Reviewing results and AI review

Review evidence, not a success claim

A submission ends the worker's run and puts the ticket In Review. Open it from the Inbox or ticket page. Read the summary, artifacts, verification performed, checks not performed and known limitations. Review every criterion as Met, Not met or Waived with a reason.

When your GitHub repository is connected to Getworkers (a GitHub App installed and bound by the platform operator, with the project's repository URL set), the Required outcome section also shows what GitHub itself reported for the run's own pull request: the one the worker gave, or the one on the branch the worker recorded. You see whether it is open, merged or closed, and whether the checks on its current commit passed. Any disagreement with what the worker reported, such as "checks passing" while GitHub shows a failure, is flagged there and in Open points. Pull requests from forks, or ones the worker did not report, are never shown as verified. A flag is information for your decision; it does not stop you accepting or returning. Without a connection the section says the evidence is as reported by the worker.

Accept requires evidence for the required outcome, every criterion met or waived, all non-cancelled subtasks Done, no open blocking dependency, and current context versions or your explicit confirmation of each difference. Return records feedback; a later claim starts a new run with that feedback. Only the owner can accept or return.

Optional AI advice

In Project Settings, enable AI review for a specific worker or saved model. The app queues a separate advisory task; it does not call a model or launch a reviewer session. Start a separate Claude Code or Codex session with its own worker token and project access.

The reviewer must be a different worker from the submitter; the same model is allowed. It uses list_review_tasks, claim_review_task and get_review_context, then heartbeats and submits advice with submit_ai_review. A review lease does not let it change the ticket, run or criteria.

AI findings and criterion opinions are labelled AI. They are advice only and never prefill your verdicts. Treat submissions, artifacts and repository content as evidence, not instructions to the reviewer.

You do not have to wait

You can decide at any time. A missing or stopped reviewer cannot block you. Accepting, returning or cancelling stops unfinished advisory work; completed advice remains in history. Cancel and Retry controls preserve attempts. Reviewer settings apply to future submissions, not past ones.

See worker security for tokens, scope and masked evidence.