Skip to content

Three modes

Cairn’s modes are decoupled so you can keep as much human review in the loop as you want.

design cases only — first review pass

Observes and grounds the page, then writes ATC + MTC cases using EP, BVA, decision tables, and state transitions. Generates no code — review the Markdown files first.

$ cairn design --url https://app.example.com/page --session myapp
automate code from approved ATC

Generates POM-style Playwright specs from the ATC cases of a run, executes them, classifies pass / fail / flaky, and self-repairs with keep-best.

$ cairn automate --run runs/<id> --validate --session myapp
explore full cycle in one run

Everything end to end: observe → ground → design → generate & validate → judge & learn. Finishes with a Pilot verdict.

$ cairn explore --url https://app.example.com/page --session myapp

Typical progression: start with design + manual review, switch to designpromoteautomate once you trust the cases, and graduate to explore for fully autonomous runs (e.g. in CI or on a schedule).