Menu
Operate a connected computer-use node to accomplish an on-screen task — screenshot, reason, act in a loop.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | drive-computer-use |
| description | Operate a connected computer-use node to accomplish an on-screen task — screenshot, reason, act in a loop. |
You can see and control a connected Mac via the computer_* tools. Accomplish the user's on-screen goal carefully and verify each step.
Call start_computer_use_session with a one-sentence goal. This requires user approval. Once a session is active, DO NOT call start again — go straight to acting.
Work the loop:
computer_screenshot to see the current state. The coordinates you pass to click/move are in the RETURNED image's pixel space (the node maps them to the display for you).computer_key {chord} — e.g. "cmd+space" (Spotlight), "enter", "cmd+a", "esc".computer_type {text} — type into the focused field.computer_click {x,y,button} / computer_move {x,y}.Failure modes to recognize and report instead of flailing:
scroll is unsupported — find another way (keyboard, or click a scrollbar target).When the goal is met or you're blocked, call end_computer_use_session with a brief reason, and report what you did. Every action is logged for the user to review.
User asked: {{input}}
Guide configuring proactive outreach so the main reaches out (digest + live decisions) and the Mac app surfaces it with notifications, an overlay list, and inline buttons.
Roll out new openAGI code to the main and to Mac nodes (iMessage / computer-use), and restart them safely.
Troubleshoot opaque model/tool failures and runaway budget — and their usual root causes.
Troubleshoot why the agent stopped replying to iMessages, in priority order.
Connect an MCP server (catalog or custom URL/stdio) and make its tools usable — including completing OAuth on a headless host.
Set up a recurring check that only does work (and spends tokens) when there's something new — e.g. "every hour, check BuildBetter for new calls".