菜单
Control the computer by taking screenshots, moving the mouse, clicking, and typing. Use this skill when you need to interact with GUI applications or perform actions that require visual feedback.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Reverse-engineer an existing software project so the user can understand the code logic, architecture, workflows, design decisions, and potential issues. Use this when the user provides a GitHub repo, local codebase, zip file, or existing project and wants architecture understanding, Mermaid diagrams, remake guidance in another language, or expert design/performance evaluation.
Perform deep research from a research requirement or instruction file and produce research outputs in the same folder, including Markdown plus an index.html visual report with Mermaid diagrams when useful.
Open and operate the local Skill Pilot WebUI through the browser for demos, walkthroughs, settings changes, task execution, and terminal-based education flows. Use when the user wants the agent to drive Skill Pilot as a human would in the browser, open a specific WebUI feature, or jump directly into shell or agent terminal views.
Use this skill when acting as a Black-Box Code Agent Supervisor. The supervisor uses coding agent CLIs such as Claude Code, Codex, OpenCode, Gemini CLI, or other AI coding agents through tmux terminal windows to finish a coding task or project. The supervisor works like a real developer using AI agents: giving instructions, checking the running result, using browser/curl/logs, finding issues, and sending feedback back to the coding agent. The supervisor does not directly inspect source code, write code, or write code-level tests.
Manage the full lifecycle of a Vibe Coding project — create, refine, brainstorm, initialize, plan, implement, review, test, deploy, update, and fix issues. Use when the user wants to start a new project from a prompt, evolve its requirements, plan or write code, validate or deploy it, or apply updates and bug fixes against an existing project.
Create interactive courses and assignments for students in the project's markdown format. Use when asked to create a new course, educational module, or assignment.
| name | use-computer |
| description | Control the computer by taking screenshots, moving the mouse, clicking, and typing. Use this skill when you need to interact with GUI applications or perform actions that require visual feedback. |
This skill enables an agent to interact with the computer's graphical user interface (GUI). It can retrieve screen information, take screenshots of specific regions, and perform mouse and keyboard actions.
As an expert in your assigned roles, you must announce your actions before performing them using the following format:
As a {Role} [and {Role}, ...], I will {action description}
This communication pattern ensures transparency and allows for human-in-the-loop oversight at key decision points.
All interactions are performed via the core/bin/use-computer CLI from the project root:
core/bin/use-computer --json_str '<JSON_COMMAND>'
The script returns a JSON response wrapped in <output></output> tags.
Use this to understand the monitor setup and primary screen resolution.
Command:
{"action": "screen_info"}
For screenshot payloads, return fields, scaling, pointer markers, and coordinate correction, refer to references/screenshot.md.
Before any mouse_move or click, use agent skill find-ui-element to locate the target UI element and derive coordinates.
Recommended flow:
draw_pointer: true.find-ui-element on that screenshot to get the element bounding box.input actions.For mouse, keyboard, hotkey, typing, and wait action payloads, refer to references/input-actions.md.
If the task needs a visible macOS Terminal window backed by tmux, refer to references/macos-terminal.md.
scale (e.g., 0.5 or 0.25) when taking screenshots for large screens to reduce processing time and token usage if sending to an LLM.