process

# Process Skill Five modes. Parse the request to pick one. Default to **retro** when ambiguous. | Signal | Mode | |--------|------| | Negative results, what didn't work, routing stats, review ROI, retro | **Retro** | | Hand off, package state, session pickup, rehydrate | **Handoff** | | Pair program, step by step, walk me through, one change at a time | **Pair** | | Subagent per task, execute plan, fresh context | **Subagent** | | Wait for, retry, backoff, poll, health check, circuit breaker | **Wait** | --- ## Mode: Retro Read-only retrospective. Two stores: `docs/what-didnt-work.md` (negative-results registry) and `learning.db` (routing/review telemetry via `scripts/learning-db.py`). ### Subcommands | Argument | Action | |----------|--------| | (none), what-didnt-work | Read `docs/what-didnt-work.md`. Group by date. Show each Decision verdict up front. If missing, report no negative results recorded. | | routing, route health | Run `python3 ~/.claude/scripts/learning-db.py route-health`, then `route-stats --by agent` (or the user's dimension). Present outcome basis before rates. | | reviews, review ROI | Run `learning-db.py review-roi` and `review-fps`. Present cost, findings, and false positives per agent in one table. | For comparisons: `learning-db.py route-delta --from SHA_OR_DATE --to SHA_OR_DATE`. --- ## Mode: Handoff Two submodes sharing one state contract. HANDOFF packages work for the next session. PICKUP rehydrates from that package. ### Handoff (endin...