How to use the Agent Manager (tmux) agent skill

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

An agent skill is a folder containing a SKILL.md file that gives an AI agent instructions for a specific kind of task. This skill, named agent-manager-skill, covers running multiple local CLI agents in parallel, each in its own tmux session.

Once installed, the agent knows how to list configured agents and existing tmux sessions, start or stop an agent, tail its logs, and send it a task while monitoring the output. The skill is described as a scoped tool for an explicitly authorized workspace.

• Run several local CLI agents in separate tmux sessions
• Start and stop agents and tail their logs
• Assign tasks to agents and monitor their output
• Prepare a plan for repeated agent work only when scheduling is explicitly requested

What you need before installing

The skill does not bundle the manager program it describes. You need a locally installed agent-manager runtime, and the upstream clone address could not be verified during review, so do not assume the examples will run as written.

The skill's notes list these requirements: tmux, python3, a locally installed agent-manager runtime, and locally configured agents under an agents/ directory. The notes also point to the repository for examples of agent configuration.

• tmux installed and available on your PATH
• python3
• A locally installed agent-manager runtime (not included in the skill)
• Locally configured agents stored under an agents/ directory

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. The skill folder contains SKILL.md. Put the folder in one of two places depending on how you want it available.

For a personal install that applies across your projects, place it in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Single project: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills

Trigger the skill with natural requests

You do not need a special command. Describe the task in plain language that matches the skill's scope, such as running agents in parallel, checking logs or assigning work. The agent will decide whether to use the skill based on its description.

Be specific about which agents and which workspace you mean. Because the skill only acts on actions you authorise, naming the target agent or session helps the agent stay within the boundary you intend.

• "List my configured agents and any running tmux sessions before doing anything."
• "Start the agent EMP_0001 and show me its log." (Replace the ID with one from your own configuration.)
• "Tail the logs of the agent in session X and tell me whether its task finished."

Safety: scripts, network and authorization

The skill runs scripts and uses the network, so treat it as a tool with real effects on your machine. The skill's own guidance says to list your actual configured agents and tmux sessions first, and that starting, assigning or stopping an agent requires authorization for that specific task and target.

Preserve sessions you did not start, record the session or PID and the command you inspected, and redact logs before sharing them. A running session does not prove the task completed, so check the result yourself. The skill includes no cron installer or automatic scheduler, so it does not support unattended runs.

• Confirm each start, assign or stop action before it runs
• Do not kill sessions unrelated to the task
• Redact secrets from logs before sharing them
• Verify the output rather than trusting a live session
• Do not run the example commands until your installed script and its CLI help confirm them

Example command shapes (verify before use)

The skill lists example commands for an upstream script. They are historical and unconfirmed in this package, so treat them as a reference for the shape of commands, not as instructions to copy. Resolve the real script path in your installation first, and replace the example employee IDs and team files with your verified configuration.

The examples cover a health check, listing agents, starting and monitoring one agent, and assigning a task from a file.

Tips for getting reliable results

Start with a read-only check. Ask the agent to list agents and sessions before any change, so you can see what exists. Keep the scope small at first, with one agent and one clearly defined task.

Use the skill only when the task clearly matches what it describes. For anything involving scheduling or unattended operation, plan the work yourself rather than relying on the skill.

• Begin with listing agents and tmux sessions
• Assign one task to one agent at a time while learning
• Check results independently of whether a session is still running