contract review

① Derived judgments

② Semver level

③ Boundary flags

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

skills(pm-dispatch): a dispatched PR carries its card's assignee (the seat's linked user), set at pr_create and moved on takeover — maintainer 2026-09-25 「同意」 #20063

Description

Maintainer's words (verbatim, this seat's chat, 2026-09-24/25, ⛔ not translated)

What lands (rule text only; ⛔ no gate, no patrol row, no new tool)

Who acts

Dedupe

Activity

objectstack-fleet commented on Sep 25, 2026

objectstack-fleet commented on Sep 25, 2026

objectstack-fleet commented on Sep 25, 2026

objectstack-fleet commented on Sep 25, 2026

Contract review

① Derived judgments

② Semver level

③ Boundary flags

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions