Devin AI SDK harness diagnostics

With debug: { enabled: true } on the HarnessAgent, the adapter reports what it is doing through the harness’s standard diagnostics channel (onLog), so a UI can show setup progress before Devin’s first token and explain failures. Each event carries sessionId, a subsystem and a short message:

`subsystem`	Level	When
`devin.bridge.attach`	info	Attaching to a bridge parked by a previous turn
`devin.bridge.attach`	warn	The attach failed (`error` attached); the bridge is respawned
`devin.bridge.spawn`	info	Installing and starting a fresh bridge in the sandbox
`devin.bridge.ready`	info	The bridge WebSocket is connected
`devin.cli`	info	Starting the Devin CLI inside the sandbox (once per bridge)
`devin.session`	info	Creating a new Devin session, or loading the resumed one
`devin.session.ready`	info	The Devin session is open
`devin.prompt`	info	The prompt was sent; the next event is Devin’s output

const agent = new HarnessAgent({
  harness: createDevin(),
  sandbox,
  debug: { enabled: true, level: "info" },
  onLog: (event) => {
    if (event.subsystem.startsWith("devin.")) console.log(event.sessionId, event.message);
  },
});

Sandbox creation and the bootstrap install happen before the adapter runs and are not reported here; the first devin.bridge.* event marks the point where the sandbox is ready. The harness-chat example builds a status line on these events.

Get started

Sessions

Integrations

Examples

SDK reference

API client

Resources

Devin AI SDK harness diagnostics