How it works in Cursor
Cursor Agent loads skills from these directories and presents them as available capabilities:
When a skill is installed, Agent reads its
SKILL.md file and automatically decides when to apply it based on what you ask. You can also invoke any skill explicitly by typing /skill-name in the Agent chat.
Skills are visible in Cursor Settings → Rules → Agent Decides.
Install the skills
- Global install (recommended)
- Project-level install
- Install a single skill
Installs the skills for all your projects. Run once:This puts all 3 skills under
~/.agents/skills/:~/.agents/skills/setup-notifyer/~/.agents/skills/automate-notifyer/~/.agents/skills/chat-notifyer/
Set up environment variables
The scripts need two variables before Agent can use them:~/.zshrc or ~/.bashrc) so they’re always available when Cursor’s terminal runs scripts. Alternatively, add a .env file to your project — Cursor Agent can read it if you reference it in your prompt.
For full environment variable reference, see Auth & environment.
Verify discovery
After installing, confirm Cursor can see the skills:- Open Cursor Settings → Rules
- Look for
setup-notifyer,automate-notifyer, andchat-notifyerin the Agent Decides section
Example Agent prompts
Once installed, use natural language in Cursor Agent chat. Agent will identify the relevant skill and execute the appropriate scripts.Workspace setup
Template management
Live chat operations
Broadcast campaigns
Manually invoking a skill
If Agent doesn’t pick up a skill automatically, invoke it with the/ command:
How Agent runs the scripts
When you give Agent a task, it:- Reads the relevant
SKILL.mdto understand available scripts and their flags - Opens a terminal in the skill’s directory (
skills/setup-notifyer/, etc.) - Runs the script with the correct flags
- Parses the JSON output (
ok/errstructure) and summarises the result
isConnected: false and automatically suggest running refresh-connection.js.