Tidy Tests

Work can produce tests that help diagnose a problem or reach Green but do not belong in the final result. This is unavoidable in agentic coding, where verification is a necessary part of the workflow. Do not try to prevent these tests from being created; tidy them up after the work.

During tidying, remove temporary tests and remove or merge duplicate tests. Do not expand the scope into general test quality improvements or a search for missing tests.

Working Mode

• If the user asks for an investigation, do not modify files. Report the problem locations, the basis for each decision, and the recommended actions.
• If the user asks for a modification, make the changes without seeking separate permission, then report the results.
• If the skill is invoked without further instruction, follow the current session's mode. If the agent was editing files, modify them. If the agent was discussing the work with the user, investigate. If the context does not indicate either mode, investigate.

Procedure

1. Set the scope: Honor the scope specified by the user. If none is specified, cover only tests added or changed during the current work and tests directly related to the changed behavior.
2. Establish the basis: Check the user's instructions, specifications, agreed public interfaces, why each test was created during the current work, and which contract it currently protects. Do not treat the implementation or whether the tests pass as the sole source of truth.
3. Remove or merge tests: Remove test cases or assertions whose temporary purpose has ended, and remove or merge truly redundant test cases that another test fully replaces. Ask the user when there is not enough evidence to determine which current contract a test protects or how its role relates to another test.
4. Clean up residue: Remove imports, fixtures, helpers, mocks, snapshots, and test-only files that became unused after tests were removed or merged. Restore any .only, .skip, retry, or timeout added temporarily during the current work to its original state.
5. Finish: Run the affected tests with project-defined commands. Fix defects introduced by the cleanup and run the tests again. Report why each change was made and the final result.