Log Error
Log agentic coding failures to identify USER mistakes in prompting, context management, or harness configuration.
Core principle: The model is constant. The user's input is the variable. Focus on the variable.
Error Categories
| Category | Symptoms |
|---|---|
| Prompt Error | Ambiguous, missing constraints, too verbose, implicit expectations |
| Context Error | Context rot, stale/missing context, didn't /clear |
| Harness Error | Wrong agent type, subagent context loss, no guardrails |
| Meta Error | Rushed implementation, assumed competence, skipped planning |
Interview Process
Ask 5-8 specific questions about what happened:
- •"Your prompt was X words. What were the 3 most important requirements?"
- •"Did you specify what NOT to do?"
- •"When did you last /clear? How full was context?"
- •"What constraints were in your head but not in the prompt?"
- •"Was this reference material or explicit requirements?"
Be critical - user is logging to learn, not to feel good.
Required Captures
- •Triggering prompt - exact, verbatim
- •Primary error category - pick ONE
- •The gap - expected vs actual outcome
- •Prevention - specific action items for next time
- •Pattern check - seen this before? (habit to break)
Logging
- •Directory:
~/agents/logs/errors/ - •Format:
error-YYYY-MM-DD-NNN.md - •Template: See
template.mdin this skill directory
After logging, user can double-escape to rewind conversation and continue working.
Common Mistakes
| Mistake | Fix |
|---|---|
| Generic questions | Ask about what ACTUALLY happened |
| Blaming model | Focus 80% on user error |
| Sanitizing logs | Be specific and honest |
| Vague prevention | Give actionable items |