I use both Cursor and Codex simultaneously. I’ve noticed that in the settings interface, Cursor’s Rule automatically loads the AGENTS.md file from the project directory. My AGENTS.md is a soft link to Cursor’s rule.mdc file. Both .mcd and AGENTS.md are loaded in the interface. Could this lead to duplicate context and occupy unnecessary context space?
Also, as shown in the image, my multi-project workspace is named “Next,” but in Cursor’s Rule interface, apart from the User Rule, it displays the name of my first project, “Frontend,” which doesn’t seem very reasonable. It would be advisable to make improvements.
Yep, this will duplicate context. Cursor treats AGENTS.md and .cursor/rules/*.mdc as separate rule sources and loads them separately. It doesn’t detect that one is a symlink to the other. So if your AGENTS.md points to rule.mdc, both will be injected into the system prompt, basically doubling token cost for the same content.
The fix is simple: keep only one. If you need AGENTS.md for Codex compatibility, keep it and delete the .mdc file (or set alwaysApply: false). If you mainly use Cursor, keep the .mdc and remove the symlink.
Workspace name in the Rules tab
This is a known cosmetic issue. In multi-root workspaces, the Rules tab shows the name of the first folder instead of the actual workspace name. A similar report is here: Bug: Settings > Rules tab label doesn't update after "Save Workspace As...". The team is aware. There’s no ETA, but it’s being tracked. This doesn’t affect functionality. Rules still apply correctly regardless of the label.
Cursor combines rules. E.g. if you have AGENTS.md and .mdc files they are loaded both. Since they contain duplicate info it does unnecessary use tokens and context.
deanrie covered the fix well. one thing i can confirm from testing: i ran the same rules in both AGENTS.md and individual .mdc files with alwaysApply: true and got identical compliance from the model either way. 2 runs each, 100% both times. so theres no behavioral reason to have both, its purely wasted tokens.
for your dual-tool setup (Cursor + Codex), keeping AGENTS.md as the single source makes sense since both tools read it natively. the tradeoff is you lose file-type scoping. i just tested this actually , a .mdc file with globs: ["*.tsx"] only fired for .tsx files and correctly ignored .js files. same rule in AGENTS.md applied to everything regardless of file type. so if you need rules that only kick in for certain files, you still need .mdc files for those. just make sure theres no overlap between whats in AGENTS.md and whats in the .mdc files.
