Context
Operator-named friction item (2026-06-12, ~04:20): "why do you not check, if a PR was already merged first?" — after TWO same-shape instances within an hour, both timeline-verified:
- PR #12950 merged
03:18:45Z; the author's [merge-eligible] — at the human merge gate broadcast landed 03:19:00Z — 15 seconds after the merge, announcing stale state to the whole swarm.
- PR #12956: the approval relay (~04:12) read the reviewer's A2A and checked
reviewDecision — but never state; the merge landed 04:13:50Z, seconds around the "waiting only on your merge" report. Undetectable staleness by construction.
Friction-capture per the self-evolving-systems discipline — task: relaying PR lifecycle signals (approval → merge-eligibility, review posts, follow-up handling); rule gap: pr-review-guide.md §2.1 mandates the state: OPEN check before reviewing only — no protocol mandates live-state before relaying, broadcasting, or acting on PR lifecycle signals; cost: stale board signals to peers + operator reading "at the merge gate" for things he already merged + correction cycles (two tonight); safer alternative: one gh pr view <N> --json state,mergedAt immediately before any PR lifecycle action or report.
Release classification: boardless — process substrate; not release-blocking.
The Problem
The merge-on-approval policy is the (correct, operator-chosen) forcing function: the faster merges happen, the staler every A2A/wake signal is by the time an agent reads it. Wake-driven flows compound it: A2A says "approved" → agent relays "merge-eligible" minutes later without re-reading live state. The existing freshness discipline exists at exactly one beat (review-start) and is absent from all the others (approval relays, re-stamp requests, merge-eligibility broadcasts, follow-up filing keyed to merge state, post-review-pickup cycle decisions).
The Architectural Reality
.agents/skills/pr-review/references/pr-review-guide.md — §2.1 (the existing OPEN-gate, review-start only), §10 (A2A comment-ID hand-off — the relay beat with no state check), §11 (post-review pickup pointer)..agents/skills/post-review-pickup/references/post-review-pickup-workflow.md — the lifecycle-cycle matrix consumes PR signals (own-PR-green → request-review, designated review, etc.) without a mandated freshness read.- The check is one cheap read-only call; it composes with the existing Verify-Before-Assert core value (a PR-state claim in any artifact IS a factual claim — the falsifying tool is
gh pr view --json state,mergedAt).
The Fix
Amend the two payloads (Progressive Disclosure: rule body in the reference payloads, trigger-line only where always-loaded):
pr-review-guide.md §10 gains the freshness gate: before relaying any review outcome or posting any merge-eligibility claim (A2A, broadcast, chat report), run gh pr view <N> --json state,mergedAt; if MERGED/CLOSED, relay the TERMINAL state instead and skip stale-eligibility language.post-review-pickup-workflow.md cycle entries that consume PR signals reference the same gate (signal = recovery hint; live state = the work gate — mirroring the context-recovery skill's existing framing).- Substrate Accretion Defense: net-add is ~6 lines across two conditionally-loaded payloads; no always-loaded bytes. Sunset: if a mechanical guard (e.g., an MCP-side state echo in A2A lifecycle messages) lands later, the manual rule compresses to a trigger.
Acceptance Criteria
Out of Scope
- Mechanical enforcement (A2A schema changes, hooks, daemons) — file separately if the manual rule proves insufficient.
- The verdict-not-enum lesson (report the §9 Strategic-Fit verdict, not
reviewDecision) — same family, already absorbed into the same §10 beat; the amendment may fold both in one sentence if natural.
Related
PR #12950 / PR #12956 (the two instances), #12856 (the thundering-herd two-phase-claim precedent — same check-at-last-moment epistemics).
Live latest-open sweep: checked latest open issues at 2026-06-12T04:24Z; no equivalent found. A2A in-flight claim sweep: lane-claim broadcast 7e9f5c98 sent first, uncontested.
Origin Session ID: e605ce21-3668-445c-bc00-45896aa9a092
Retrieval Hint: "PR state freshness check merge-eligible stale relay merged first"
Context
Operator-named friction item (2026-06-12, ~04:20): "why do you not check, if a PR was already merged first?" — after TWO same-shape instances within an hour, both timeline-verified:
03:18:45Z; the author's[merge-eligible] — at the human merge gatebroadcast landed03:19:00Z— 15 seconds after the merge, announcing stale state to the whole swarm.reviewDecision— but neverstate; the merge landed04:13:50Z, seconds around the "waiting only on your merge" report. Undetectable staleness by construction.Friction-capture per the self-evolving-systems discipline — task: relaying PR lifecycle signals (approval → merge-eligibility, review posts, follow-up handling); rule gap:
pr-review-guide.md §2.1mandates thestate: OPENcheck before reviewing only — no protocol mandates live-state before relaying, broadcasting, or acting on PR lifecycle signals; cost: stale board signals to peers + operator reading "at the merge gate" for things he already merged + correction cycles (two tonight); safer alternative: onegh pr view <N> --json state,mergedAtimmediately before any PR lifecycle action or report.Release classification:boardless — process substrate; not release-blocking.The Problem
The merge-on-approval policy is the (correct, operator-chosen) forcing function: the faster merges happen, the staler every A2A/wake signal is by the time an agent reads it. Wake-driven flows compound it: A2A says "approved" → agent relays "merge-eligible" minutes later without re-reading live state. The existing freshness discipline exists at exactly one beat (review-start) and is absent from all the others (approval relays, re-stamp requests, merge-eligibility broadcasts, follow-up filing keyed to merge state, post-review-pickup cycle decisions).
The Architectural Reality
.agents/skills/pr-review/references/pr-review-guide.md— §2.1 (the existing OPEN-gate, review-start only), §10 (A2A comment-ID hand-off — the relay beat with no state check), §11 (post-review pickup pointer)..agents/skills/post-review-pickup/references/post-review-pickup-workflow.md— the lifecycle-cycle matrix consumes PR signals (own-PR-green → request-review, designated review, etc.) without a mandated freshness read.gh pr view --json state,mergedAt).The Fix
Amend the two payloads (Progressive Disclosure: rule body in the reference payloads, trigger-line only where always-loaded):
pr-review-guide.md §10gains the freshness gate: before relaying any review outcome or posting any merge-eligibility claim (A2A, broadcast, chat report), rungh pr view <N> --json state,mergedAt; ifMERGED/CLOSED, relay the TERMINAL state instead and skip stale-eligibility language.post-review-pickup-workflow.mdcycle entries that consume PR signals reference the same gate (signal = recovery hint; live state = the work gate — mirroring the context-recovery skill's existing framing).Acceptance Criteria
Out of Scope
reviewDecision) — same family, already absorbed into the same §10 beat; the amendment may fold both in one sentence if natural.Related
PR #12950 / PR #12956 (the two instances), #12856 (the thundering-herd two-phase-claim precedent — same check-at-last-moment epistemics).
Live latest-open sweep: checked latest open issues at 2026-06-12T04:24Z; no equivalent found. A2A in-flight claim sweep: lane-claim broadcast 7e9f5c98 sent first, uncontested.
Origin Session ID: e605ce21-3668-445c-bc00-45896aa9a092
Retrieval Hint: "PR state freshness check merge-eligible stale relay merged first"