Default output: return only the result, blockers, and required evidence. Omit preambles, process narration, repeated context, confidence scores, and follow-up offers. Use at most five bullets unless a required artifact or schema needs more.

Test Fix Loop

Smallest cooking-routines skill. Hand the agent a failing test (or a set of them) and a marker file; the cooking Stop hook will keep the agent looping through fix-and-rerun cycles until either the tests pass or the iteration counter hits 10. Marker file is the trust boundary; remove it to end the loop.

When to Use

• A unit/integration test is failing and the fix is mechanical (assertion update, missing import, simple refactor)
• You want the agent to iterate without supervision until green
• You explicitly do not want the agent to ship the fix anywhere — cross-repo-refactor handles branch+push; this skill only edits the working tree

When NOT to Use

• Tests are failing because of a real design bug — fix the design first, don't loop
• Tests touch multi-repo state — use cross-repo-refactor instead
• You don't have a clean working tree — the loop edits files; commit or stash first

How to Run