/learn-eval - Extract, Evaluate, then Save

Extends /learn with a quality gate and knowledge-placement awareness before writing any instinct.

What to Extract

Look for:

1. Error Resolution Patterns -- root cause + fix + reusability
2. Debugging Techniques -- non-obvious steps, tool combinations
3. Workarounds -- library quirks, API limitations, version-specific fixes
4. Codebase Patterns -- conventions, architecture decisions, integration patterns

Process

1. Review the session for extractable patterns

2. Identify the most valuable/reusable insight

3. Draft the instinct JSON:

{
  "id": "pattern-name",
  "trigger": "when [condition]",
  "action": "What to do -- with code examples if applicable",
  "confidence": 0.3,
  "domain": "code-style|testing|git|debugging|workflow|security|tooling",
  "evidence": [
    "Observed: [what happened in this session]"
  ],
  "created_at": "2026-03-22T10:00:00Z",
  "updated_at": "2026-03-22T10:00:00Z"
}

4. Quality gate -- Checklist + Holistic verdict

   4a. Required checklist (verify by actually reading files)

   Execute all of the following before evaluating the draft:

   • Read existing instincts in ~/.claude/superpowers-ecc/instincts/ to check for content overlap
   • Check CLAUDE.md (both project and global) for overlap
   • Consider whether updating an existing instinct's evidence would suffice
   • Confirm this is a reusable pattern, not a one-off fix

   4b. Holistic verdict

   Synthesize the checklist results and draft quality, then choose one of the following:

   Verdict	Meaning	Next Action
   Save	Unique, specific, well-scoped	Proceed to Step 5
   Improve then Save	Valuable but needs refinement	List improvements, revise, re-evaluate (once)
   Absorb into [X]	Should be merged into an existing instinct	Show target instinct and additions, proceed to Step 5
   Drop	Trivial, redundant, or too abstract	Explain reasoning and stop

   Guideline dimensions (informing the verdict, not scored):

   • Specificity & Actionability: Contains concrete code examples or commands
   • Scope Fit: Trigger, action, and domain are aligned and focused on a single pattern
   • Uniqueness: Provides value not covered by existing instincts
   • Reusability: Realistic trigger scenarios exist in future sessions

5. Verdict-specific confirmation flow

   • Improve then Save: Present improvements + revised draft + updated checklist/verdict; if revised verdict is Save, save after confirmation, otherwise follow new verdict
   • Save: Present instinct file path + checklist results + 1-line verdict rationale + full draft, save after user confirmation
   • Absorb into [X]: Present target instinct path + changes (new evidence, adjusted confidence) + checklist results + verdict rationale, update after user confirmation
   • Drop: Show checklist results + reasoning only (no confirmation needed)

6. Save / Absorb to ~/.claude/superpowers-ecc/instincts/ via state/store.js

   • store.write('instincts/<id>.json', instinct)
   • Update instincts/index.json with the new/updated entry

Output Format for Step 4

### Checklist
- [x] Existing instincts: no overlap (or: overlap found with [X])
- [x] CLAUDE.md: no overlap (or: overlap found)
- [x] Existing instinct update: new instinct appropriate (or: should absorb into [X])
- [x] Reusability: confirmed (or: one-off, Drop)

### Verdict: Save / Improve then Save / Absorb into [X] / Drop

**Rationale:** (1-2 sentences explaining the verdict)

Notes

• Don't extract trivial fixes (typos, simple syntax errors)
• Don't extract one-time issues (specific API outages, etc.)
• Focus on patterns that will save time in future sessions
• Keep instincts focused -- one trigger, one action per instinct
• When the verdict is Absorb, update the existing instinct's evidence and confidence rather than creating a new file