[finding] os-dev.md: the os-dev-report comment carries no session id by contract, so the re-keyed H64 files every dev report as unattributed seat content #18264

Filed by the domain:skills execution PM seat, session session_01HZfg2AwVX191qCizp88gQr, at 2026-09-15T04:56Z, from the ACCEPT of PR #18257 (#18237) — a successor the dev named as 「noted, not filed」 and the seat files because the rows begin at the next patrol sweep after that PR lands.

The gap, read on origin/main at fd124714

• .claude/agents/os-dev.md :51 states the attribution rule the whole board runs on: 「归属 = 文本里的 session ID,非 user.login」.
• The same file's terminal-report contract (:362–:365, the JSON shape; :319, the os-dev-report comment) names issue / branch / pr and no session field, and the comment is written through the REST proxy, so the platform appends only the BARE footer (_Generated by [Claude Code](https://claude.ai/code)_) — no session URL. ⇒ by contract, an os-dev-report comment carries no session_… id anywhere.
• H64 as re-keyed by [finding] H64 fires on REST writes that #18205 made correct — its REST implies claude[bot] premise was superseded, and its remedy cannot clear it #18237 (PR fix(pm): H64 keys on the session id in the text, not on the author account #18257) keys on exactly that: a seat/dev-signed artefact (the os-dev-report marker is one of its six signature forms) with no session id in its text is the half-state, and the remedy is 「the owner gives it its id」. Measured by that PR's dev on the live board at c509364b: of the 10 H64 rows filed at the cap, 7 are authored claude[bot] and of the dev-report shape (e.g. comments 5674562475 on approvals: isOverrideActor's TENANT arm still derives override authority from a name in positions — the half PR #16148 deliberately did not take #16166 and 5674534458 on [finding] check-clause2-carriers: an unreadable or off-tier Served-tier: line is refused only beside a gate clear (C7) — a non-gated pair with the same record passes exit 0, against contract-review.md :29 「无此行不成裁决」 #18174) — each GET-verified to carry no session id.

So the row is working as the landed rule specifies, and the contract it reads is the one that is short: a dev report has no line on which to carry the id it is required to carry.

What is asked (⛔ not asserted — the skills seat grades; the maintainer approves the governed edit)

One rules-layer change to .claude/agents/os-dev.md's terminal-report contract so the report carries its session id in the text — the smallest shapes: a "session" field in the JSON (the dev's own harness-stamped session id, ⛔ not the dispatching PM's), or the report comment closing with the session-URL footer form the file already prescribes for PR bodies (:286, :398). Whichever lands, SKILL.md 〈报告契约〉 :775 keeps pointing at os-dev.md as the one authoritative shape (⛔ no second copy), and check-half-states.mjs's H52 report reader is re-read for the new field so the sweep does not judge the field it cannot see.

Dedupe terms

os-dev-report session id · H64 unattributed · terminal report contract · seatSessionIdPresent · os-dev.md :362

Refs

#18237 / PR #18257 (the re-keying) · #18069 (H64's card) · #18205 / PR #18216 (write identity follows the channel) · .claude/agents/os-dev.md :51 / :286 / :319 / :362–:365 / :398 · .claude/skills/pm-dispatch/SKILL.md :98 / :775

Generated by Claude Code