Summary
Run the organization review from runbooks/org-review.md for three real weekly cycles, compare signal and noise, tune the prompt only from repeated evidence, and then make an explicit maintainer decision: keep the review manual, schedule it, or retire it.
Linear mirror: ZSH-8
Acceptance criteria
- Run three reviews against live GitHub and Linear state, each covering a distinct weekly window.
- Keep generated labels, comments, closures, merges, and follow-up issues draft-only unless a maintainer authorizes them separately.
- Record actionable findings, false positives, duplicate/missing tracker records, evidence gaps, time cost, and maintainer decisions for every cycle.
- Change the prompt/output only after comparative evidence shows a repeated precision, routing, or safety problem.
- After cycle 3, record the manual/scheduled/retired decision in this issue and ZSH-8 before closure.
Cycle 1 — 2026-07-11 through 2026-07-18
Completed on 2026-07-18 against live organization issue/PR state and workflow runs for the 11 repositories in the current workspace inventory. No labels, comments, closures, merges, or follow-up issues were applied.
Actionable draft findings
- #452 / ZSH-22 remains the dependency-automation blocker; Zi PR #352 must remain draft until hosted Renovate processing is proved.
- zunit scheduled run 29391939822 failed all six jobs before tests because a pinned
docker/setup-buildx-action SHA could not be resolved.
- Zi's deprecated lock workflow failed throughout the window; the draft recommendation is removal under organization workflow policy, not repair.
- Wiki PR #793 has Trunk tool-execution failures despite reporting no lint findings and needs diagnosis or a clean rerun.
- Six old human-authored, non-draft PRs need a bounded maintainer disposition pass; routine bot dependency PRs should not be mixed into that queue.
- Private security-alert follow-through remains tracked in Linear ZSH-20; advisory details must stay private.
These are review findings only. Owning-repository issues or other external actions remain maintainer decisions.
Signal and noise
High signal:
- live strategic GitHub/Linear mirror checks;
- current failing checks on active human PRs;
- repeated scheduled failures collapsed to one cause;
- cross-repository patterns tied to a specific next action.
Noise or misleading behavior:
- connector issue search did not reliably distinguish PRs, so an accurate PR inventory needed a dedicated PR search;
- unfiltered age lists were dominated by bots and abandoned work;
- raw seven-day failure lists included failures superseded by green reruns.
Prompt hypotheses for cycles 2 and 3
- exclude bots and drafts from “waiting for review,” summarizing bots separately;
- collapse identical scheduled failures with an occurrence count and latest run;
- require a current-state refresh before calling a historical failure urgent;
- separate “needs review” from “needs disposition” for PRs older than 30 days.
Remaining pilot
- Cycle 2: next distinct weekly window, no earlier than 2026-07-25.
- Cycle 3: following weekly window, no earlier than 2026-08-01.
- After cycle 3: compare signal/noise and decide manual continuation, scheduling, or retirement.
- Keep this issue open and ZSH-8 In Progress until that maintainer decision is recorded.
Prior evidence
A preliminary dry-run summary was recorded on #410; it does not replace the three live weekly cycles above.
Summary
Run the organization review from
runbooks/org-review.mdfor three real weekly cycles, compare signal and noise, tune the prompt only from repeated evidence, and then make an explicit maintainer decision: keep the review manual, schedule it, or retire it.Linear mirror: ZSH-8
Acceptance criteria
Cycle 1 — 2026-07-11 through 2026-07-18
Completed on 2026-07-18 against live organization issue/PR state and workflow runs for the 11 repositories in the current workspace inventory. No labels, comments, closures, merges, or follow-up issues were applied.
Actionable draft findings
docker/setup-buildx-actionSHA could not be resolved.These are review findings only. Owning-repository issues or other external actions remain maintainer decisions.
Signal and noise
High signal:
Noise or misleading behavior:
Prompt hypotheses for cycles 2 and 3
Remaining pilot
Prior evidence
A preliminary dry-run summary was recorded on #410; it does not replace the three live weekly cycles above.