Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
Project-level instructions in .cursorrules (at repo root) are not applied in Agent (Chat). The model’s replies never follow the rule—e.g. they never append a required blockquote “nudge” at the end of responses when the rule’s trigger conditions are met. The same rule works when placed in .cursor/rules/*.mdc with alwaysApply: true, so the issue appears to be that .cursorrules is either not loaded for Agent (Chat) or is applied differently than .cursor/rules.
Steps to Reproduce
Steps to Reproduce
- Add a rule to
.cursorrulesat the project root that requires the model to append a specific line at the end of certain responses (e.g. “Append> **Nudge:** [suggestion]when code was written or when the user jumps into building without clarifying”). - Open Agent (Chat) and send a prompt that triggers the rule (e.g. “Add a ‘Last synced’ timestamp to the dashboard header” or “Add export to CSV to the time-audit dashboard”).
- Let the model complete its response.
- Export the conversation (Cursor → Export) and inspect the model’s reply.
Result: The model’s reply does not include the required nudge; it ends after the substantive content (summary, code review, etc.). Repeating in a new chat with the same prompt yields the same result—no nudge.
Workaround: Move the same rule into .cursor/rules/<name>.mdc with alwaysApply: true. After that, the model’s replies in Agent (Chat) include the nudge as specified.
Expected Behavior
When a rule in .cursorrules instructs the model to append a blockquote (e.g. > **Nudge:** …) at the end of responses when certain conditions are met, the model should do so. For example:
- After implementing a code change: response should end with e.g.
> **Nudge:** Good stopping point — /review to catch anything before we move on? - After completing a feature/review: response should end with e.g.
> **Nudge:** This is in a good state — /done to lock it in? - When the user gives a vague build request without clarifying: response could end with e.g.
> **Nudge:** Want to run /discovery first to lock down what we're building?
Actual behavior
In full exported conversations, every model reply ends with the substantive content only. There is no > **Nudge:** line in any of the model responses—so the missing nudge is not due to the UI hiding it; the model output never included it. The rule in .cursorrules does not appear to be applied (or its output is not present) in Agent (Chat).
Operating System
MacOS
Version Information
- Cursor IDE: 2.6.14
For AI issues: which model did you use?
Additional Information
- Same rule in
.cursor/rules/build-cycle-nudges.mdcwithalwaysApply: truefixes the issue; nudges then appear in Agent (Chat). - Suggests
.cursorrulesat project root is either not used for Agent (Chat) or is applied with lower priority/different behavior than.cursor/rules. Documenting this in the Rules docs (or aligning behavior) would help users.
Does this stop you from using Cursor
No - Cursor works, but with this issue