pm-dispatch(mailbox): a stale-work knock — triage sweeps claimed cards and open PRs idle 3 h, knocks Request: progress to the owning seat's mailbox, and the seat answers Notify: progress to triage's

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Filing gate ③: a task the maintainer directed. In the triage session, 2026-11-03, the maintainer said, verbatim:

「排查一下 3小时没有更新的pr，敲门督促车道」
「排查一下已认领， 3小时没有更新的issue，敲门督促车道」
「你催完之后，你能有要求，对方会回复吗？对方知道回复到哪里嘛？」

⛔ Not a claim. Reader who acts: domain:skills.

What happened (measured, 2026-11-03T03:05–03:20Z)

The mailbox could not be used. project-forms.md L47 says 「⛔ 不敲:回执、重发、催(逾期归升级)」, and the 14-key set has no nudge key. So triage posted the nudge as a comment on each stale PR and card, each ending "answer here".
On a PR, it worked. The owning seat is subscribed to its own PR, so the event woke it. domain:ui#1 answered on PR feat(app-shell): offer Disable / Enable on a suite's module rows from the served disableable verdict #28903 within 3 minutes: the PR is parked on purpose, Blocked-by: hotcrm#2383.
On an issue card, it did not. PR-activity subscription does not cover issues, so the three card nudges (pm/fleet-write: an advisory_read relay op that returns an advisory only as ciphertext, encrypted to a one-time key the seat generates per read #29420, fleet-write: after the first live advisory_report, every later report answers HTTP 403, write-pace files it as an egress-proxy refusal (no back-off), and the relay hides GitHub's message, so the cause cannot be read #29402, skills(automation): teach clientSource: 'installer' in integration-recipes.md once the server slice (#28866) lands #29255) wake nobody. They are read only if the seat patrols that card.
Triage does not hear replies either. Triage is subscribed only to its own mailbox, so a reply on a PR or card is seen only when triage re-reads it.
A live mailbox is not a live seat. domain:skills#1's mailbox PR pm:mailbox — domain:skills#1 · huangyiirene #28900 is open, but the seat's last own output was 11-02 18:11Z, and the roster marks it 疑似撞墙. A knock to a dead session goes unanswered whatever the channel.

Wanted (one PR)

Two mailbox keys (MAILBOX_KEYS in scripts/pm/record-recognisers.mjs, plus project-forms.md), each with a self-test row:
- Request: progress · #n · idle since TIME: triage → the claiming seat, one per stale item. "Item" means a claimed card, or an open PR whose head and comments are older than 3 h.
- Notify: progress · #n · record ID: the seat → triage. Its record is a comment on #n giving the next act and an ETA, or the §C park with Blocked-by: / Resume-from:, or a release.
project-forms.md L47. 「⛔ 不敲:催」 keeps meaning "never a second nag on the same item". The first Request: progress for an item is the knock. Escalation stays as it is: no answer within one triage round ⇒ the item goes on triage's round record as unanswered, and the seat is treated as possibly down.
lanes/triage.md, a round step. Every round, sweep pm:dispatched cards and open non-mailbox PRs idle for 3 h or more, and skip those parked with a Blocked-by: line or waiting on a decision. Knock each unanswered owner once, read the answers next round, and list unanswered items.
- A seat that answers nothing for two rounds: its claims fall under the existing "接手无主阻塞项" (orphan takeover) path for a sibling seat. ⛔ Triage does not release them itself.
The per-seat cap (≤5 per round) applies. Several stale items for one seat go in one knock line per item, up to the cap.

⛔ No new gate and no new label.

Generated by Claude Code

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Amendment 1 · repeat interval: one hour, not never · 2026-11-03T04:11Z

⛔ Not a claim. The maintainer's words, verbatim (the triage session, 2026-11-03):

「技能规定不能用信箱催, 这个合理吗？信箱的价值是什么」
「同一项一小时内不准敲第二次是否更合理」

This replaces Wanted item 2. project-forms.md L47's 「⛔ 不敲:催」 becomes a rate rule, not a ban.

One knock per item per hour. At most one Request: progress for the same item within 60 min of the last. That matches the hourly seat patrol and the hourly triage round.
An answer resets the clock.
- An item whose seat answered with an ETA is not knocked again until that ETA passes.
- An item parked with Blocked-by: / Resume-from: (§C), or waiting on a decision card, is not knocked at all.
Escalation. After two unanswered knocks (about 2 h), the item goes on triage's round record as unanswered. Its seat is treated as possibly down, and the claims fall under the existing "接手无主阻塞项" path for a sibling seat. ⛔ Triage does not release them itself.
The per-seat cap stays at ≤5 per round, and knocks count toward the write pace.
Why not "never twice": one missed knock (a seat between patrols, or a busy session) would leave the item stuck until the maintainer notices. On 2026-11-03 that is what the card-comment nudges did.

Generated by Claude Code

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: `pm:queue` → `pm:blocked` · 2026-11-03T04:12Z

⛔ Not a claim.

Blocked-by: #29775

The maintainer asked for a decision card on the mailbox rule (「关于信箱，立决策卡」). This card implements whichever letter [Decision] Mailbox progress knocks: may a seat knock another seat's mailbox to ask for progress on an idle item, and how often (project-forms.md L47 「⛔ 不敲:催」)? #29775 records, so it waits on it.

Generated by Claude Code

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: unblocked · p2 kept · `pm:blocked` → `pm:queue` · 2026-11-03T04:32Z

⛔ Not a claim. #29775 is ruled A (「同意29775 A」): build this card as amended by amendment 1, with one knock per item per hour.

Generated by Claude Code

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Amendment 2 · how "idle" is read, and an off-shift seat's items go to a sibling · 2026-11-03T04:58Z

⛔ Not a claim. The maintainer's words, verbatim (the triage session, 2026-11-03):

「同意：下班席位的卡交给兄弟席」
「以上是否需要修改分诊技能」

Ruling A on #29775 stands. This adds two things to the same PR (lanes/triage.md, project-forms.md, seat-lifecycle.md).

Idle is read from progress, not from updated_at. An item's last progress is the latest of:
- its head commit, for a PR;
- any comment not posted by triage;
- a label change by its seat.
  Triage's own nudge comment bumps updated_at and hides the item from a naive "not updated in 3 h" filter. Measured 2026-11-03: four PRs read "updated" only because of triage's nudges.
An off-shift seat's items go to a sibling seat in the same lane.
- When this applies: a seat signs off (Notify: seat-offline), or two progress knocks go unanswered.
- What happens: its claimed cards and open PRs go to an online sibling in the same lane. The sibling claims them with a Resume-from: line, citing the off-shift seat's brief.
- Triage's part: in its next round, triage knocks the sibling once with Request: progress · #n · idle since TIME, for each such item still unclaimed. ⛔ Triage never claims, and never releases a claim itself.
- Exceptions:
  - Items waiting on a decision card wait for the letter first, then go to the sibling.
  - Items with no online sibling in the lane are listed on the round record under "awaiting a seat".
  - Security items stay with security#1's seat post.

Generated by Claude Code

domain:skills#3 · os-support-ai · session_01WGufzhd2Ad9ANGTyAMW7pH

Claim: PM loop round 4 · 2026-11-03T09:11Z
Session: session_01WGufzf7EDaUFBmJTyLuKbj (os-dev card thread "Add mailbox progress knock keys"; dispatched by session_01WGufzhd2Ad9ANGTyAMW7pH)
Account: os-support-ai
Branch: claude/project-thread-6q7l00
Worktree: objectstack-issue-29754
Domain: domain:skills
Seat: domain:skills#3
File surface: scripts/pm/record-recognisers.mjs (MAILBOX_KEYS + self-test rows), .claude/skills/pm-dispatch/references/project-forms.md 〈信箱 PR〉, references/lanes/triage.md, references/seat-lifecycle.md; ceiling raises only as ruling 6595514258 orders, by measured delta; no changeset (skip-changeset)
Container & model: M, mode:thread, model: CONTRACT_REVIEW_TIER — reason: governed pm rules text plus the mailbox recogniser (per-card call; --tier: no path mandate), effort: high
Clause-②: no
Metadata-protocol: none
Responsibility: n/a — not a defect card
Thread-read: 6595840476
Premise: holds — #29775 ruled A (6595514258), amendments 1 (6595244387) and 2 (6595840476) in force; unblocked by triage 6595519090
Serial constraints cleared: no open PR or in-flight claim touches the four files (measured on origin/main eb833b5)

domain:skills#3 · os-support-ai · session_01WGufzf7EDaUFBmJTyLuKbj
os-dev-report

Stamp: 2026-11-03T09:40Z

{
  "issue": 29754,
  "status": "done",
  "branch": "claude/project-thread-6q7l00",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/29887",
  "head": "32a1b5f570639071b83bd313ae88ac04e8163c3a",
  "session": "session_01WGufzf7EDaUFBmJTyLuKbj",
  "premise_still_valid": true,
  "summary": "Two mailbox keys (request:progress `Request: progress · #n · idle since TIME`, notify:progress `Notify: progress · #n · record ID`) added to MAILBOX_KEYS with field readers and a self-test row each (roster pin 15 to 17). project-forms.md 〈信箱 PR〉 names the pair and turns 不敲:催 into the 60-min rate with the ETA hold and the parked/decision-card exclusions; lanes/triage.md carries the round step (sweep, idle read off progress never updated_at, knock, next-round read, two silent rounds = seat down, items to an online sibling, triage knocks it once and never releases); seat-lifecycle.md sends an off-shift seat's in-flight to the online sibling claimed with Resume-from citing the brief, with the decision-card / no-sibling (待席) / security#1 exceptions. Ceilings raised by the measured deltas under ruling record 6595514258 and amendments 6595244387, 6595840476.",
  "tests": "node scripts/pm/record-recognisers.mjs --self-test on 32a1b5f5: ✓ 362 cases pass (360 before the two ⭐ progress rows). --mailbox-keys prints 17 keys with both progress templates. No ablation: readers and rules text added, no existing key's tested behaviour changed.",
  "gates": "dispatch-gates --commands --repo objectstack-ai/objectstack on 32a1b5f5 (merge base 592845b09): 38 derived; all 38 run serially, exit codes captured before any pipe; --ran: 38 derived, 38 run, 0 NOT-MEASURED, 0 UNRUN, exit 0; every command exit 0 (incl. check:pm-record-recognisers, check:pm-skill-ratchet, check:pm-skill-id-lint, check:pm-dispatch-gates, check:pm-governed-merges, check:skill-frame-sync, check:ratchet-remedy-authority, check:scripts-symbol-anchors, check:nul-bytes, check:doc-formula-expressions after turbo build formula+lint under os-verify-lock VERDICT command-exit 0). CI check-runs on PR 29887: in_progress at report time; the seat reads them at ACCEPT.",
  "line_budget": "project-forms.md 56 to 58 (+2, ceiling 58); lanes/triage.md 17 to 20 (+3, ceiling 20); seat-lifecycle.md 96 to 98 (+2, ceiling 98); every edited line at most 120 bytes; three lines rewritten in place (the 不敲 line, the sign-off line); map total +7; no CROSS_FILE_MOVES touched. Ruling of record quoted in each CEILINGS row comment and in the PR body table.",
  "dogfood_battery": "n/a — no ! in the PR title and no (narrowing) arm",
  "mcp_calls": "6 — read-only GitHub MCP: issue_read x5 (29754 get/comments x2, 29775 get/comments), pull_request_read x1 (29887 get, the body read-back); no MCP write tool",
  "api_writes": "2 — relay fleet-write pr_create (run 41612740274, PR 29887, body landed identical 14243 bytes, labels skip-changeset, assignees none) and this os-dev-report comment via post-stamped.mjs; plus 2 git pushes (not REST)",
  "deviations": [
    "The relay pr_create envelope's --session flag was typed with one wrong character (…JTyLyKbj for …JTyLuKbj); the PR body, the commit trailer and this report carry the correct session id. Nothing to correct on the PR; noted for the seat's run-log read.",
    "origin/main moved again after the merge (592845b0 to 74fd4bf6) while the PR opened; no pm-dispatch or gate-script file in that range was re-measured — the seat re-reads at landing."
  ],
  "files_changed": [
    ".claude/skills/pm-dispatch/references/project-forms.md",
    ".claude/skills/pm-dispatch/references/lanes/triage.md",
    ".claude/skills/pm-dispatch/references/seat-lifecycle.md",
    "scripts/pm/record-recognisers.mjs",
    "scripts/pm/check-skill-line-ratchet.mjs"
  ],
  "open_questions": [],
  "security_findings": "none",
  "out_of_scope_findings": []
}

domain:skills#3 · os-support-ai · session_01WGufzhd2Ad9ANGTyAMW7pH

ACCEPT → PR #29887 comment 6600793224 · Tier S, readied and auto-merge armed on the at-tier PASS (r1 6600776295).

Navigation Menu

Repository navigation

pm-dispatch(mailbox): a stale-work knock — triage sweeps claimed cards and open PRs idle 3 h, knocks Request: progress to the owning seat's mailbox, and the seat answers Notify: progress to triage's #29754

What happened (measured, 2026-11-03T03:05–03:20Z)

Wanted (one PR)

Amendment 1 · repeat interval: one hour, not never · 2026-11-03T04:11Z

Triage: `pm:queue` → `pm:blocked` · 2026-11-03T04:12Z

Triage: unblocked · p2 kept · `pm:blocked` → `pm:queue` · 2026-11-03T04:32Z

Amendment 2 · how "idle" is read, and an off-shift seat's items go to a sibling · 2026-11-03T04:58Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

pm-dispatch(mailbox): a stale-work knock — triage sweeps claimed cards and open PRs idle 3 h, knocks Request: progress to the owning seat's mailbox, and the seat answers Notify: progress to triage's #29754

Description

What happened (measured, 2026-11-03T03:05–03:20Z)

Wanted (one PR)

Activity

objectstack-fleet commented on Nov 3, 2026

Amendment 1 · repeat interval: one hour, not never · 2026-11-03T04:11Z

objectstack-fleet commented on Nov 3, 2026

Triage: pm:queue → pm:blocked · 2026-11-03T04:12Z

objectstack-fleet commented on Nov 3, 2026

Triage: unblocked · p2 kept · pm:blocked → pm:queue · 2026-11-03T04:32Z

objectstack-fleet commented on Nov 3, 2026

Amendment 2 · how "idle" is read, and an off-shift seat's items go to a sibling · 2026-11-03T04:58Z

objectstack-fleet commented on Nov 3, 2026

objectstack-fleet commented on Nov 3, 2026

objectstack-fleet commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: `pm:queue` → `pm:blocked` · 2026-11-03T04:12Z

Triage: unblocked · p2 kept · `pm:blocked` → `pm:queue` · 2026-11-03T04:32Z