Test audit

Find existing tests that cost more than they protect, prove it with evidence, and remove or repair one coherent batch after the user approves. Optimize for confidence, not deletion count.

Two modes share one value bar:

• Audit (default): a focused sweep for a few high-confidence candidates.
• Campaign: prune every test one subsystem owns (one package, plugin, or core area) in one PR. Before starting one, read campaign.md.

Adapted from openclaw test-audit (MIT).

Read when

• Every run: code-quality.md, code-structure.md, and testing.md. The mandatory No tautological tests rule, junk patterns, and retention bar live there, not here.
• Throughout: keep judgment in the main context and hand large reads to a subagent (main-context.md). Discovery and ledgers are subagent work.
• Before asking the user anything: Asking the user. Findings follow Plain language.
• Running checks: tooling.md. Shipping: shipping.md.

Contract