Cloud Agent (GitHub, Slack, Linear,…)
When creating a self hosted pool, it should be possible to inject a system prompt / pre prompt to give specific rules to the agent running on this system. This would enable as an example to have an agent running on a production system with strict rules for harmless debugging always injected before any conversation
Hey, thanks for the feature request. Pool-level control of the system prompt is a solid idea, and we don’t have a dedicated mechanism for that right now.
For now, a workaround is that rules are pulled from the repo the worker checks out. If you put them in AGENTS.md or in .cursor/rules/ with alwaysApply: true, they’ll apply to every agent conversation in that environment. For your use case, like production box, harmless debugging only, read-only, this is the closest to what you described:
---
alwaysApply: true
---
This is a production environment. Read-only debugging only, no destructive operations.
More on rules: Rules | Cursor Docs
The difference is that today rules live in the repo, not in the pool or worker config. We’ll track pool-level injection as the feature request. No timeline yet, but we’ll keep the thread open for votes. Let me know if the workaround doesn’t cover your scenario.