| name | co-loop |
| description | Autonomous background refinement loop for claude-codex-orchestration. Runs /co-eval + /co-review, if candidate ≥ 6 runs /co-promote and commits, then schedules next cycle via ScheduleWakeup (270s, stays within cache TTL). Never self-stops — continues until user manually interrupts. Darwin ratchet: next session scoring the promoted dimension must improve or rule is reverted. Strategy escalation for persistent weak points (5/8/12/15+ sessions). |
/co-loop — Autonomous Background Refinement
Full protocol: ~/.claude/skills/claude-codex-orchestration/references/self-correction.md §Layer 3 "Autonomous cron loop".
Quick run
- Run
/co-eval on last session
- Run
/co-review to find promote candidates
- If candidate ≥ 6 →
/co-promote → git commit (triggers Layer 0 score)
- CC runtime:
ScheduleWakeup(270s, "continue /co-loop", "autonomous refinement") → repeat until user interrupts
- Codex runtime (no ScheduleWakeup): Single execution only. Log:
"[co-loop] Single execution complete. Re-run manually for next cycle." No auto-scheduling — user must re-invoke.
Strategy escalation (same weak point ≥ N sessions)
| Stuck sessions | Escalate to |
|---|
| 5 | Micro-fix: rewrite only failing bullet/sentence |
| 8 | Section rewrite: restructure whole failing section |
| 12 | Radical restructure: reconsider section's purpose |
| 15+ | Flag to user: possible structural design flaw |
Darwin ratchet
After promoting: next session scoring same dimension improves → keep. Doesn't improve → rule was wrong, revert and reclassify root cause.
