Prune

Verification accretes. Each delivery adds tests, checks, and the seams they need, and nothing retires them. Prune one scope of that surface so what remains is cheaper to run and maintain and still proves every contract. Optimize for confidence, not deletion count.

Run to the end on your own, and write the report and response in the user's active language.

Value bar

Read the test value reference, references/test-value.md, in the installed strategic-programming skill; when it is not installed, stop and say so. Its keeper, junk patterns, and retention bar judge every test in scope.

A check is a lint or type rule, an architectural test, or a script or step in the verification or CI command. It earns its cost when the mechanism it guards still exists, it goes red on a planted violation, and no other check catches the same violation. A gate that runs the same suite twice, or a step that cannot fail, is a check without value.

Baseline

1. Resolve the scope: the module, package, subsystem, or kind of check the user names. Without one, take the area with the strongest signal, such as the slowest or intermittent suites, test code growing faster than the production code it covers, or checks added since the area was last pruned, and state the choice.
2. Read root and scoped agent instructions, coding standards, gate configuration, and CI routing.
3. At a pinned revision, record the scope's test and support line counts, the result and duration of each suite, and each check. List baseline failures separately: they are possible product bugs, not stale tests.

Done when every test file and check in scope has a recorded baseline.