phantom
Computer use toolkit for desktop automation via Claude’s vision and action API.
Overview
Phantom enables Claude to interact with desktop environments through screenshot capture, mouse/keyboard control, and an autonomous agent loop. It wraps Claude’s Computer Use API for sandboxed GUI automation workflows.
Security Precautions
Computer use grants Claude direct control over mouse, keyboard, and screen reading. Follow these precautions:
- Run in a sandboxed environment (VM, container, or dedicated machine). Never run on a machine with access to production systems or sensitive credentials.
- Review tasks before execution. The
/control-desktopcommand displays the planned actions. Confirm before allowing execution. - Limit network access. Restrict outbound connections from the sandbox to prevent data exfiltration if the agent navigates to an unintended URL.
- Do not store credentials in the sandbox environment. If a workflow requires login, use temporary tokens with narrow scope.
- Monitor active sessions. The desktop-pilot agent runs autonomously. Watch for unexpected navigation or input actions and terminate if behavior deviates.
Installation
/plugin install phantom@claude-night-market
Skills
| Skill | Description | When to Use |
|---|---|---|
computer-control | Desktop automation via screenshot capture, mouse/keyboard control | Automating GUI tasks in sandboxed environments |
Commands
| Command | Description |
|---|---|
/control-desktop | Run a computer use task on the desktop |
Agents
| Agent | Description |
|---|---|
desktop-pilot | Autonomous desktop control with multi-step GUI workflows |
Usage Examples
Control a Desktop
# Run a GUI automation task
/control-desktop "Open the browser and navigate to example.com"
# Use the agent for multi-step workflows
Agent(phantom:desktop-pilot)