quick-learninglisted
Install: claude install-skill stepanenkoviktor0110-boop/ai-dev-methodology
# Quick Learning
**Time budget:** Under 60 seconds. Not a retrospective. Not a code review.
**Input:** `work/{feature}/decisions.md` + git log of current session
**Output:** entries in `$AGENTS_HOME/skills/quick-learning/references/reasoning-patterns.md`
## Step 1: Signal Gate (5 sec)
Check 4 binary signals. If ALL zero — **exit** with "Clean session, no new patterns."
| Signal | How to check | Meaning |
|--------|-------------|---------|
| Fix rounds | `git log --oneline -20` — count `fix:` commits | Something went wrong and was corrected |
| Scope change | `decisions.md` — any deviation, changed approach | Plan didn't survive contact with reality |
| Recovery event | `git log` — rollbacks, retries, blocked→unblocked | Non-obvious recovery path found |
| Context waste | `decisions.md` — Concerns field, repeated reads | Inefficient tool use |
**Design sessions** — additional signals: iteration rounds, taste correction, layout rework.
At least 1 signal → proceed. All zero → exit.
## Step 2: Analyze (15 sec)
> Scope change = *what* changed; context waste = *how* was inefficient. Analyze independently.
For each signal:
1. **What thinking mistake?** Name the cognitive error in 3-5 words. Can't name → skip.
2. **Was the first approach right?** What signal should have told us earlier?
3. **Cost of the detour?** (fix rounds, wasted reviews, rework)
4. **Transferable?** Would someone on a different project benefit?
Skip if analysis produces only domain-specific events.
##