Autopilot
Understand how 143 prioritizes issues and decides when agent work can run automatically.
Autopilot is the part of 143 that decides which work can safely move without a manual session kickoff. It should make priority, eligibility, blocked states, and human review gates visible before an agent starts.
How Autopilot decides
Autopilot evaluates gates in order:
- The organization autonomy setting allows automatic work.
- The issue complexity fits the configured autonomy level.
- The aggressiveness limit allows this complexity tier.
- Organization concurrency is below the active run cap.
If any gate fails, the issue stays available for manual kick-off.
What you see
The queue should make state explicit:
- Ready for manual kick-off.
- Queued or running.
- Needs review.
- Opened a PR.
- Failed with actionable next steps.
- Completed or shipped.
Good operating model
Start with conservative automation and raise autonomy after the team trusts the repo config, validation commands, preview setup, and review expectations.
For agents
Automation eligibility is not the same as priority. An issue can be high priority but blocked from autorun by complexity, autonomy, or concurrency gates.