Hello! I’ve been using the new GitHub Cloud Agent and it’s a fantastic addition to the workflow. However, I’ve noticed that every request is prefixed with a hardcoded instruction:
“Using the context from Github, focus on this specific message. Make sure to be proactive and take action if needed”
The “take action” instruction is quite broad. I’ve found the agent occasionally performs unintended actions that require me to manually add guardrails to every single prompt I write. This doesn’t scale well and creates unnecessary friction.
Are there plans to allow users to customize or override this default github agent prompt?
thanks!
Steps to Reproduce
comment @cursor [prompt] in Github PRs
prompt: provide analysis and recommendations for these code changes
Expected Behavior
i thought it would just add a PR comment, but it also updated the PR description and added a unnecessary commit. problematic when i’m using it on a teammates PR.
Hey, thanks for the report. I can see the screenshot with the prompt, and yeah, “take action if needed” is way too broad when you only want analysis.
Right now you can’t directly override the hardcoded prompt for the GitHub and Linear integrations. But there are two options that can help:
AGENTS.md
Cloud Agents read AGENTS.md from the repo. It won’t replace the system prompt, but it adds extra instructions the agent will follow. You can put something like “only comment, never modify PR descriptions or push commits unless explicitly asked”. Docs: Rules | Cursor Docs
Automations with GitHub triggers
These give you full control over the prompt, MCP tools, and scope. The downside is triggers run on events like PR created or comment added, not via @cursor mentions. If your workflow can adapt, this gives you the customization you want. Docs: Automations | Cursor Docs
