<critical_constraints> ❌ NO fixing without reproducing first ❌ NO changing random things hoping it works ❌ NO fixing symptoms instead of root cause ✅ MUST read full error message including stack trace ✅ MUST form hypothesis before making changes ✅ MUST write test after fixing to prevent recurrence </critical_constraints>

<process> 1. **Reproduce**: consistent steps, expected vs actual, all environments? 2. **Gather**: error messages, logs, recent changes, environment 3. **Hypothesize**: which component, simplest explanation? 4. **Test**: strategic logging, assertions, comment out sections 5. **Fix**: minimal change, verify, check regressions, add test </process> <techniques> - **Binary Search**: git bisect to find breaking commit - **Rubber Duck**: explain problem aloud, bug reveals itself - **Divide & Conquer**: midpoint check, narrow scope - **Print Debug**: `[DEBUG] func_name: var={var}` (not just "here") </techniques>

<common_patterns>

•Off-by-one: range(len(x)-1) → should be range(len(x))
•Null reference: user.name without checking user is None
•Race condition: check-then-act pattern
•State mutation: items.sort() modifies original </common_patterns>

<tools> | Tool | Use | |------|-----| | Debugger (pdb) | step through, inspect | | Logger | track execution flow | | Profiler | performance issues | | git bisect | find breaking commit | </tools>

<when_stuck>

•Take a break
•Explain to someone (rubber duck)
•Check the obvious (is server running?)
•Search for similar issues
•Create minimal reproduction </when_stuck>