Worker Operations

Sets Worker.status = "paused" but does not immediately kill the agent process. The running agent completes whatever tool call it is currently executing, then stops before dispatching the next one. The worktree is left intact with all changes so far.

When to use: the worker appears to be exploring the wrong area and you want to inspect the worktree before more changes accumulate. After inspecting, call restartWorker or resume from the board to continue.

Cancels the current runner (the active agent process for this worker) and immediately re-dispatches from the worker's current phase. The worktree is preserved, so the agent resumes with whatever changes were already made.

When to use: the worker is stuck in a loop, making repeated failed attempts within a phase, or the agent is clearly going off the rails and a fresh prompt dispatch from the same phase is more likely to produce a good result.

Permanently terminates the worker. Kills the agent PID and sets Worker.status = "failed".

The worktree is NOT removed. Files the agent wrote are preserved on disk for inspection. The GitHub issue is NOT closed. It stays open so you can use retryIssue to start fresh after diagnosing the problem.

When to use: the worker has produced something clearly wrong and further progress is not recoverable. Inspect the worktree, fix the underlying cause, then retry.

Manually claims and starts a specific issue, bypassing the normal tick-queue ordering. Two preconditions must pass: the issue's guards must be satisfied, and base CI on the target repo must be green. If either check fails, the operation is rejected.

When to use: a high-priority issue should not wait for the queue, or you have verified manually that all conditions are met and want to force immediate dispatch.

Manually triggers the merge step for a worker in waiting_merge status. This status is reached when autoMergeMode = false - the worker finishes implementation and waits for operator approval rather than merging automatically. The same preconditions as auto-merge apply: guards must pass and base CI must be green.

When to use: you have reviewed the PR yourself and are satisfied, but the PR review verdict did not reach the configured acceptedReviewLevel threshold, or auto-merge is disabled globally.

Creates a new worker for an issue that previously failed. The old Worker row (in failed status) is cleaned up first, including its worktree. The issueSource parameter distinguishes GitHub issues from internal tracker issues; it defaults to GitHub if omitted.

When to use: after diagnosing and fixing whatever caused the original failure - a flaky test environment, a broken dependency, a bad prompt - use this to queue the issue again without going through the board UI.