Problem
A Devin worker spawned through an Agent Relay fleet node can start successfully at the process level but never execute its assigned task because Devin stops at an interactive first-run prompt:
Do you trust the authors of this directory?
1. Yes, I trust the authors
2. No, exit
Observed worker:
- Agent: cloud-cleanup-lead-devin-r3
- Node: sf-frame
- Working directory: /Users/khaliqgant/Projects/AgentWorkforce/chief
- Result: worker remains live in the fleet roster but is blocked before task execution
Agent Relay already handles equivalent first-run/bootstrap behavior for other CLI harnesses. Devin needs the same deterministic non-interactive setup.
Expected behavior
A fleet-spawned Devin worker should accept the explicitly supplied working directory through a safe broker-managed trust/bootstrap mechanism and proceed directly to its assigned task without human terminal interaction.
Acceptance criteria
- The Devin harness adapter handles the directory-trust prompt during fleet spawn.
- Trust is scoped to the explicit spawn working directory; it must not broadly trust unrelated paths.
- A fresh-home or first-run integration test proves the worker becomes ready and begins the task.
- Failure to establish trust returns a clear spawn failure instead of leaving a live-but-blocked worker.
- Existing Devin installations that have already trusted the directory continue to work.
Problem
A Devin worker spawned through an Agent Relay fleet node can start successfully at the process level but never execute its assigned task because Devin stops at an interactive first-run prompt:
Observed worker:
Agent Relay already handles equivalent first-run/bootstrap behavior for other CLI harnesses. Devin needs the same deterministic non-interactive setup.
Expected behavior
A fleet-spawned Devin worker should accept the explicitly supplied working directory through a safe broker-managed trust/bootstrap mechanism and proceed directly to its assigned task without human terminal interaction.
Acceptance criteria