diagnosing-bugslisted

# Diagnosing Bugs - Tight Feedback Loop Diagnose hard bugs by first building a command that can catch the user's exact symptom. Do not jump to a theory before a red-capable loop exists. **Default command:** `$diagnosing-bugs <symptom>` **Primary output:** a fix verified by the original loop, or a clear blocked diagnostic note listing what evidence is missing. --- ## Commands | Command | Usage | Purpose | |---------|-------|---------| | `loop` | `$diagnosing-bugs loop <symptom>` | Build or tighten the red-capable feedback loop. | | `perf` | `$diagnosing-bugs perf <symptom>` | Establish timing/profiling evidence before fixing a performance regression. | | `fix` | `$diagnosing-bugs fix <symptom>` | Reproduce, minimize, fix, and verify the bug. | | `postmortem` | `$diagnosing-bugs postmortem` | Capture cause, regression coverage, and prevention notes after the fix. | Default to `fix` when the user reports a concrete bug and has not requested a diagnosis-only pass. --- ## Core Rule No red-capable command, no confident diagnosis. A valid loop is: - **Specific:** it asserts the user's symptom, not just "does not crash". - **Runnable:** it can be run unattended by the agent. - **Deterministic enough:** it gives the same verdict, or raises a flaky reproduction rate high enough to debug. - **Fast enough:** seconds if possible; narrow minutes only when the system genuinely requires it. When no loop can be built, stop and report what was tried. Ask for the missing artifac