Sync user rules and agents to Cloud Agents

Ideas Feature Requests
, , ,
1

Feature request for product/service

Agents Window and Cloud Agent

Describe the request

Cursor already supports syncing local user level skills (~/.cursor/skills) to Cloud Agents.

Could you also support syncing ~/.cursor/rules and ~/.cursor/agents to the Cloud Agents. More generally, why not support syncing everything under ~/.cursor/ to the Cloud Agents?

5

Feature request for product/service

Cloud Agent (GitHub, Slack, Linear,…)

Describe the request

Cloud Agent launch setting to sync Rules, AGENTS.md, and CLAUDE.md

Feature request for product/service

Cloud Agent

Describe the request

Cursor already has Sync Skills for Cloud Agents (Settings → Agents), which copies ~/.cursor/skills/ so personal skills work in the cloud. I want the same kind of opt-in setting for instruction files and rules when a Cloud Agent starts.

Problem

Locally, agent behavior is defined in files like:

  • AGENTS.md (shared agent playbook for Cursor / Codex / other agents)
  • CLAUDE.md (Claude Code; Cursor Desktop already reads this the same way as AGENTS.md)
  • .cursor/rules/*.mdc and ~/.cursor/rules

Those files are the actual rules for how AI agents should work. Cloud Agents do not have a first-class way to use the same local files:

  • Cloud Agents clone git, so uncommitted AGENTS.md / CLAUDE.md / project rules never arrive
  • ~/.cursor/rules stays on the machine and does not sync (account User Rules do; file-based rules do not)
  • Move to Cloud transfers chat history, not dirty files
  • Copying the same text into a Skill is a workaround, and skills are not always-on rules

Account User Rules and committed repo AGENTS.md help, but they are a different store. I do not want to commit personal or mixed-tool playbooks into every repo just so Cloud Agents behave like local Cursor / Claude Code.

Request

Add an opt-in Cloud Agent setting, similar to Sync Skills, for example:

  • Sync file-based Rules (~/.cursor/rules, and optionally uncommitted project .cursor/rules)
  • Sync instruction markdown: AGENTS.md, CLAUDE.md (project, plus optional user-level such as ~/.claude/CLAUDE.md)
  • Apply them as always-on instructions for that Cloud Agent run, not only as a Skill the model may ignore

A checkbox at Cloud Agent launch, or a toggle under Settings → Agents, would be enough. Default off for privacy.

Related existing request (narrower: ~/.cursor/rules and ~/.cursor/agents only):

This request is specifically about instruction files (AGENTS.md / CLAUDE.md) plus a launch-time setting, matching how local Cursor already loads CLAUDE.md.

Operating System (if it applies)

macOS

Operating System (if it applies)

MacOS

8

Hey Archit, thanks for the feature request.

A couple things that can help right now:

  • User Rules (Customize > Rules) are tied to your account, so Cloud Agents will automatically pick them up on every run. You don’t need to commit anything to your repo. But file-based rules in ~/.cursor/rules and custom subagents in ~/.cursor/agents are still local to your machine for now.
  • If you’re using a custom Cloud Agent environment, you can bake those files into the environment image so the agent can find them there.

Related threads where this is discussed: Global rules for Cursor Cloud Agents? Global rules for Cursor Cloud Agents? and Rules in cloud agent Rules in cloud agent.

Syncing the rest of ~/.cursor/ the same way skills work makes a lot of sense. I’ll keep the thread open so other users can add their votes.

9

Hey guys, I do not claim I found the perfect work arround, and I’m still testing, but I always had some computers laying arround in my house, I plugged one in, installed all the rules and repository I want the exact way I wanted, and I plan to use them as “remote machine” I don’t yet know if it will work, but I’ll post updates here

10

Hey, thanks for sharing. What you’re describing is pretty close to an existing feature called My Machines, which is part of Self-Hosted Machines for Cloud Agents. The idea is exactly this: the agent loop runs in the cloud, but file edits, terminal commands, and tool calls run on your machine, which is already set up the way you want, including your local ~/.cursor/rules and repositories. No open ports or port forwarding is needed since the worker opens an outbound connection.

How to set it up:

So you can use your home computer not as a DIY hack, but as the standard execution environment. That way, your local rules and agents files are available to the agent as-is.

That said, the request to sync file-based rules or AGENTS.md in managed Cloud Agents still makes sense, so I’m leaving the thread open for votes. Let me know how it goes with the worker.

11

I exactly used this, in fact, I haven’t posted back, since there’s couple of problems, first of all, I cannot /use the rules i have in this repo, but since I’m using a project, it’s okay, still, the feature request could be cool.