Where does the bug appear (feature/product)?
Cloud Agent (GitHub, Slack, Web, Linear)
Describe the Bug
Cloud Agent with /goal did not stop after an explicit stop/cancel request. It continued coding for about 6 hours and consumed a large amount of tokens. A later stop command then caused the agent to reply three times to a single message. The session was only stopped after repeated attempts.
I started a Cloud Agent using /goal and instructed it to implement code from a spec document. After some time I told it to stop and cancel the work. The agent ignored the stop request and kept writing/editing code in the cloud for approximately 6 hours, continuing to burn usage/tokens the whole time.
When I sent another stop command, the UI/agent produced three duplicate responses to that single instruction. I was eventually able to halt the session, but stop was not reliable.
This is a billing and control issue: a user stop should immediately terminate the run and stop token consumption.
Steps to Reproduce
Start a Cloud Agent from Cursor (select Cloud under the agent input, or from cursor.com/agents).
Use /goal and instruct the agent to implement features based on a spec document.
Let the agent run and make code changes.
Send an explicit follow-up telling it to stop / cancel / do no more work.
Observe that the agent continues coding in the cloud for hours and keeps consuming tokens.
Send another stop command.
Observe that the agent answers the same stop message three times instead of stopping cleanly.
Expected Behavior
An explicit stop/cancel message (or the Stop button) should immediately halt the Cloud Agent.
No further tool calls, file edits, or token usage should occur after stop.
/goal should not keep the agent alive against a user cancel.
One user message should produce one response, not three duplicate replies.
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?
Cloud Agent with /goal
Model: grok-4.6-xhigh-fast
I used this model for a long-running /goal session that implemented code from a spec document.
For AI issues: add Request ID with privacy disabled
Cloud / Background Agent ID: bc-01a0419a-128d-7383-a6bb-58c52e653f2c
This was a Cloud Agent /goal run, not a short local chat.
Please check why the agent continued running for ~6 hours after I explicitly asked it to stop, and why a later stop message produced 3 duplicate replies.
Additional Information
I started a Cloud Agent with /goal and told it to implement code based on a spec document.
After some progress I explicitly told it to stop and cancel the remaining work.
It did not stop. It kept coding in the cloud for about 6 hours and consumed a large amount of tokens.
When I sent another stop command, the agent replied three times to that single message.
I was only able to halt it after repeated attempts.
This makes /goal unsafe to use: a user stop should immediately end the run and stop billing.
I can provide screenshots of the stop messages, the continued activity, and the usage spike if needed.
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor