Safe Web Fetch for Save Token

安全的智能网页抓取技能，节省 50-80% Token。替代内置 web_fetch，自动使用 Jina Reader 清洗服务获取干净 Markdown。内置 URL 白名单验证、SSL 强制验证、敏感数据检测，防止 SSRF 和数据泄露。

This skill is mostly coherent but has a critical implementation vs. documentation mismatch: it will call the external Jina Reader (r.jina.ai) and send the original URL for server-side fetching and cleaning without running the script's sensitive-data checks first. That can leak API keys, tokens, or other secrets present on the page to an external service. Before installing or using it: - If you cannot tolerate sending pages (or their URLs) to an external service, do not use the 'Jina Reader' path; require the skill to fetch locally first and run sensitive-data detection before any external call. - Inspect or modify scripts/safe_fetch.py to enable detect_sensitive_data on the original page before calling Jina, or to skip Jina entirely and perform local cleaning. - Restrict allowed_domains in config.json to a small explicit allowlist if you must use this skill in a sensitive environment. - Be aware that the Jina request uses a path of the form https://r.jina.ai/http://{original}, which may not preserve the original scheme and could cause unexpected fetch behavior. If you make the code change so the script checks for sensitive patterns prior to any external request (or removes the Jina step entirely), the concerns above would be resolved and the skill would be coherent with its stated security guarantees.

Type: OpenClaw Skill Name: safe-web-fetch-for-save-token Version: 1.0.0 The skill bundle provides a utility for fetching and cleaning web content using the Jina Reader service or direct requests. Analysis of `scripts/safe_fetch.py` confirms the implementation of robust security features, including SSRF protection via IP resolution checks, mandatory SSL verification, and regex-based sensitive data detection (e.g., API keys, private keys) to prevent data leakage. The code uses standard Python libraries and lacks any indicators of malicious intent, obfuscation, or unauthorized data exfiltration.