Skip to content

Mac Computer Use Router

Routes Mac control tasks to Interceptor or a supervised vision agent, with deny lists and a verifier.

Stars
19,682
Licence
MIT
Files
14 files

What it does

This skill teaches an AI agent how to operate a Mac on your behalf. It first decides how each task should be done: anything the system can reach directly (through structured interfaces, commands or APIs) goes to Interceptor, while purely visual, pixel-only GUI work goes to a vision agent (OpenAI Astra or a tool-less Claude). The run is supervised with a deny list of credential and messaging apps and denied web hosts, step and time caps, an app allowlist, and a verifier script whose exit code alone decides pass or fail.

Great for

  • Developers and power users who already run Interceptor and want a documented, auditable way to let an agent drive Mac apps
  • Teams evaluating which vision model handles canvas or pixel-only app interfaces best, using the included bake-off tooling
  • People who want strict guardrails around GUI automation, such as blocking credential apps from running during a task

Skip it if

  • Anyone without a Mac, since the scripts depend on macOS APIs, Apple Events and Interceptor; the skill is not useful elsewhere
  • Users who want unsupervised automation that sends, buys, deletes or publishes things without review, because those actions are excluded and stop the run by design; a misconfigured deny list or verifier could still expose data or cause unintended actions
  • Simple web scraping or data extraction, which the skill itself routes to a separate scraping skill

Try asking

  • “Use computer use to fill in the invoice form in this app and save it as a PDF”
  • “Drive the canvas editor in this app to create a new layout, then verify the result”
  • “Run the computer-use bake-off to compare Claude and OpenAI on the design app”
  • “Probe whether the Astra vision leg is working on this machine”
Step-by-step guide: how to use Mac Computer Use RouterInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 14 original files, the MIT licence and an attribution note.

    ComputerUse.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip ComputerUse.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip ComputerUse.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Complete Setup.md, including a dry-run proof, before running real tasks, and create a careful Config.md deny list first.
  • Give each task a clear, checkable verifier so the run verdict reflects real outcomes rather than the agent's own report.
  • Keep the default 60-step and 10-minute caps unless you have a specific reason to raise them.

What's inside

  • Config.example.md
  • Instructions.md
  • Intents/BakeOff.md
  • Intents/Drive.md
  • Intents/Probe.md
  • References/VisionLegs.md
  • Setup.md
  • SKILL.md
  • Tools/BakeOff.ts
  • Tools/ClaudeLeg.ts
  • Tools/Probe.ts
  • Tools/Route.ts
  • Tools/Run.ts
  • Tools/WindowBounds.ts

Redistributed unmodified from danielmiessler/LifeOS under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.