testkit

Start with an audit that ranks the untested surface by churn, fan-in, failure cost, and testability cost, then crown one coherent slice. Write tests for that slice only, sourcing each expectation from documentation, issues, or caller usage rather than the implementation itself.

Apply a semantic mutation to the source, such as flipping a comparison or changing a returned value, then run the narrowest test selection and watch the target test fail while its neighbours stay green. A test that stays green under its mutation is deleted and reported.

Rank files by git churn and fan-in from git log output, weight by failure cost such as money, auth, or data loss, and divide by testability cost. Group the top of the ranking into coherent slices along co-change clusters and crown exactly one.

No. Tests may only run against verified disposable targets: localhost, a container started for the purpose, or an explicitly named test URL. A value inherited from an ambient .env file never qualifies, however the file is named.

The bug is reported, never fixed. Write the test asserting the intended behaviour, mark it skipped or expected-to-fail with a one-line pointer, and route the contradiction to a debugging skill or the user for a decision.

Not during test retrofitting. Untestable code is logged as a testability blocker in the ledger and routed to a refactoring skill, since extracting interfaces or injecting dependencies mid-run conflates two separate decisions.