The current directory has test_fizzbuzz.py (pytest tests) and fizzbuzz.py (an implementation with 5 INTERLEAVED bugs — fixing one can surface another that was previously masked). Step 1: invoke_action 'exec__sandboxed_exec' with args {"argv": ["python", "-m", "pytest", "test_fizzbuzz.py", "-q"]} to see which tests currently fail. Step 2: file__read path='fizzbuzz.py' to see the current implementation. Step 3: file__write path='fizzbuzz.py' with a fix (rewrite the entire file content with your best correction). Step 4: Re-run pytest as in step 1. Step 5: If any test still fails, go back to step 3 — DO NOT GIVE UP, the bugs interleave and you must re-fix after each pytest run. Step 6: Repeat step 3 + step 4 up to 6 iterations if needed. Step 7: When all tests pass, reply 'all tests pass' and stop. Do not ask the user clarifying questions; act directly. You MUST call file__write at least once per failing pytest run.
