A skeptical QA agent with memory: baseline → delta runs (NEW/REGRESSED/RESOLVED), flaky quarantine with expiry, and evidence-cited release verdicts you can defend.
Skills 5 in the Claude listing
Verdict flaky triage Classify an intermittent test failure with evidence — REAL_DEFECT, STALE_EXPECTATION, BRITTLE_TEST, ENVIRONMENT or FLAKY — and decide quarantine with a mandatory expiry. Verdict release risk Run a measured, skeptical QA pass on a repository and produce a release verdict you can defend — evidence-cited findings, a stated not-tested list, baseline → delta memory. Verdict root cause Trace a failure to its root cause with proof, not a story — a four-link chain (symptom, mechanism, origin, class), a counterfactual in an isolated scratch copy, trigger separated from cause, and a dia Verdict spec review Judge a specification, issue or PRD for testability before any code exists — inventory the requirements, find the untestable, contradictory, unbounded and silent ones with a verbatim quote each, and r Verdict verify fix Prove that a fix fixed the defect — re-run the guarding test at the previous commit and at HEAD, re-inject the defect in an isolated scratch copy, and let the harness record fix_verified from measurem