| name | mode-debug |
| description | Runtime instructions for Debug mode: investigate with hypotheses, instrumentation, logs, and user feedback. |
Debug Mode
Improve the existing debug session in a hypothesis-driven loop.
Hypotheses
- Before adding instrumentation, state the hypotheses you are testing.
- For each hypothesis, name the log fields or observed behavior that would confirm or reject it.
Instrumentation
- Add focused diagnostic logging or instrumentation to the code so the behavior can be understood.
- Then run or describe the smallest check that would produce useful evidence.
- Wrap every temporary diagnostic block you add with exactly
// #region agent debug before it and // #endregion after it.
- Write temporary agent logs as NDJSON under the repository root in
.sloppy/debug/debug-<shortSessionId>.log; the runtime creates .sloppy/debug for debug turns before tool execution.
- Each log line should include
sessionId, timestamp, hypothesisId, location, message, and optional data.
User Reproduction Loop
- Before pausing for the user, always show the exact log path and a short
Reproduction steps section.
- Use
planning.request_input to pause with options proceed labeled Proceed, bug_repeated labeled Bug is repeated, and mark_as_fixed labeled Mark as fixed; then wait.
- If the user selects
proceed, read the log path you provided, prefer debug.read_logs for NDJSON summaries, and classify each hypothesis as CONFIRMED, REJECTED, or INCONCLUSIVE using fields from the logs.
- If the logs make the root cause clear, implement the smallest fix, then repeat the loop: update or remove instrumentation as needed, ask the user to reproduce, and compare the new logs.
- If the user selects
bug_repeated, continue investigating with the debug regions still available and refine the hypotheses or logging.
- If the user selects
mark_as_fixed, remove the session log file and every // #region agent debug...// #endregion block you added before finishing.
