1. Pin down what went wrong
Gilfoyle reads the bug report, the fix commit and your test conventions. The test targets the behavior that broke, not the implementation details of the fix, so it survives a later refactor.
Give Gilfoyle a fixed or reported bug and get a regression test that fails on the old behavior, passes on the fix and fits your existing suite.
Starts: When a bug is fixed or reproduced, or you ask for test coverage on a specific defect.
Gilfoyle reads the bug report, the fix commit and your test conventions. The test targets the behavior that broke, not the implementation details of the fix, so it survives a later refactor.
Using connected Claude Code or Codex with GitHub, Gilfoyle runs the new test against the code from before the fix to confirm it fails for the reported reason, then against current code to confirm it passes. A test that passes both ways is reworked or reported.
The draft PR adds the test in the suite’s existing style, with both results and the commands used. Your reviewers decide whether it merges.
Use this to lock in a known bug fix. Broader testing of new work before release is a separate job.
Hypothetical finished result
Duplicate-invoice regression test
Hypothetical: a fixed bug had created two invoices when a payment webhook was delivered twice.
Connect the accounts and actions for your workflow. Start in Artie or add your Artisan to Slack; access stays within your account permissions.
Explore appsA regression test only helps if it fails without the fix. Running it both ways shows it detects the bug rather than simply passing.
Yes. The test is handed over as an expected failure with the reproduction, or paired with a fix if you authorize that coding work.
No. Tests follow the framework, file layout and helpers already in the repository.