Browser Automation Toolbox

Problem Background

Browser automation in data collection and testing frequently encounters challenges such as anti-scraping mechanisms, login session maintenance, and SPA content extraction. Specific issues include:
- Anti-detection: Tools like Playwright being blocked by风控 systems, exposing automation fingerprints.
- Login session loss: Sessions expiring when scraping login-required platforms, leading to incomplete data.
- Incomplete SPA scraping: Dynamic content in single-page applications hard to capture via static parsing.
- High maintenance costs: Frequent selector changes across platforms require manual updates.
- Special handling for AI platforms: Stricter login and anti-scraping measures on platforms like Doubao, GPT, and Gemini.

Core Capabilities and Workflow

Browser Automation Toolbox addresses these issues through an intelligent fallback strategy, with the following core workflow:

Engine Auto-Fallback

The toolbox includes four browser engines, switching automatically by priority to ensure task success:
1. CloakBrowser: Local anti-detection browser, suitable for real websites and sensitive pages.
2. browser-act: Cloud browser pool, providing session management and human collaboration.
3. Kimi WebBridge: Assists via browser plugins in specific failure scenarios.
4. Playwright: Acts as a fallback for standard tests and simple pages.

For example, during execution, the tool first attempts CloakBrowser; if it fails, it falls back to browser-act, and so on. Default max_attempts_per_engine is 2, increased to 3 only for transient failures.

Platform-Aware Priority

For different platforms, the tool dynamically adjusts engine order to optimize success rates:
- Xiaohongshu: Prioritizes browser-act as its cloud pool more effectively bypasses anti-scraping.
- AI platforms (Gemini, Doubao, GPT): Prioritizes CloakBrowser to avoid cloud browsers losing local login cookies.
- Bilibili, Douyin, Weibo: Default CloakBrowser priority, leveraging persistent profiles.

Users can specify platforms via the --platform parameter, or the tool auto-detects domains. User preferences (e.g., favoring browser-act) can be recorded and override default orders.

Key Step Support

• Login and CAPTCHA handling: When encountering login walls, pauses the engine and prompts users to complete manually, then resumes or switches engines. Public scraping defaults to visible browser mode.
• SPA infinite scrolling: For Bilibili, Xiaohongshu, and Douyin, uses scroll-based collection; Weibo real-time search prioritizes URL pagination parameters.
• Experience accumulation: Changes in selectors or strategy adjustments are recorded to references/platform-scraping-patterns.md for reuse.

Applicable Scenarios and Considerations

Suitable Scenarios

• Web scraping and data collection, especially public sentiment monitoring.
• Automated form filling and screenshot evidence collection.
• AI platform automation, such as content generation and chain-of-thought detection.
• Evaluating and enhancing other browser automation skills.

Considerations

• Engine selection: Default priorities may vary by platform; adjust based on target sites. Users can always force an order via --engine-order ....
• Login session maintenance: For login-required sites, prefer CloakBrowser's persistent profiles to avoid cookie loss.
• Failure handling: The tool saves error logs and screenshots for debugging, but only switches engines after multiple failures in the same one to avoid endless retries.
• Complementary tools: For tasks needing only user-logged-in browsers without anti-scraping, recommend BrowserSkill (Tencent) over this toolbox's fallback chain.
• Skill evaluation: The toolbox can assess other skills' browser automation capabilities, providing gap reports and integration suggestions, but requires user activation.