Find a Regression

Objective

Identify a minimal causal boundary with a trustworthy test oracle and leave the repository in a known state.

Use When

• Run manual or automated git bisect between verified good and bad revisions.
• Design a classifier that distinguishes good, bad, skip, and infrastructure failure.
• Handle builds, generated artifacts, dependencies, submodules, or environment setup across revisions.
• Validate the reported first bad commit and nearby boundary.

Do Not Use / Route Elsewhere

• Use investigate-history for text/provenance questions without an executable oracle.
• Use a dedicated SWE review workflow to assess a known diff for product correctness.
• Use diagnose-repository when current repository state itself is confusing.
• Do not patch the regression as part of the search unless separately requested.