This file defines a test-audit skill for the openclaw codebase with three operating modes: an authoring gate for new tests, focused audit sweeps, and full subsystem campaign pruning.

It specifies fourteen junk patterns that cause a test to fail the gate, alongside a retention bar listing which contracts justify keeping a test. The document prescribes a read-only discovery process, a candidate-evidence checklist required before any deletion, and a six-step validation sequence using Vitest and repository scripts.