automating-browser

thkt 的头像作者 thkt

使用 claude-in-chrome MCP 工具进行交互式浏览器自动化。适用于:演示、文档 GIF、手动测试、实时浏览器控制。对于 CI/CD 中的自动化端到端测试,请使用 webapp-testing 技能。触发词:浏览器自动化、ブラウザ自動化、屏幕截图、スクリーンショット、表单填写、フォーム入力、点击、导航、GIF 录制、GIF 録画、浏览器操作、Chrome、演示、デモ、实时浏览器。

📖 中文导读:使用 claude-in-chrome MCP 工具进行交互式浏览器自动化。适用于:演示、文档 GIF、手动测试、实时浏览器控制。对于 CI/CD 中的自动化端到端测试,请使用 webapp-testing 技能。触发词:浏览器自动化、ブラウザ自動化、屏幕截图、スクリーンショット、表单填写、フォーム入力、点击、导航、GIF 录制、GIF 録画、浏览器操作、Chrome、演示、デモ、实时浏览器。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: automating-browser description: > Browser automation using agent-browser CLI. Best for E2E testing, demos, screenshots. Use when automating browser interactions, creating E2E tests, or when user mentions browser automation, E2E test, ブラウザ操作, スクリーンショット, demo, manual testing. allowed-tools: [Read, Glob, Bash(agent-browser:*)] context: fork user-invocable: false

Browser Automation (agent-browser)

Core Commands

Command Purpose
agent-browser --headed open <url> Open URL (visible browser)
agent-browser open <url> Open URL (headless)
agent-browser snapshot -i Get interactive elements (refs)
agent-browser click @ref Click element
agent-browser fill @ref "text" Clear and fill
agent-browser type @ref "text" Type into element
agent-browser press <key> Press key (Enter, Tab, etc.)
agent-browser get text @ref Read element text
agent-browser wait <sel|ms> Wait for element or time
agent-browser find <loc> <val> <a> Find by role/text/label + act
agent-browser screenshot [path] Capture screenshot
agent-browser close Close browser session

Debugging Commands

Command Purpose
agent-browser console Display console output
agent-browser errors Show JS error stack traces
agent-browser network requests List captured requests
agent-browser network requests --filter api Filter by keyword
agent-browser trace start <path> Begin recording trace
agent-browser trace stop Save trace file

Workflow

Step Action
1 agent-browser --headed open <url>
2 agent-browser snapshot -i
3 Use @ref for operations
4 Re-snapshot after DOM changes
5 Check console/errors regularly

Key Points

Point Description
Always snapshot Refs are only valid after snapshot
Re-snapshot on DOM After click/fill, get new refs
Check DevTools Run console/errors proactively, not on fail
Mode switch Close first when switching headed ↔ headless

Patterns

Pattern Commands
Form Filling snapshot -i → fill @ref "value" → click submit
Navigation click @ref → wait → snapshot
Screenshot snapshot → screenshot path.png
Debug action → console → errors → network requests