It runs the code before it judges it
Every substantive review gets an ephemeral sandbox with the pull request checked out. inline installs, builds, runs your suite, and writes throwaway tests for changed code nothing covers. You get the commands, the output, and the artifacts.
The terminal,
not a summary of it
Every command inline ran is on the left, its full output on the right, ANSI color and all. If a step could not run, the run is marked partial and says why rather than quietly reporting green.
- Install dependenciesnpm ci
- Type checknpx tsc --noEmit
- Existing suitenpx vitest run
- Authored testsnpx vitest run charges.spec.ts
- Browser flownpx playwright test checkout
A run that
does not give up
Install failures fall back through the package managers a repo might use. Authored tests are re-run and repaired until they hold or the attempt is reported as failed. Browser flows record video and screenshots and keep them as artifacts.
- Idempotency key is written after the chargethread resolved
- Replay cache key omits the tenant idcode changed by 8f4c1ab
- Retry budget is never reset between attemptsawaiting a reply
Included
Put it on one repository
Connect GitHub, enable a repo, and the next pull request opened gets a review.