agentleFS
songkeyssongkeys/claude-codex-computer-use/skills/computer-use/SKILL.md

computer-use

Skill13 starsChanged 3 months agoMITView on GitHub

Control local macOS applications through the installed Computer Use MCP tools. Use for GUI work that is not better served by a dedicated API, connector, or CLI.

68 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use competitor-news-monitor, claude-buddy and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • aivrarSkill

    competitor-news-monitor

    Watch named companies for material news; cited digests.

    Also in 6 projects that use this

  • autonomous-aiSkill

    claude-buddy

    Coordinate with the Claude Buddy companion (Claude Desktop + Claude Code on the user's Mac) — voice-approve tool permission prompts and stay aware of Claude activity.

    Also in 3 projects that use this

  • autonomous-aiSkill

    face-enroll

    Manage the device's face recognition roster — enroll new faces (3 paths: user-supplied photo, agent-captured snapshot on user request, or HAL's familiar-stranger prompt) and maintain the enrolled set (status / remove / reset). All enrolled persons are friends; strangers stay unnamed until promoted via one of the enroll flows.

    Also in 3 projects that use this

  • +9 more from autonomous-ai

The file

We split this skill into the rules it gives an agent

8 rules. 0 are common, 3 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 63 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Start with get_app_state for the named application.

    line 18
  2. Distinctive:

    Read the returned accessibility text first.

    line 20
  3. Very specific:

    Prefer actions using the latest element_index.

    line 22
  4. Very specific:

    Call get_app_state again before deciding the next action.

    line 25
  5. Very specific:

    No separate launch step is needed.

    line 49
  6. Very specific:

    If targeting by display name fails, call list_apps, find the bundle identifier, and retry once with that identifier.

    line 53
  7. Distinctive:

    If an action behaves unexpectedly, immediately fetch a fresh app state.

    line 60
  8. Distinctive:

    If a tool result says the application session was explicitly stopped by the user, the user pressed Esc on the Computer Use overlay.

    line 64

“Common” is measured: kept by at least 6 in 10 of the 63 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Computer Use

This plugin exposes OpenAI's locally installed Computer Use service directly as Claude Code MCP tools. Use the tools whose names begin with mcp__plugin_claude-codex-computer-use_computer-use__.

The bridge automatically handles the service's app-access elicitation. Do not ask the user to approve Computer Use access and do not attempt to manage the OpenAI app's allowlist.

Required observation and action loop

  1. Start with get_app_state for the named application. Pass the app's display name or bundle identifier directly.
  2. Read the returned accessibility text first. Use the accompanying screenshot when the accessibility tree is incomplete or visual context matters.
  3. Prefer actions using the latest element_index. Use coordinates only when the target is unavailable or unreliable through accessibility.
  4. Perform one or more tightly related actions.
  5. Call get_app_state again before deciding the next action. Re-derive element indices from the latest state and never assume an old index is still valid.

The get_app_state result may be a diff from the preceding state. Request a full state only when the earlier tree is unavailable or a diff is insufficient.

Tools

  • list_apps: discover apps only when the requested app cannot be identified from its name or bundle identifier.
  • get_app_state: read accessibility state and capture a screenshot.
  • click: click by element_index, or by coordinates as a fallback.
  • set_value: replace the value of an editable accessibility element.
  • type_text: type into the target app.
  • press_key: send a key or key combination such as Return, Tab, super+c, or Escape.
  • scroll: scroll an element or view.
  • drag: drag between coordinates.
  • select_text: select matching text or place the cursor around it.
  • perform_secondary_action: invoke an accessibility action explicitly exposed by an element. Never guess the action name.

App targeting

  • No separate launch step is needed. get_app_state launches the app when necessary.
  • Try the app name or known bundle identifier directly before calling list_apps.
  • If targeting by display name fails, call list_apps, find the bundle identifier, and retry once with that identifier.
  • press_key and type_text target the selected app; they are not global keyboard shortcuts.

Recovery

If an action behaves unexpectedly, immediately fetch a fresh app state. If the accessibility tree is sparse, inspect the screenshot and use coordinates. Keep the loop observable: state, action, state.

If a tool result says the application session was explicitly stopped by the user, the user pressed Esc on the Computer Use overlay. Stop the Computer Use loop, tell the user, and continue only when they ask. The bridge recycles the client automatically, so the next attempt starts a fresh session.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.