Adversarial test sweep

Audit and harden an existing automated test suite as a finite, evidence-led campaign. Try to falsify the product's contracts, expose tests that can pass while behavior is broken, repair confirmed defects, preserve each failure as a deterministic regression, and remove only tests proven to add no distinct protection.

This skill is deliberately language-agnostic. Detect and use the repository's languages, frameworks, commands, conventions, and supported environments rather than importing a preferred stack.

Route the request

Use this skill when the user asks for one or more of these outcomes:

• a comprehensive or adversarial audit of an existing test suite
• a focused hardening sweep over a risky component, incident area, or subsystem
• systematic edge, invalid-state, concurrency, dependency-failure, or resource-boundary testing
• an evidence-based review of weak, flaky, redundant, or false-positive tests
• a regression campaign that turns discovered defects into durable protection

Use ordinary test guidance instead for routine tests around one feature or a single known fix. If available, load the installed maintainable-tests skill when writing or editing individual tests; it is a soft companion, while this skill owns the campaign, risk ledger, adversarial discovery loop, and completion gate.

Route production load or endurance work to performance testing, deployed fault campaigns to chaos engineering, and exploit-seeking work to security testing. This skill may simulate resource and dependency failures locally, but it does not authorize stress against shared systems, production mutation, destructive effects, or penetration testing.