A model-invoked engineering skill that runs a structured debugging loop: reproduce → minimize → hypothesize → instrument → fix (see pattern-systematic-debugging). A feedback-loop answer to failure mode #3: don’t guess, shrink the problem and test hypotheses against evidence. It ends with a “Cleanup” phase.
Redact. Redaction is the first move on every command, output, and captured artifact the skill shows: write <REDACTED>, build loops against env vars so credentials stay in the environment, and quote only the signal-carrying lines. Phase 1 asks the user for a redacted captured artifact.
After the fix. mp-ask-matt routes you to mp-retro once the fix is in, “to ask what would have prevented the bug”, and to mp-improve-codebase-architecture only where the finding is a missing seam.
Its cross-framework counterpart is GSD’s gsd-debugger sub-agent, which tracks hypotheses and evidence in persistent session state — see equivalent_to.
See Also
- gsd-debugger — GSD’s equivalent debugging capability.
- mp-tdd — the test-first feedback loop.
- mp-retro — the post-fix follow-up.
- addy-debugging — Addy’s debugging skill; same evidence-driven, no-guessing loop.
- stage-implement — the canonical stage this implements.
Version history
v1.3.1 (2026-10-04)
- mp-ask-matt routes a post-fix mp-retro instead of the old architecture hand-off.
v1.3.0 (2026-10-04)
- Phase 6 post-mortem hand-off to mp-improve-codebase-architecture removed: that skill is user-invoked, so it was unreachable from an often-unattended bug-fixing flow. Phase 6 became “Cleanup” only.
v1.2.3 (2026-08-06)
- Added the Redact section; the Phase 1 completion criterion changed from “paste the invocation and its output” to show it redacted.
Initial ingest (2026-06-28)
- Documented with the reproduce → minimize → hypothesize → instrument → fix loop.