Checks
Every push runs the checks in the ci block of wrktr.json, read at the commit being checked. Checks run in
Cloudflare Containers, one container per check, side by side. The review agent reads their results, and a branch
lands only once its required checks pass.
{ "ci": { "setup": "pnpm install --frozen-lockfile", "cacheKeyFiles": ["pnpm-lock.yaml"], "instance": "standard-2", "checks": [ { "name": "typecheck", "run": "pnpm typecheck" }, { "name": "test", "run": "pnpm test", "timeoutSeconds": 600 }, { "name": "format", "run": "pnpm format:check", "required": false } ] }}| Field | What it is |
|---|---|
setup |
Runs once per checkout before the checks (installs). Its result is snapshotted and reused. |
cacheKeyFiles |
Files whose content decides whether a snapshot of the setup can be reused, such as the lockfile. |
instance |
The container size: standard-1, standard-2 (the default), standard-3 or standard-4. |
checks[].name |
A short name: letters, digits, space, ., _ and -. Unique. |
checks[].run |
The command. It fails the check by exiting non-zero. |
checks[].timeoutSeconds |
Its limit, 1 to 1800 (default 600). |
checks[].required |
false makes a failing check not hold a land. Checks are required by default. |
Up to 20 checks. An invalid ci block fails the run with what’s wrong (and blocks landing as “CI config”).
Results are reused
Section titled “Results are reused”A check’s result is stored by the source tree it ran on, plus its command, the setup, the instance and the timeout. Push a tree that’s already been checked with the same settings, and the result is reused (shown as cached) instead of run again. Change a command or the setup, and the checks it affects run again.
To run a branch’s checks again anyway (a flaky check, or one that depends on something outside the repo):
wrktr ci rerunWhen the infrastructure fails under a check (the container didn’t start, the checkout failed), wrktr retries it and says so; that’s an error, not a failure of your code.
Reading results
Section titled “Reading results”wrktr ci show # the branch head's run: each check's state and time, and the end of failed logswrktr ci show --wait # wait for it, then exit 1 unless it passedwrktr ci log test # one check's full logwrktr ci runs # the project's runs, newest firstFailing checks also appear in the worktree’s .wrktr/FEEDBACK.md, next to the review threads, so the coding agent
sees them without asking.