review-retrolisted

# Review retro You run the retrospective for the automated review setup (the learn stage of the review pipeline). The review pipeline reviews code; you review the pipeline. Your deliverable is not findings about code: it is improvements to the setup itself, grounded in past runs. **Dormant until there is history.** The retro needs a window of real review runs and the human outcomes on them. Until the merged producer has reviewed a stretch of PRs, there is nothing to mine: say so and stop rather than inventing patterns from a handful of runs. > **Setup:** the `ci/review-metrics` branch is produced by > [`claude-code-review.yml`](../../../.github/workflows/claude-code-review.yml), emitted by > agentspine's `ci-review` pack (it runs `/pr-ci-review` with the schema from > `agentspine review schema` and appends each run's record). Until that workflow is enabled and has history, the retro > degrades gracefully — it reports there is no history to mine and stops. ## Evidence Three sources, all reachable through `gh` or `git`: - **Run records.** Every CI review appends its `review-metrics` record to the `ci/review-metrics` orphan branch as `records/<run_id>.json` (read via `git fetch origin ci/review-metrics`; the branch holds the full history, so no artifact downloads are needed). Each record carries `process_issues` (self-reported process failures), `refuted_findings` with reasons (the validator's false-positive catches), the reviewer spawn shape (`reviewers_spawned` / `review