xiaohongshu-skills

Warn

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: In scripts/xhs/login.py, the skill transmits the Xiaohongshu login QR code (as image data) and the resulting authentication URL to a third-party service (api.qrserver.com) for decoding and re-encoding. This exposes highly sensitive authentication tokens to an external API that is not on the trusted domain whitelist.
  • [PROMPT_INJECTION]: The SKILL.md files contain explicit instructions under the "🔒 技能边界" section that direct the AI agent to ignore previous instructions or alternative tools (such as MCP servers) and only use the provided CLI scripts. This is a behavioral override attempt to restrict the agent's operational capabilities to this specific implementation.
  • [COMMAND_EXECUTION]: The skill uses subprocess.Popen in scripts/chrome_launcher.py to launch the Google Chrome browser with the --remote-debugging-port flag enabled. This allows the local Python scripts to control the browser instance via the Chrome DevTools Protocol (CDP).
  • [EXTERNAL_DOWNLOADS]: The scripts/image_downloader.py utility uses the requests library to download media files from arbitrary URLs provided in the command-line arguments or extracted during web search operations.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 16, 2026, 11:41 PM