What tests does
The tests skill detects the stack, reuses an existing test runner or installs the native choice, adds one small working example, records the command in AGENTS.md, and runs the build and suite.
The opt-in switch
Once AGENTS.md declares a test command, logic-bearing implementation steps must include relevant tests and keep the suite green. UI and integration surfaces continue to use browser evidence and builds.
What it does not promise
The skill does not generate a complete test suite. It creates a trustworthy foundation and makes future logic testing an explicit workflow requirement.
Good unit-test targets
Parsers, validators, formatters, slug builders, server actions, and other pure logic with meaningful edge cases are good targets. Visual markup and browser flows are not.