Skip to content
Documentation: Cursor

Cursor

Where rmk puts each type of item for Cursor, and how it works with Claude Code.

From the Documentation in Ronne AI Marketplace 0.3.0.

Where each type goes

With --target cursor, rmk writes each item where Cursor reads it, for the editor and its agent CLI: in the project, or in your home folder with --scope user.

TypeWhere it goes
skill.agents/skills/<name>/ (your home folder: ~/.agents/skills/)
agent.cursor/agents/<name>.md
rule.cursor/rules/<name>.mdc (projects only: Cursor keeps your own rules in its settings)
command.agents/skills/<name>/, run as /<name>
hookhooks in .cursor/hooks.json; a script under .cursor/hooks/<name>/
mcp-servermcpServers in .cursor/mcp.json
permission-policypermissions in .cursor/cli.json (home: ~/.cursor/cli-config.json), for the CLI
output-stylenot supported: skipped with a warning
statuslinenot supported: skipped with a warning
lsp-servernot supported: skipped with a warning
bundlenothing of its own: its items are installed one by one

rmk export --from cursor reads the agents, rules, commands and MCP servers you wrote for Cursor from the same places: Exporting your own items.

With Claude Code

Cursor also reads Claude Code's files: its skills, its agents, and the hooks in its settings, which Cursor runs as well as its own. That's Cursor's Third-Party Imports setting, on by default.

So when you install for both (--target claude-code,cursor), Cursor leaves skills, commands and hooks to Claude Code's copy instead of writing a second one, and rmk says so for each. If you turn Third-Party Imports off in Cursor, install for Cursor alone to get its own copies. Agents, rules, MCP servers and permissions are written for both tools.

Good to know

  • Skills and commands go to .agents/skills/, the folder Codex reads too, so one copy serves both. Commands become skills you run as /name; Cursor doesn't pass them arguments, so a note in the skill says what each placeholder is.
  • Rules become .cursor/rules/ files: always on, for matching files, picked by the AI from their description, or added when you mention them with @name. Cursor keeps your personal rules in its settings, so a rule installed with --scope user is skipped.
  • Agents have no tool list in Cursor. An agent whose tools don't change files or run commands is written as read-only; its model is Cursor's default unless the item sets a Cursor model.
  • MCP servers reference their secrets as ${env:NAME}: nothing secret is written, and you set the variables yourself.
  • Permission policies go to the agent CLI's config, which has allow and deny but no ask, so ask rules are left out. Cursor documents these permissions only for its CLI, not the editor.
  • Output styles, status lines and language servers have no place in Cursor, and are skipped with a warning.

Plugins

A Cursor team admin can import a git mirror of this marketplace, written by rmk feed build, as a team marketplace, so the team installs items as plugins. Plugin marketplaces explains how.