技能介绍
先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。
做什么
Chrome DevTools 协议 自动连接到现有Chrome的CLI,用于浏览器自动化而无需重量级运行时间管理
何时用
官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。
代理如何加载
按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Chrome DevTools CLI、Why Use This、Installation、macOS (recommended)、Or via Cargo、Prerequisites。 其中含规范建议的小节:分步指令。
文件分析
文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。
官方 description(原文)
Chrome DevTools Protocol CLI that auto-connects to existing Chrome for browser automation without heavyweight runtime overhead
Chrome DevTools CLI Why Use This Installation macOS (recommended) Or via Cargo Prerequisites Auto-Connect Behavior Environment variables Or flags Daemon Architecture Page Targeting Pattern Step 1: Navigate and capture target name
来源分类:skills.sh agent-skill
来源与网址
namechrome-devtools-cli
descriptionChrome DevTools Protocol CLI that auto-connects to existing Chrome for browser automation without heavyweight runtime overhead
发现技能 客户端向 Agent 提供名称与描述目录。
匹配与调用 用户指定或任务匹配后,载入 SKILL.md 指令。
按需加载 按步骤读取参考文档、使用脚本与素材。
具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗
先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。
交给 Agent 安装
复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。
把 Agent Skill「chrome-devtools-cli」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-603da7bb7aac5fde-Chrome-Devtools-Cli-%E8%8B%B1%E8%AF%AD.html
请存为 .cursor/skills/chrome-devtools-cli/SKILL.md 或 .claude/skills/chrome-devtools-cli/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
复制安装提示词 GitHub 完整包 ↗
终端安装 · Skills CLI
需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。
npx skills add 'https://github.com/reason-machines/devtools-skills' --list
复制查看命令
目标客户端
安装时交互选择 Codex Claude Code Cursor
安装范围
当前项目 我的所有项目
npx skills add 'https://github.com/reason-machines/devtools-skills' --skill 'chrome-devtools-cli' 复制安装命令
CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。
SKILL.md 原文 可复制,不会改动源文件
复制 SKILL.md
恢复原文
阅读排版
---
name: chrome-devtools-cli
description: Chrome DevTools Protocol CLI that auto-connects to existing Chrome for browser automation without heavyweight runtime overhead
triggers:
- "automate Chrome browser"
- "control Chrome DevTools"
- "navigate browser and take screenshot"
- "interact with web page using Chrome"
- "click element in Chrome"
- "run JavaScript in browser"
- "inspect page accessibility tree"
- "fill form in Chrome browser"
---
# Chrome DevTools CLI
> Skill by [ara.so](https://ara.so) — Devtools Skills collection.
Chrome DevTools CLI is a lightweight Rust binary that connects directly to an existing Chrome browser via the DevTools Protocol. It auto-connects to your Chrome instance, requires no separate browser process, and provides a minimal-context alternative to MCP-based browser automation.
## Why Use This
- **Zero context overhead**: One command in, one result out. No MCP layer, no large protocol payloads
- **Uses your Chrome**: Works with your existing browser session, credentials, and extensions
- **Persistent connection**: Background daemon keeps WebSocket open, browser prompts for access only once
- **Deterministic targeting**: Friendly target names (`[target:red-snake]`) for reliable multi-page workflows
## Installation
```bash
# macOS (recommended)
brew install aeroxy/tap/chrome-devtools
# Or via Cargo
cargo install chrome-devtools-cli
```
The binary is named `chrome-devtools`.
## Prerequisites
Enable Chrome remote debugging:
1. Open Chrome
2. Navigate to `chrome://inspect/#remote-debugging`
3. Enable the remote debugging server
## Auto-Connect Behavior
The CLI automatically discovers Chrome's WebSocket endpoint by reading `DevToolsActivePort` from:
- macOS: `~/Library/Application Support/Google/Chrome/`
- Linux: `~/.config/google-chrome/`
- Windows: `%LOCALAPPDATA%\Google\Chrome\User Data\`
Override with environment variables or flags:
```bash
# Environment variables
export CHROME_WS_ENDPOINT=ws://localhost:9222/devtools/browser/abc123
export CHROME_USER_DATA_DIR=~/custom-chrome-profile
export CHROME_CHANNEL=beta # stable, beta, canary, dev
# Or flags
chrome-devtools --ws-endpoint ws://localhost:9222/... navigate https://example.com
chrome-devtools --user-data-dir ~/custom-profile navigate https://example.com
chrome-devtools --channel canary navigate https://example.com
```
## Daemon Architecture
First command spawns a background daemon that:
- Connects to Chrome WebSocket (one-time approval prompt)
- Listens on Unix socket: `/tmp/chrome-devtools-daemon.sock`
- Auto-exits after 5 minutes of inactivity
- All subsequent commands reuse the persistent connection
Kill manually: `pkill -f __daemon__` or delete the socket file.
## Page Targeting Pattern
**CRITICAL**: Always capture and reuse the `[target:name]` from navigation commands.
```bash
# Step 1: Navigate and capture target name
chrome-devtools navigate https://github.com
# Output: Navigated to https://github.com
# [target:green-dog]
# Step 2: Pin all subsequent commands to this target
chrome-devtools --target green-dog screenshot --output /tmp/gh.png
chrome-devtools --target green-dog evaluate "document.title"
chrome-devtools --target green-dog click "a[href='/login']"
```
Without `--target`, commands default to page index 0, which may change as Chrome reorders tabs.
### List All Pages
```bash
chrome-devtools list-pages
# Output:
# [0] (green-dog) GitHub — https://github.com
# [1] (red-snake) Example Domain — https://example.com
# [2] (bold-stag) Local Dev — https://localhost:3000
```
You can also use `--page ` for quick one-offs:
```bash
chrome-devtools --page 1 evaluate "window.location.href"
```
## Core Commands
### Navigation
```bash
# Navigate to URL (waits for load event)
chrome-devtools navigate https://example.com
# History navigation
chrome-devtools --target red-snake navigate --back
chrome-devtools --target red-snake navigate --forward
chrome-devtools --target red-snake navigate --reload
# Tab management
chrome-devtools new-page https://github.com
chrome-devtools close-page 2
chrome-devtools select-page 0 # Bring tab to front
```
### Inspection
```bash
# Screenshot (viewport only)
chrome-devtools --target green-dog screenshot --output /tmp/page.png
# Full-page screenshot (scrollable area)
chrome-devtools --target green-dog screenshot --full-page --output /tmp/full.png
# JavaScript evaluation
chrome-devtools --target green-dog evaluate "document.title"
chrome-devtools --target green-dog evaluate "Array.from(document.querySelectorAll('h1')).map(h => h.textContent)"
# Handle JavaScript dialogs (alert/confirm/prompt)
chrome-devtools --target green-dog evaluate "confirm('Proceed?')" --dialog-action accept
chrome-devtools --target green-dog evaluate "prompt('Name?')" --dialog-action "John Doe"
# Accessibility tree snapshot
chrome-devtools --target green-dog snapshot
```
### Interaction
```bash
# Click by CSS selector
chrome-devtools --target green-dog click "#submit-button"
chrome-devtools --target green-dog click "a[href='/login']"
# Click at coordinates
chrome-devtools --target green-dog click-at 100 200
# Fill input fields
chrome-devtools --target green-dog fill "#email" "
[email protected] "
chrome-devtools --target green-dog fill "#password" "secret123"
# Fill dropdown (select element)
chrome-devtools --target green-dog fill "#country" "United States"
# Toggle checkbox/radio
chrome-devtools --target green-dog fill "#terms" "true"
chrome-devtools --target green-dog fill "#newsletter" "false"
# Type text into focused element
chrome-devtools --target green-dog type-text "Hello World"
chrome-devtools --target green-dog type-text "Search query" --submit-key Enter
# Press keys
chrome-devtools --target green-dog press-key Enter
chrome-devtools --target green-dog press-key "Control+A"
chrome-devtools --target green-dog press-key Escape
# Hover over element
chrome-devtools --target green-dog hover ".dropdown-menu"
```
### Viewport & Waiting
```bash
# Resize viewport
chrome-devtools --target green-dog resize 1920 1080
# Wait for text to appear (default 30s timeout)
chrome-devtools --target green-dog wait-for "Login successful"
chrome-devtools --target green-dog wait-for "Dashboard" --timeout 60000
```
### Third-Party Developer Tools
For pages that inject custom DevTools via `window.__dtmcp`:
```bash
# List available custom tools
chrome-devtools --target green-dog list-3p-tools
# Execute custom tool
chrome-devtools --target green-dog execute-3p-tool "myTool" '{"param": "value"}'
```
## JSON Output Mode
Add `--json` to any command for machine-readable output:
```bash
chrome-devtools --json --target green-dog evaluate "document.title"
# {"success": true, "result": "Example Domain"}
chrome-devtools --json list-pages
# {"success": true, "pages": [{"index": 0, "friendlyName": "green-dog", ...}]}
```
## Real-World Workflows
### Automated Form Submission
```bash
#!/bin/bash
set -e
# Navigate and capture target
TARGET=$(chrome-devtools navigate https://example.com/form | grep -oP '(?<=target:)[a-z-]+')
# Fill form fields
chrome-devtools --target "$TARGET" fill "#name" "Alice Smith"
chrome-devtools --target "$TARGET" fill "#email" "
[email protected] "
chrome-devtools --target "$TARGET" fill "#country" "Canada"
chrome-devtools --target "$TARGET" fill "#terms" "true"
# Submit and wait for confirmation
chrome-devtools --target "$TARGET" click "#submit"
chrome-devtools --target "$TARGET" wait-for "Thank you" --timeout 10000
# Capture confirmation screenshot
chrome-devtools --target "$TARGET" screenshot --output /tmp/confirmation.png
```
### Data Extraction
```bash
#!/bin/bash
TARGET=$(chrome-devtools navigate https://news.ycombinator.com | grep -oP '(?<=target:)[a-z-]+')
# Extract top stories
chrome-devtools --target "$TARGET" evaluate "
Array.from(document.querySelectorAll('.athing')).slice(0, 5).map(item => ({
title: item.querySelector('.titleline > a').textContent,
url: item.querySelector('.titleline > a').href
}))
" --json > /tmp/hn-stories.json
cat /tmp/hn-stories.json
```
### Multi-Page Session
```bash
#!/bin/bash
# Open multiple pages
TARGET_HOME=$(chrome-devtools navigate https://github.com | grep -oP '(?<=target:)[a-z-]+')
TARGET_REPO=$(chrome-devtools new-page https://github.com/torvalds/linux | grep -oP '(?<=target:)[a-z-]+')
# Interact with different pages
chrome-devtools --target "$TARGET_HOME" click "a[href='/login']"
chrome-devtools --target "$TARGET_REPO" evaluate "document.querySelector('.repository-content').textContent.includes('Linux kernel')"
# Screenshot both
chrome-devtools --target "$TARGET_HOME" screenshot --output /tmp/home.png
chrome-devtools --target "$TARGET_REPO" screenshot --output /tmp/repo.png
```
### Accessibility Audit
```bash
#!/bin/bash
TARGET=$(chrome-devtools navigate https://example.com | grep -oP '(?<=target:)[a-z-]+')
# Get full accessibility tree
chrome-devtools --target "$TARGET" snapshot > /tmp/a11y-tree.txt
# Check for specific roles
chrome-devtools --target "$TARGET" evaluate "
document.querySelectorAll('[role=\"button\"]').length
"
# Find missing alt text
chrome-devtools --target "$TARGET" evaluate "
Array.from(document.querySelectorAll('img:not([alt])')).map(img => img.src)
"
```
## Common Patterns for AI Agents
### 1. Always Capture Target Names
```bash
# ❌ BAD: Target may change
chrome-devtools navigate https://example.com
chrome-devtools screenshot --output /tmp/page.png # Might screenshot wrong page!
# ✅ GOOD: Explicit target
TARGET=$(chrome-devtools navigate https://example.com | grep -oP '(?<=target:)[a-z-]+')
chrome-devtools --target "$TARGET" screenshot --output /tmp/page.png
```
### 2. Understand the Page First
```bash
# Get context before interacting
chrome-devtools --target "$TARGET" snapshot # See accessible structure
chrome-devtools --target "$TARGET" screenshot --output /tmp/context.png
chrome-devtools --target "$TARGET" evaluate "document.body.innerText" # Read visible text
```
### 3. Wait for Dynamic Content
```bash
# Don't assume instant rendering
chrome-devtools --target "$TARGET" click "#load-more"
chrome-devtools --target "$TARGET" wait-for "Showing 20 results" --timeout 5000
chrome-devtools --target "$TARGET" evaluate "document.querySelectorAll('.result-item').length"
```
### 4. Handle Errors Gracefully
```bash
# Check page state before interaction
HAS_BUTTON=$(chrome-devtools --target "$TARGET" evaluate "!!document.querySelector('#submit')" --json | jq -r '.result')
if [ "$HAS_BUTTON" = "true" ]; then
chrome-devtools --target "$TARGET" click "#submit"
else
echo "Submit button not found"
fi
```
## Troubleshooting
### "Could not connect to Chrome"
- Ensure Chrome is running
- Verify remote debugging is enabled at `chrome://inspect/#remote-debugging`
- Check `DevToolsActivePort` exists in user data directory
- Try explicit WebSocket: `chrome-devtools --ws-endpoint ws://localhost:9222/... navigate https://example.com`
### "Daemon connection failed"
```bash
# Kill stale daemon
pkill -f __daemon__
rm /tmp/chrome-devtools-daemon.sock
# Retry command (will spawn fresh daemon)
chrome-devtools navigate https://example.com
```
### "Element not found" on click/fill
```bash
# Verify element exists
chrome-devtools --target "$TARGET" evaluate "!!document.querySelector('#my-button')"
# Get all matching elements
chrome-devtools --target "$TARGET" evaluate "document.querySelectorAll('#my-button').length"
# Wait for element to appear
chrome-devtools --target "$TARGET" wait-for "Submit" --timeout 10000
chrome-devtools --target "$TARGET" click "#submit"
```
### Target name confusion
```bash
# List all pages with friendly names
chrome-devtools list-pages
# Use explicit page index if needed
chrome-devtools --page 0 evaluate "document.title"
# Or use raw target ID (from list-pages output)
chrome-devtools --target "E4F5A6B7C8D9" evaluate "document.title"
```
## Best Practices for AI Agents
1. **Always use `--target`** after initial navigation
2. **Capture screenshots** before and after interactions for debugging
3. **Use `snapshot`** to understand page structure before clicking
4. **Wait for content** instead of assuming instant load
5. **Verify elements exist** before interaction
6. **Use JSON output** (`--json`) for programmatic parsing
7. **Check `list-pages`** when multi-page workflows get confusing
8. **Kill daemon** between unrelated sessions to avoid stale connections
## Integration with AI Coding Agents
When helping a user automate Chrome:
1. Install `chrome-devtools-cli` if not present
2. Verify Chrome remote debugging is enabled
3. Start with `navigate` and capture the target name
4. Use `snapshot` or `screenshot` to understand the page
5. Perform interactions with explicit `--target`
6. Extract results with `evaluate` and `--json`
7. Clean up with `close-page` if needed
Example agent workflow:
```bash
# User: "Login to example.com with my credentials"
# 1. Navigate
TARGET=$(chrome-devtools navigate https://example.com/login | grep -oP '(?<=target:)[a-z-]+')
# 2. Understand page
chrome-devtools --target "$TARGET" snapshot | head -50
# 3. Fill credentials (from env vars)
chrome-devtools --target "$TARGET" fill "#username" "$EXAMPLE_USERNAME"
chrome-devtools --target "$TARGET" fill "#password" "$EXAMPLE_PASSWORD"
# 4. Submit
chrome-devtools --target "$TARGET" click "#login-button"
# 5. Verify success
chrome-devtools --target "$TARGET" wait-for "Dashboard" --timeout 10000
chrome-devtools --target "$TARGET" screenshot --output /tmp/logged-in.png
```
This skill enables zero-context browser automation directly from your existing Chrome session.
相关技能
Agent 工作流
创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.
Agent 工作流
使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.
Agent 工作流
管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.
Agent 工作流
Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…