菜单
Browser automation for development testing, web scraping, and UI verification. USE WHEN you need to interact with web pages, verify UI changes, take screenshots, or extract data from websites.
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
| name | AgentBrowser |
| description | Browser automation for development testing, web scraping, and UI verification. USE WHEN you need to interact with web pages, verify UI changes, take screenshots, or extract data from websites. |
| allowed-tools | Bash(agent-browser:*) |
You control a headless Chromium browser for web automation tasks. This skill provides CLI commands to navigate, interact with, and inspect web pages. The daemon starts automatically on first use — no explicit start command needed.
Determine Action: Based on the user's request:
open then snapshotclick, fill, or evalscreenshotwaitExecute Commands: Use the appropriate CLI commands via Bash tool
Report Results: Summarize what you found or accomplished
Sessions persist browser state (cookies, localStorage) across commands. Use --session before the subcommand to specify a named session.
--session is omitted, the "default" session is used| Command | Usage | Description |
|---|---|---|
open | agent-browser --session <id> open <url> | Navigate to URL |
snapshot | agent-browser --session <id> snapshot -i | Get interactive elements with refs |
click | agent-browser --session <id> click <ref|selector> | Click element |
fill | agent-browser --session <id> fill <ref|selector> <text> | Clear and fill input |
screenshot | agent-browser --session <id> screenshot [path] | Take screenshot |
eval | agent-browser --session <id> eval <script> | Run JavaScript |
wait | agent-browser --session <id> wait <ref|selector> | Wait for element |
close | agent-browser --session <id> close | Close browser session |
See commands.md for the full command set.
| Workflow | Trigger Words | When to Use |
|---|---|---|
| BrowseWeb | "browse", "open", "navigate", "check", "verify", "screenshot", "scrape" | User wants to interact with or inspect a web page |
snapshot -i after navigation to understand page structure--session ID across related commands to maintain statescreenshot --annotate for visual context alongside text snapshotsTest NixOS configuration changes using the feedback ladder — from instant syntax checks to full system builds. USE WHEN the user wants to verify, test, or check NixOS changes before deploying.
Review and fix documentation surprises discovered by the surprise hook. USE WHEN the user wants to fix documentation issues, review surprises, or address items in .claude/surprises/.
Query systemd journal logs, check service status, diagnose failed units, and analyze boot performance. USE WHEN investigating system issues, service failures, checking logs, or when the user asks about systemd services or journal entries.
Systematically explore and test a web application to find bugs, UX issues, and other problems. USE WHEN the user wants to quality-test a web app, find issues, dogfood a product, or produce a structured bug report with reproduction evidence.
Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) via Chrome DevTools Protocol. USE WHEN the user needs to interact with an Electron desktop app, automate a native app, connect to a running Electron app, or test an Electron application.
Interact with Slack workspaces using browser automation. USE WHEN the user needs to check unread channels, send messages, search conversations, navigate Slack, or extract data from Slack workspaces without using the Slack API.