The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external URLs, creating a surface for potential prompt injection attacks.
Ingestion points: The fetch_wx_article function in scripts/fetch_wx_article.py fetches content from user-provided WeChat URLs (mp.weixin.qq.com).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish the fetched content from system instructions, which could allow malicious text within an article to influence the agent's behavior.
Capability inventory: The skill performs network reads using the scrapling library and outputs the converted content to the console.
Sanitization: The script uses html2text to convert HTML to Markdown and is configured to ignore links and images, which serves as a basic filter but does not sanitize potential natural language instructions.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs network operations to fetch content from external domains.
Evidence: The script uses Fetcher.get(url) to download content from the mp.weixin.qq.com domain, which is a well-known service for official WeChat articles and is consistent with the skill's primary purpose.

fetch-wx-article