pm-dispatch skill: on a private repository (objectstack-ai/cloud), a security finding is an ordinary issue labelled security, not a draft advisory

repo:cloud#2 · os-zhuang · session_0123kRqqdpAJWYMgsgdAJX4G

Filing gate ③, a task on the maintainer's word. Sam, 2026-11-04 14:45Z in the cloud seat 2 thread: 「cloud 私有仓库可以直接用带 security 标签的普通 issue，立卡修改 skills」.

Reader: the domain:skills lane (.claude/skills/pm-dispatch/references/security-findings.md, plus any file that points to it). Suggested priority p2; triage decides.

维护者速读

现在 pm-dispatch 规定，安全发现「永不立卡」，只能写成仓库的草稿安全公告。但中继对 cloud 建公告会返回 404，所以 cloud 的发现一直是在对话里口头交给你，修复走无卡派发令。你的裁定是：cloud 这类私有仓库，直接用带 security 标签的普通 issue。这张卡就是把 skill 里的规则改成这样。

Ruling to encode

For a finding whose affected repository is private (today: objectstack-ai/cloud):

The private record is an ordinary issue in that repository, carrying the security label. It is filed through the relay's issue-create like any card, and it holds the mechanism.
The fix is dispatched from that issue like any card, at effort xhigh. It needs no card-less ORDER and no draft advisory.

Unchanged for public repositories (objectstack, objectui, hotcrm, the docs and site repos): a draft advisory through the relay, never an issue (§2–§3 as written).

Work

security-findings.md. Add the private-repository branch above. Name how a seat tells the two cases apart: the repository's private flag, read from the API at filing time, never recalled. Update §2's "⛔ 不立 issue" and §3's fallback A so they apply to public repositories only, and keep §8 ("一处成文"). security already exists as a label on cloud, so no new label is needed.
What still stays rule-level on a private repository. Pick one answer and write it down, because a private repository can be made public later and its history goes with it. Recommended: the issue holds the mechanism; PR titles, PR bodies, commit messages and changesets still state only the rule (§6), and link the issue.
The relay log. fleet-write runs on this public repository. Before the rule ships, confirm that an issue-create stroke's body never reaches the public run log (the validator, the executor rows and the summary), and pin that with a test in scripts/pm/fleet-write. If any part of the body can reach the log, the rule must say how the mechanism gets into the issue without passing through it.
Update every file that points to security-findings.md (gate, lanes, os-dev, checklist-test) only where a pointer's wording would now be wrong. Restate nothing.

Done when

security-findings.md carries the private-repository branch, and the skill's line and byte limits pass.
The relay-log question is answered and pinned by a test.
A cloud seat that reads the skill files a cloud finding as a security issue.

Dedupe

repo:objectstack-ai/objectstack is:issue security-findings private: 166 hits. None of them changes where a private repository's finding is recorded. The nearest are:

pm/fleet-write: the relay publishes a security advisory on the maintainer's list: reopen, set the affected and patched versions, publish #30110 (the relay publishes advisories): about public advisories, not private repositories.
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 (advisory_report 403s): a relay defect.
pm-dispatch(security): one fleet-wide rule for security findings — never a card, never a public mechanism; tracked as a GitHub draft security advisory, fixed by a no-card pm-dispatch order #29111 and pm-dispatch(security): a dev, seat or QA runner that meets a security finding reports it through the security channel and never fixes it in place — security-findings.md states it once, os-dev.md points to it #29731 (closed): the rule this card amends.

The security seat (#29740, security#1) owns findings end to end, so tell it about this change.

triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U

Path: none

Triage: first-touch audit · p2 kept · `domain:skills` kept · `pm:queue` kept · 2026-11-04T14:55Z

⛔ Not a claim.

Read fresh. A rule the maintainer gave (「cloud 私有仓库可以直接用带 security 标签的普通 issue」): on a private repository, a security finding is an ordinary issue labelled security, filed through the relay. Public repositories keep the draft-advisory rule. p2 kept.
Serial with pm/fleet-write: the relay publishes a security advisory on the maintainer's list: reopen, set the affected and patched versions, publish #30110 and pm-dispatch skills: the security seat's triage of a non-finding, its ledger states after the close-on-merge order, a release's contents read by ancestry, and the CI recoveries the relay leaves #30111 on security-findings.md.

Generated by Claude Code

domain:skills#1 · os-elon-musk · session_016WV4d1mkgEris4MHHLjtnk

Claim: PM loop round 1 · 2026-11-05T22:53Z
Session: session_016WV4d1mkgEris4MHHLjtnk
Account: os-elon-musk (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/project-thread-shi9ig (the card thread's designated branch; dev session session_016WV4cxfNdCqsVzihQoX3BE, amended by the seat)
Worktree: objectstack-issue-30114
Domain: domain:skills
Seat: domain:skills#1
File surface: .claude/skills/pm-dispatch/references/security-findings.md (the private-repository branch; §2 and §3 fallback A scoped to public repositories; §6 rule-level for PR titles, bodies, commits and changesets on a private repository too); pointer wording only where it would now be wrong in .claude/skills/pm-dispatch/references/lanes/security.md, .claude/agents/os-dev.md and other files that point to security-findings.md; scripts/pm/fleet-write/** tests only (the pin that an issue_create body never reaches the public run log: validator, executor rows, summary). ⛔ A behaviour change in fleet-write itself: stop and report (stop on breach)
Container & model: M, mode:thread, model: TIER_DEFAULT (dispatch-gates --tier: no path-derived mandate), effort: high
Clause-②: no
Metadata-protocol: none
Responsibility: n/a — not a defect card (a maintainer ruling to encode: 「cloud 私有仓库可以直接用带 security 标签的普通 issue，立卡修改 skills」)
Thread-read: 6633882869
Premise: holds: security-findings.md §2 still reads 「⛔ 不立 issue」 for every repository on origin/main 788715ef; #30110 and #30111 (same file) are closed
Serial constraints cleared: #30110 and #30111 closed; no open PR touches security-findings.md (open PRs read at claim time). Next on that file: #30271 (serial)

Tier S (.claude/**) for the rules text; the seat lands it on an at-tier contract-review PASS with every check green. The security seat (security#1) is told at landing.

Generated by Claude Code

domain:skills#1 · os-elon-musk · session_016WV4d1mkgEris4MHHLjtnk

Release: session_016WV4d1mkgEris4MHHLjtnk · seat clock-out (maintainer 「skills 处理完当前任务就下班」, 2026-11-06, project chat) · back to pm:queue for domain:skills#2 · 2026-11-06T07:53Z

Claim 6671655834 is released whole. Nothing is in flight on it, and no PR is open.

Done so far: the dev measured item 3 (relay log). Answer: yes, on one channel. A platform refusal whose message quotes the body (for example a 422) reaches the public run log unfiltered: requestLanded returns HTTP {status} — {json.message} (scripts/pm/fleet-write/execute.mjs ~L593). It lands in the row, in the ✗ action log line and in both step-summary lines. Every other channel withholds the value.
Waiting on: a maintainer ruling, put to the maintainer on 2026-11-05 and not yet answered. Option A (recommended): fix the relay in this card by redacting or withholding json.message on a refusal, then hold pm-dispatch skills: say who confirms a pending-changeset correction (the seat's record, never a maintainer wait) everywhere a dev or the gate is told to "get it confirmed", and keep PR bodies from narrating pre-fix behaviour on security fixes #30271 behind it. The next seat asks for the ruling again before any code; ⛔ no autonomous ruling.
Resume-from: claude/project-thread-shi9ig@5274bbc6 (pushed, no PR). The local gate log gates-ran-5274bbc6.txt and the full reading report-30114-relay-log.md are in the seat folder skills-lane-seat1/30114/ of the fleet project files.
Serial: pm-dispatch skills: say who confirms a pending-changeset correction (the seat's record, never a maintainer wait) everywhere a dev or the gate is told to "get it confirmed", and keep PR bodies from narrating pre-fix behaviour on security fixes #30271 waits on this card.

Generated by Claude Code

domain:skills#1 · os-elon-musk · session_016WV4cxfNdCqsVzihQoX3BE

Handoff: maintainer ruling recorded · 2026-11-07T04:52Z

⛔ Not a claim. This is a record only. The card stays in pm:queue for the next domain:skills seat.

Ruling. On the relay-log question (comment 6680051129), the maintainer picked option A2 on the decision card in the card thread, verbatim: 「A2 本卡里一起修中继」. They then added: 「你已经下班了，只需要记录」. So no work was done in that session.

What A2 means for the next seat:

This card also fixes the relay. When a request fails and the platform's message quotes the payload, the generic failure path in scripts/pm/fleet-write/execute.mjs (requestLanded) withholds that message. Today it prints the message unfiltered into the row, the action log line and both step-summary lines. The advisory rows already withhold such a message through advisoryRefusalCause. Pin the fix in the executor's self-test.
It lands in the same PR as the security-findings.md private-repository rule.
By this ruling, the claim's file surface widens to relay behaviour in scripts/pm/fleet-write/execute.mjs and its self-test. The new claim must name that surface.

Resume-from: claude/project-thread-shi9ig@5274bbc6 (pushed, no PR). It already carries:

the rule text, in the final form where the mechanism rides issue_create
the pointer edits (filing gate, security lane charter, lane table)
the line-ratchet raise 64 → 67, citing this card's body as the ruling of record
an executor self-test battery pinning the three channels that never print the title or body, mutation-proven

All 46 derived gates and pnpm lint were green on that head. Merge origin/main before resuming.

Serial: #30271 waits on this card.

Generated by Claude Code