capture-learning

Reflect on the current session and capture learnings. Context: $ARGUMENTS

Process

1. Reflect

• What was the main problem being solved?
• What non-obvious insights emerged?
• What patterns are reusable in future work?
• What failed and why?

2. Structure the Learning

Format each learning as:

### [Date]: [Title]
- **Issue**: [The problem encountered]
- **Solution**: [What worked]
- **Pattern**: [The reusable principle extracted]
- **Files**: [Key file references, if applicable]
- **Freshness**: HOT (actively relevant) | WARM (occasionally relevant) | COLD (reference only)

3. Store

Write the learning to the appropriate location:

• Agent-specific patterns: Update the relevant agent's project memory
• Project-wide rules: Add to .Codex/rules/ as a new file or append to an existing one
• Debugging insights: Add to auto-memory MEMORY.md

4. Calibrate

Check existing learnings for staleness:

• Are any HOT items now WARM or COLD?
• Are any previous learnings contradicted by what we learned today?
• Remove or update anything that's no longer accurate.