跳到正文
cccskills
免费在 GitHub 公开

go-playwright

Expert capability for robust, stealthy, and efficient browser automation using Playwright Go.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(2)

  • SKILL.md4.8 KB
  • resources/implementation-playbook.md3.1 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Playwright Go Automation Expert

Overview

This skill provides a comprehensive framework for writing high-performance, production-grade browser automation scripts using github.com/playwright-community/playwright-go. It enforces architectural best practices (contexts over instances), robust error handling, structured logging (Zap), and advanced human-emulation techniques to bypass anti-bot systems.

When to Use This Skill

  • Use when the user asks to "scrape," "automate," or "test" a website using Go.
  • Use when the target site has complex dynamic content (SPA, React, Vue) requiring a real browser.
  • Use when the user mentions "stealth," "avoiding detection," "cloudflare," or "human-like" behavior.
  • Use when debugging existing Playwright scripts.

Safety & Risk

Risk Level: 🔵 Safe

  • Sandboxed Execution: Browser contexts are isolated; they do not persist data to the host machine unless explicitly saved.
  • Resource Management: Designed to close browsers and contexts via defer to prevent memory leaks.
  • No External State-Change: Default behavior is read-only (scraping/testing) unless the script is explicitly designed to submit forms or modify data.

Limitations

  • Environment Dependencies: Requires Playwright drivers and browsers to be installed (go run github.com/playwright-community/playwright-go/cmd/playwright@latest install --with-deps).
  • Resource Intensity: Launching full browser instances (even headless) consumes significant RAM/CPU. Use single-browser/multi-context architecture.
  • Bot Detection: While this skill includes stealth techniques, extremely strict anti-bot systems (e.g., rigorous Cloudflare settings) may still detect automation.
  • CAPTCHAs: Does not include built-in CAPTCHA solving capabilities.

Strategic Implementation Guidelines

1. Architecture: Contexts vs. Browsers

CRITICAL: Never launch a new Browser instance for every task.

  • Pattern: Launch the Browser once (singleton). Create a new BrowserContext for each distinct session or task.
  • Why: Contexts are lightweight and created in milliseconds. Browsers take seconds to launch.
  • Isolation: Contexts provide complete isolation (cookies, cache, storage) without the overhead of a new process.

2. Logging & Observability

  • Library: Use go.uber.org/zap exclusively.
  • Rule: Do not use fmt.Println.
  • Modes:
    • Dev: zap.NewDevelopment() (Console friendly)
    • Prod: zap.NewProduction() (JSON structured)
  • Traceability: Log every navigation, click, and input with context fields (e.g., logger.Info("clicking button", zap.String("selector", sel))).

3. Error Handling & Stability

  • Graceful Shutdown: Always use defer to close Pages, Contexts, and Browsers.
  • Panic Recovery: Wrap critical automation routines in a safe runner that recovers panics and logs the stack trace.
  • Timeouts: Never rely on default timeouts. Set explicit timeouts (e.g., playwright.PageClickOptions{Timeout: playwright.Float(5000)}).

4. Stealth & Human-Like Behavior

To bypass anti-bot systems (Cloudflare, Akamai), the generated code must imitate human physiology:

  • Non-Linear Mouse Movement: Never teleport the mouse. Implement a helper that moves the mouse along a Bezier curve with random jitter.
  • Input Latency: never use Fill(). Use Type() with random delays between keystrokes (50ms–200ms).
  • Viewport Randomization: Randomize the viewport size slightly (e.g., 1920x1080 ± 15px) to avoid fingerprinting.
  • Behavioral Noise: Randomly scroll, focus/unfocus the window, or hover over irrelevant elements ("idling") during long waits.
  • User-Agent: Rotate User-Agents for every new Context.

5. Documentation Usage

  • Primary Source: Rely on your internal knowledge of the API first to save tokens.
  • Fallback: Refer to the official docs playwright-go documentation ONLY if:
    • You encounter an unknown error.
    • You need to implement complex network interception or authentication flows.
    • The API has changed significantly.

Resources

  • resources/implementation-playbook.md for detailed code examples and implementation patterns.

Summary Checklist for Agent

  • Is Debug Mode on? -> Headless=false, SlowMo=100+.
  • Is it a new user identity? -> NewContext, apply new Proxy, rotate User-Agent.
  • Is the action critical? -> Wrap in SafeAction with Zap logging.
  • Is the target guarded (Cloudflare/Akamai)? -> Enable HumanType, BezierMouse, and Stealth Scripts.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Use when manage multiple local CLI agents via tmux sessions (start/stop/monitor/assign) with cron-friendly scheduling.

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

Use when a hybrid memory system that provides persistent, searchable knowledge management for AI agents (Architecture, Patterns, Decisions).

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

Reference for how BDB structures autonomous software engineering work — the seven-node dispatcher graph (Architect, TechLead, UI/UX, Engineering, Media/EventTech, Reviewer, Shipping) that /startcycle-graph actually runs. Use when you need the high-level lifecycle framing without inventing your own process.

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

Tools are how AI agents interact with the world. A well-designed tool is the difference between an agent that works and one that hallucinates, fails silently, or costs 10x more tokens than necessary. This skill covers tool design from schema to error handling.

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

Harness patterns for coding agents — memory, permissions, context engineering, delegation, skills, hooks, bootstrap.

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

agenttrail: live map of a multi-agent build in the browser: which plan component is being worked on, by which agent or harness, what is done and what is stuck. Use when a multi-agent pipeline starts (/startcycle, /startcycle-graph, /teamwork-preview) or after a plan-canvas approve, when the user asks to see what the agents are doing ("live map", "build board", "who is running now"), or by yourself whenever a multi-agent run is under way — no user prompt needed.

中文简介正在准备中。当前显示原文说明。

hybridlabor-api/aos62026年10月16日 更新

查看 hybridlabor-api 的全部技能

报告此技能的问题