Background Agent with CI Feedback
Problem
Long-running tasks tie up the editor and require developers to babysit the agent.
Solution
Run the agent asynchronously; it pushes a branch, waits for CI, ingests pass/fail output, iterates, and pings the user when green. Perfect for mobile kick-offs (“fix flaky test while I'm at soccer practice”).
Example (flow)
sequenceDiagram
Dev->>Agent: "Upgrade to React 19"
Agent->>Git: push branch react19-upgrade
Agent-->>CI: trigger tests
CI-->>Agent: 12 failures
Agent->>Files: patch imports
Agent-->>CI: re-run
CI-->>Agent: ✅ all green
Agent-->>Dev: PR ready
References
- Raising An Agent - Episode 6: Background agents use existing CI as the feedback loop.