Skip to main content
Launching the agent is only the first half of the job. The review loop is where orchestration either holds together or falls apart.

Diffs

Inspect the branch and changed files without switching to a different tool just to understand what happened.

CI State

Track whether the branch is actually heading toward merge or just produced a local patch that looks plausible.

Feedback

Send targeted guidance back into the session instead of relaunching the whole workflow from scratch.

Review as a first-class state

Conductor tracks review, CI, and merge status so handoff from agent to human is visible in one place.

Feedback loop

Use the dashboard or CLI to push review feedback back into a session.

Practical workflow

1

Let the session finish in its own worktree

The runtime boundary matters because reviewers need to know exactly which branch and diff belong to the current task.
2

Inspect the diff, branch state, and CI

Look at the actual changes and the merge path together instead of treating review as a blind yes-or-no gate.
3

Send targeted feedback

Keep the feedback specific so the next retry addresses the missing edge case, failing test, or code-shape concern directly.
4

Retry only the task that needs work

Recover one task without destabilizing the rest of the board or losing the audit trail for the current session.
Last modified on March 6, 2026