computer-use Skill (Orca): Let Claude Control App Windows
Lets your coding agent click, type and take screenshots in a visible app window on your computer, through the Orca CLI.
- Installs
- 241K
- on skills.sh
- Stars
- 112.4K
- on GitHub
- Updated
- Nov 2026
- last repo commit
This skill is a short loader. It tells the agent how to pick the right Orca program on your machine, then run orca skills get computer-use to print the full guide that matches your installed Orca version. That guide is not in this repo file, so this page cannot describe its commands.
The skill description says it drives the GUI of a visible local app window: accessibility tree, clicks, typing, menus, dialogs and screenshots, in native apps and in Chrome, Edge or Safari windows and webviews. It says to use it only when a visible window needs GUI control that a shell, filesystem, git, HTTP call or existing CLI cannot do. It is not for Orca's embedded browser, which uses orca-cli. It needs the Orca app installed, because the real instructions come from it.
Good for
- Orca users whose agent needs to drive a desktop app or browser window
- Developers who want an agent to test a native app or webview
- Anyone with a task that no shell command or API can reach
Install computer-use
Claude Code
- 1
Copy this prompt.
Install the computer-use skill in Claude Code. Run this command: npx skills add https://github.com/stablyai/orca --skill computer-use Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires Node.js.
From its listing on skills.sh. Source.
What's in the skill
When it loads
Drives the GUI of a visible local app window through `orca computer`: accessibility tree, clicks, typing, menus, dialogs, and screenshots in native apps and external browser windows (Chrome, Edge, Safari) or webviews. Prefer a programmatic path (shell, filesystem, git, HTTP, existing CLIs) whenever it can complete the task. Use only when a visible window needs GUI control those cannot reach. Do not use for Orca's embedded browser (`orca-cli`).
Sections in SKILL.md
- Resolve the CLI for this session
- Load the version-matched guide before running Orca commands
Key facts
- Author
- stablyai (Orca)
- License
- MIT
- Installs on skills.sh
- 241,024 (Nov 2, 2026)
- GitHub stars
- 112,427 for the stablyai/orca repo (Nov 2, 2026)
- Last repo commit
- Nov 1, 2026
What it includes
Picks which Orca program to run: ORCA_CLI_COMMAND, orca-dev, orca-ide or orca
The agent chooses one executable and reuses it. It uses ORCA_CLI_COMMAND if set, orca-dev in a dev checkout, orca-ide on Linux outside an Orca terminal, otherwise orca. If it cannot run, the agent reports the error and stops.
Runs code ·
Resolve the Orca commandLoads the full GUI-control guide that matches your Orca version
Runs the Orca program to print the real guide. The agent must read it before running any other Orca command. Prefer the --json flag.
Runs code ·
orca skills get computer-useStarts Orca if a command says it is not running
Used once, then the failed command is retried.
Runs code ·
orca open --jsonOn runtime_access_denied, re-run with escalated permissions; do not restart Orca
If your sandbox blocks the connection to Orca, the skill tells the agent to retry with escalated permissions instead of running open or restarting.
Permission failure handlingUse shell, files, git or HTTP when they can finish the job; GUI only as a last resort
The description tells the agent to prefer non-GUI routes and not to use this skill for Orca's embedded browser.
Programmatic first ruleAccessibility tree, clicks, typing, menus, dialogs, screenshots
The skill description lists these actions. Their exact commands come from the guide Orca prints, which this page has not seen.
GUI actions via the loaded guide
From the official docs
Permissions
Medium risk
The agent can click and type in your open windows as if it were you. Close sensitive apps first and watch the first runs.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
It can click and type in any open app window
The agent acts as you in visible windows. Close banking, mail and admin apps before you run it, and watch what it does.
It may ask the agent to retry with escalated permissions
If your sandbox blocks the connection to Orca (error runtime_access_denied), the skill tells the agent to re-run the command with escalated permissions. Approve that only if you trust the Orca command being run. github.com
It runs the orca program from your machine
The real instructions come from the installed Orca version, not from this file. Read the guide it loads if you want to know what it will do. github.com
Two of three skills.sh audits warn
Gen Agent Trust Hub and Snyk show warn, Socket shows pass. Open the audit page on skills.sh before you install. skills.sh
Questions
What does the computer-use skill do?
It lets your agent operate a visible app window: read its accessibility tree, click, type, open menus, handle dialogs and take screenshots. It does this through the orca computer commands. The SKILL.md itself is only a loader for the guide stored in your Orca install.
How do I install it in Claude Code?
Use the install box above, which lists the skills CLI command from skills.sh. You also need the Orca app, since the skill calls its orca program.
Does it work in the Claude app or Codex?
The skill does not name specific clients. It runs shell commands, so it needs an agent that can run commands on your machine and reach Orca.
Does it need an API key?
The SKILL.md mentions no API key. It does need the Orca app and, in managed WSL sessions, the ORCA_CLI_COMMAND variable that Orca sets.
Is computer-use safe?
It gives the agent control of real windows on your computer, so treat it like handing over your mouse and keyboard. The skill is a loader, so what it does next depends on the guide your installed Orca version prints. Watch the first runs and keep banking, mail and admin apps closed.
Why does it say not to run bare orca on Linux?
On Linux outside an Orca terminal, orca normally points to the GNOME Orca screen reader and would start speech on your machine. The skill says to use orca-ide there.
Information
updated Nov 2, 2026
- Repository
- stablyai/orca
- Folder
skills/computer-use
Sources
Docs, issue trackers and discussions this page is built from.
Show all 1Hide
- stablyai/orca computer-use SKILL.md · github.com