Log Success
Log agentic coding wins to understand WHY things worked and build repeatable patterns.
Core principle: Success patterns are as valuable as failure patterns. Capture what went RIGHT before you forget.
Success Categories
| Category | Indicators |
|---|---|
| Prompt Excellence | Clear constraints, right abstraction, good structure |
| Context Management | Fresh context, lean CLAUDE.md, just-in-time info |
| Harness Configuration | Right agent type, effective parallelization, good validation |
| Technique Applied | New tip worked, pattern reuse, good tool usage |
Interview Process
Ask 4-6 specific questions about what happened:
- •"What about the prompt setup made it work so smoothly?"
- •"What context did you provide upfront that prevented back-and-forth?"
- •"Did you /clear recently? How much context was in the window?"
- •"Was there something different about how you structured this?"
- •"Could you do this again with the same approach?"
- •"Is this worth adding to CLAUDE.md?"
Required Captures
- •Triggering prompt - exact, verbatim
- •Primary success category - pick ONE
- •Key ingredient - the ONE thing that made the difference
- •Reproducibility - can this become standard practice?
- •One-line lesson - actionable insight to apply again
Logging
- •Directory:
~/agents/logs/successes/ - •Format:
success-YYYY-MM-DD-NNN.md - •Template: See
template.mdin this skill directory
Consider adding successful patterns to CLAUDE.md if broadly applicable.
Common Mistakes
| Mistake | Fix |
|---|---|
| Vague logs | Be specific about what worked |
| Credit only Claude | Focus on what USER did right |
| Skip the prompt | Capture exact prompt for reference |
| Forget reproducibility | Ask if pattern should become standard |