How to use the PM2 Process Manager CLI agent skill

What is an agent skill and what does this one do?

An agent skill is a folder of instructions, bundled with any scripts it needs, that an AI agent can load when a task calls for it. This skill is described by a single file called SKILL.md. Once it is installed, the agent knows how to run a command-line wrapper named cli-anything-pm2 to control PM2 on your machine.

The skill lets the agent list the PM2 processes that are running, look at their metrics, start or stop them, restart them, delete them, and read their recent logs. It also covers a few system-level tasks: saving the current process list, generating a startup script, and reporting the PM2 version.

The wrapper is stateless. It sends each request to the PM2 CLI as a subprocess and keeps no session of its own, so every command stands alone.

Install the cli-anything-pm2 skill

Download the skill as a ZIP file from this page, then unzip it. Put the resulting folder in one of two places. For a personal setup, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

The skill also needs its command-line package installed. The SKILL.md file gives this command, run from the package directory:

• Python 3.10 or newer must be available.
• PM2 must be installed globally with npm. The SKILL.md file uses: npm install -g pm2

Check that the wrapper works before you trust the agent

Before asking an agent to act, run a few commands yourself. The --help flag lists what the CLI offers. Adding --json before a command makes the output machine-readable, which is the format the agent uses.

If these commands return your process list, the agent will be able to reach PM2 too.

• Show help: cli-anything-pm2 --help
• List processes as JSON: cli-anything-pm2 --json process list
• Check the PM2 version as JSON: cli-anything-pm2 --json system version

Trigger the skill with natural requests

You do not need to name the skill or type commands. Describe the outcome you want in plain language, and the agent should pick the matching commands. Be specific about the process name, because restart, stop and delete all act on a named process.

The skill's command groups map to common requests. Inspection covers list, describe and metrics. Lifecycle covers start, stop, restart and delete. Logs covers view and flush. System covers save, startup and version.

• 'Show me every PM2 process and whether it is online.'
• 'Show the metrics for all my PM2 processes.'
• 'Read the last 50 log lines for the api process.'
• 'Restart the worker process and then check its logs for errors.'
• 'What version of PM2 is running on this server?'

Know the command groups

The agent uses these groups. You can run the same commands yourself to check its work.

The process group inspects processes. The lifecycle group starts, stops, restarts and deletes them. The logs group views or flushes logs. The system group saves the process list, generates a startup script, and reports the version.

• process: list, describe <name>, metrics
• lifecycle: start <script> --name <name>, stop <name>, restart <name>, delete <name>
• logs: view <name> --lines 50, flush [name]
• system: save, startup, version

Safety notes before you let an agent act

Restarting, stopping or deleting a process can take down a live service. Use the skill only on a machine you own or administer, and be clear with the agent about which process it may touch. Reading processes, metrics and logs is the lower-risk side of the skill.

The skill runs scripts on your machine, so it is not a read-only tool. The source states that it does not use the network. Review the SKILL.md file and the package before installing if your environment requires it.

The skill's own guidance for agents is to always use the --json flag, check return codes (0 means success and anything else means an error), and parse stderr for error messages when a command fails.

• Use it only on servers you control.
• Name the exact process for any stop, restart or delete.
• Read logs first when diagnosing a problem.
• Confirm the agent's commands match what you expected before it runs destructive ones.