← ClaudeAtlas

test-auditlisted

Review requested test suites or regression-protection claims for wrong expectations, missed defects and brittle checks. Skip ordinary implementation and test execution alone.
Muratovnik/assay · ★ 1 · Testing & QA · score 77
Install: claude install-skill Muratovnik/assay
# Test audit Determine what the scoped tests actually protect and where their verdict can mislead. Audit both false acceptance of wrong behavior and false rejection of supported behavior. A clean result is valid; test smells are leads, not findings. ## Frame the review Read the original request and owner contracts. Identify the actual repository or frozen artifact, relevant source/test/configuration state, consumer boundary and decision: changed tests, a named capability or a wider suite. Establish the expected behavior independently of the test author's verdict where possible. Tests and implementation agreeing does not establish that either is correct. A defect inventory or remediation report is also not a complete acceptance contract unless the original request explicitly made it exhaustive. Without explicit requirements, infer provisional criteria from supported APIs, verified examples and consumer behavior, and label uncertainty. Separate characterization and intentional compatibility from approved product rules. An unresolved consequential policy choice is not permission to invent intent. Name exclusions and sampling. Whole-suite conclusions require an inventory of the relevant tests and execution configuration; inspecting a few files or a green CI summary does not cover every test. A test audit is not automatically a product, security, release or architecture audit. ## Preserve the subject and authority Do not repair the subject's code, tests, snapshots, configur