Agent troubleshooting
Fixes for common Agent issues with file access, rules, terminal commands, and more.
How can I improve Agent accuracy?
- Be specific in your prompts: include expected behavior and constraints
- Use Plan mode first for tasks that touch many files
- Start a new chat when you finish a feature or switch tasks
- Add rules in .cursor/rules/ for patterns you want Agent to follow every time
- Type @ followed by a file or folder name to give Agent targeted context
What if Agent doesn't pick up my files?
- Check your .cursorignore file in the project root. Files listed there are blocked from Agent, semantic search, and @ mentions.
- Check .gitignore. Patterns there can also prevent Agent from discovering files.
- Reindex your project: open the command palette and search for "Reindex."
- Attach files directly by typing @ followed by the filename in the chat input.
How do I undo Agent changes?
Hover over a previous message and click Restore Checkpoint in the bottom right to roll back all changes Agent made after that point.
Checkpoints are stored locally and are separate from git. Use git for permanent version control.
What if Agent sets CI=1 in terminal commands?
Cursor sets CI=1 as an environment variable when running terminal commands through Agent. This helps terminal tools produce cleaner output. If your project behaves differently when CI=1 is set (for example, skipping interactive prompts or using CI-specific paths), you can unset it in your commands:
Or add this to a rule so Agent always unsets it:
What if I see "Agent Execution Timed Out"?
This error means Cursor's extension host didn't finish starting within 60 seconds, so Agent features couldn't initialize. The same root cause shows up as "Timeout waiting for EverythingProvider" in network diagnostics.
Causes vary, so collect logs before changing anything:
- Press Cmd/Ctrl+Shift+P and run Developer: Export Logs...
- Select Main, Window, and Extension Host
- Also check Output → Extension Host (Cmd/Ctrl+Shift+P → "Output", then pick "Extension Host" from the dropdown). If it's empty, take a screenshot; an empty panel is a strong diagnostic signal.
- Share the exported zip and screenshot with support, along with your OS and Cursor version.
On managed or enterprise machines, endpoint security software (antivirus, EDR) is one possible cause. If your IT team confirms a security agent is active, share the Endpoint Security Configuration page so they can add the right process and path exclusions.
How do I report a bad Agent response?
- Click the ... menu at the bottom of the response and select Copy Request ID.
- Post to forum.cursor.com with the request ID, steps to reproduce, and your system info (found in Cursor > About Cursor on macOS or Help > About on Windows/Linux).