investigate

Trace the failure back through the call chain to the shared function where the bad behavior originates, then fix it there. This Skill enforces that workflow: investigate, analyze, hypothesize, and only then implement, so callers sharing the same defect are fixed too.

Root cause analysis identifies the underlying defect responsible for a failure rather than the location where it surfaced. For example, a crash in one script may stem from a shared parsing helper, and fixing only the crash site leaves other callers broken.

Yes. It verifies the reported behavior against the actual code and data before changing anything. If the output is correct given the inputs — such as a negative margin reflecting a genuine loss — it rejects the bug framing instead of inventing a fix.

Invoke it when you see errors, stack traces, 500 responses, unexpected behavior, or phrases like "it was working yesterday". It is designed for any situation where something stopped working and the cause is not yet known.

Multiple call sites often share the same helper function, so the defect lives in the helper, not the caller. Patching one caller leaves every other caller exposed to the same failure under the same inputs.