kookr-codex-pr-statelisted

## When to Use This skill is invoked after the critic phase completes for a batch. It persists the observations and updates tracking state. ## Non-Negotiable Rules | # | Rule | Violation Example | Correct Pattern | |---|------|-------------------|-----------------| | 1 | Append to learnings-raw.md, never overwrite | Writing fresh file that drops prior learnings | Append batch section to end of file | | 2 | Update state.json atomically | Writing partial state on error | Read → modify → write complete JSON | | 3 | Add ALL batch PR IDs to processed_prs | Only adding merged PRs | Add every analyzed PR, merged or not | | 4 | Add skipped PR IDs to skipped_prs | Forgetting skipped PRs, re-evaluating them | Track skips separately from processed | | 5 | Never commit state files | `git add state.json` | All state files live in ~/.claude/codex-pr-lessons/ | | 6 | Timestamp each batch entry | Undated learnings | Include ISO date in batch header | ## State File Schema (state.json) ```json { "repo": "openai/codex", "version": 1, "processed_prs": [101, 102, 103], "skipped_prs": [50, 51], "total_processed": 3, "total_skipped": 2, "distillation_count": 0, "last_batch_at": "2026-03-26T10:00:00Z", "last_distillation_at": null, "cursor": 2, "notes": "State file for codex-pr-lessons playbook." } ``` > **`repo` is required.** Kookr's dashboard reads this field to match PR lessons state to projects. Without it, the project's PR lessons count will not display. ### Field