Skip to main content

Writing Behavioral Tests

Behavioral fixtures validate that the expected output of an AI agent passes the domain-specific validator.

Fixture structure​

tests/behavioral/my-test/
├── prompt.md     ← The exact prompt you'd give an agent
├── expected.md   ← The golden output the agent should produce
└── skill.txt     ← Routes to the right validator (c4designer or adr-scribe)

Requirements​

  • prompt.md must contain an ## Acceptance Criteria section or CI fails
  • skill.txt must contain exactly c4designer or adr-scribe
  • expected.md must pass the corresponding validator

Negative test fixtures​

Prefix with negative- to assert that the validator rejects the output:

tests/behavioral/negative-my-bad-case/

The test runner expects these to fail validation. If they pass, the test fails — proving the linter actually works.

Running fixtures​

make test