菜单
Agent Skill for browser automation over the Chrome DevTools Protocol connection. Use when a task requires interacting with or inspecting live web pages, forms, UI behavior, or any scraping tasks like data extraction, form filling, etc, or any browser automation tasks.
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
Agent Skill for planning implementation work into comprehensive, testable steps. Use before implementation to create a clear roadmap for execution.
Agent Skill for writing and reviewing durable design specs. Use when the user asks to write, formalize, save, review, or approve a spec, design document, RFC, or implementation-ready proposal. If the direction is still vague or unsettled, use `brainstorming` first before producing the spec. If the direction is already clear and no spec is needed, use `plan`.
Agent Skill for spawning new AI agents with interactive TUI in tmux. Use when work can be offloaded to new or existing AI agents (exploration, code review, second opinions, parallel work, etc.). Can also be used if spawning new agents with different model families (e.g. Spawn GPT models when you are Claude Code, Spawn Claude models when you are Codex, Or spawn any other models (e.g. DeepSeek, Gemini) with Pi). Prefer harness's own subagent tools when no need for a different model.
Agent Skill for OpenCLI. Use when the user wants to control websites through the `opencli` CLI, run or inspect OpenCLI site/app/external adapters, use the OpenCLI browser bridge/CDP automation
Agent Skill for collaborative brainstorming and design exploration. Use when the user wants to think through an idea, compare options, clarify trade-offs, or get a quick recommendation, without forcing a written artifact. If the user wants a saved, reviewed design spec, use `design-spec`.
Agent Skill for working with Python. Use when writing, editing, or reviewing Python code and scripts.
| name | agent-browser |
| description | Agent Skill for browser automation over the Chrome DevTools Protocol connection. Use when a task requires interacting with or inspecting live web pages, forms, UI behavior, or any scraping tasks like data extraction, form filling, etc, or any browser automation tasks. |
agent-browser command is available
Public Internet Access is preferred but not required
A Chromium-based browser (e.g. Chrome, Edge, Chromium, Brave) must be running
Only --auto-connect and --cdp connection are allowed. If both options fail, notify the user to fix it.
--auto-connect (Preferred)This will attempt to automatically find the debugging port and connect to it.
Example: agent-browser --auto-connect get-cdp
--cdp (Explicit)Use --cdp when you already know a CDP endpoint.
This is the best choice for a known local port, a full CDP WebSocket URL, or a browser-as-a-service provider over ws:// or wss://.
Use curl to check if http://<ip>:<port>/json/version returns a websocket URL when connecting by local port.
Examples:
agent-browser --cdp 9222 get cdp-url
agent-browser --cdp ws://127.0.0.1:9222/devtools/browser/<id> get cdp-url
agent-browser --cdp wss://your-provider.example.com/devtools/browser/<id> get cdp-url
If both --auto-connect and --cdp fail, notify the user to start their browser with remote debugging enabled.
If you are on Windows, read windows-specific-convenience before running your first agent-browser command.
agent-browser --help to see the latest manual and all available options.agent-browser skills as mentioned in the help message, as it provide false positive.agent-browser with --color-scheme no-preference to override the default light mode in agent-browser.agent-browser with at least one of --auto-connect or --cdp.agent-browser install to install other browsers.