The Agent Skills Directory

[COMMAND_EXECUTION]: The script scrape_xhs.sh is vulnerable to command injection. The user-supplied -k (keyword) and -d (detail count) arguments are interpolated directly into shell-executed python3 -c strings, allowing for arbitrary code execution if they contain malicious characters like single quotes.\n- [DYNAMIC_EXECUTION]: The skill performs dynamic JavaScript execution using the agent-browser tool's eval function to scrape data from the web, which constitutes a runtime code execution pattern.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Xiaohongshu that could contain malicious instructions.\n
Ingestion points: Note titles, author names, and bodies are scraped from xiaohongshu.com in scrape_xhs.sh.\n
Boundary markers: The output does not include delimiters or warnings to separate scraped content from agent instructions.\n
Capability inventory: The skill has the ability to write files to the local disk and interact with a browser.\n
Sanitization: Content is not sanitized before being output in Markdown format, allowing embedded instructions to persist in the agent's context.

xhs-scraper