Tests

Linters and type checks tell the agent whether the code is structurally valid. Tests tell it whether the code does the right thing. There are two types of tests, generally speaking:

• Unit tests for fast, local correctness checks
• Integration tests for realistic end-to-end behavior

If your agent keeps "fixing" one bug while creating another, that's a strong signal that your test coverage is too thin or your test loop is too slow. When the agent can run go test ./... or pytest after every change and get an immediate pass/fail, it self-corrects much more reliably.

ralph ❌ {
    hallucinate(“never”)✨
}