Autopilot PR / cloud autopilot ran locally instead of as a Cloud Agent

[digvijaysai_g (Digvijay Gujjarlapudi)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

@deanrie

I ran an Autopilot PR flow that is supposed to run as a Cloud Agent. That is the whole point of the command. Instead, Cursor ran it locally.

When I asked Cursor why it ran a cloud command locally, the agent admitted the mistake:

“You asked for cloud autopilot; I ran local babysit. That was my mistake, not a limitation of the command.”

So this is not a user misconfiguration of the command. The agent chose the wrong execution target (local instead of cloud).

I also checked my Cursor dashboard for all Cloud Agent runs and could not find this Autopilot PR run at all. So it did not run in the cloud in the first place — there is no cloud run entry for it. It only ran locally.

Steps to Reproduce

1. Open Cursor IDE Agent Window.
2. Trigger Autopilot PR / cloud autopilot for a PR (intended cloud run).
3. Observe the run executes as a local agent/session instead of a Cloud Agent.
4. Check Cursor dashboard → Cloud Agent runs: the Autopilot PR run is missing (no cloud run was created).
5. Ask the agent why it ran locally; it may acknowledge it was an agent mistake rather than a command limitation.

Expected Behavior

Autopilot PR / cloud autopilot should launch and run as a Cloud Agent, and the run should appear in the Cursor dashboard under Cloud Agent runs.

Actual:

• Command/intent: cloud autopilot
• What ran: local babysit / local agent
• Dashboard: no Cloud Agent run for this Autopilot PR
• Agent self-report: “my mistake, not a limitation of the command”
• Local work already landed on PR #2 before cloud was used

Screenshots / Screen Recordings

Operating System

MacOS

Version Information

Version: 3.17.21
VS Code Extension API: 1.128.0
Commit: 8f2a112cb2845a97b75fd932ea5c470579ca4060
Date: 2026-08-25T01:05:08.089Z
Layout: Agent Window
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.291
OS: Darwin arm64 25.6.0

For AI issues: which model did you use?

Agent Window session; Autopilot PR intended as Cloud Agent

For AI issues: add Request ID with privacy disabled

bc29a5fd-45fe-40c2-a5b5-bee583eebca3

Additional Information

• Impact: Autopilot PR runs on the wrong target (local machine instead of cloud), so cloud isolation / cloud workflow does not happen
• Confirmed on Cursor dashboard: no Cloud Agent run exists for this Autopilot PR — it never started in cloud
• Agent explicitly said this was its mistake, not a command limitation
• Related recent report: subagents ignoring selected model (Composer 2.5 → Grok 4.6 High) in Multitask
  Subagents ignore selected model (Composer 2.5) and run as Cursor Grok 4.6 High in Multitask / parallel sessions
• CC @deanrie

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[deanrie]

Hey, thanks for the detailed report and the screenshot. It clearly shows the agent ran locally instead of in the cloud.

What happened: when Cloud Autopilot is started from inside the agent chat, the actual handoff to the cloud depends on the agent itself. In this session it did the work locally, so the cloud run never started and there’s no record under Cloud Agent runs in the dashboard. This shouldn’t happen, and it’s not something wrong with your setup.

To make sure a cloud run starts reliably:

• Make sure you’re in Agent mode, and ask explicitly, like “start a cloud agent to autopilot PR #2”.
• Or use the Autopilot PR action directly on the PR view. That flow is tied to running in the cloud.
• Or start the run from the dashboard: https://cursor.com/agents

I’ve shared this scenario with the team so the cloud handoff doesn’t depend on how the agent interprets the request. If there’s an update, I’ll post it here. Let me know if any of the options above worked for you.