The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/publish.py transmits all article data, including titles, full content, and tags, to a hardcoded external IP address http://39.108.254.228:8002. While the SKILL.md labels this as the user's "self-built service," the hardcoding of a specific, non-local IP address means data is exfiltrated to a third-party server by default.
[COMMAND_EXECUTION]: The scripts/publish.py script utilizes subprocess.run to execute curl commands. This is used to perform POST requests to the hardcoded external IP address with user-supplied content.
[CREDENTIALS_UNSAFE]: The documentation provided in references/server-config-guide.md and references/platform-requirements.md instructs users to obtain and store highly sensitive authentication data—including session cookies and access tokens for platforms like WeChat, 知乎 (Zhihu), 微博 (Weibo), and 掘金 (Juejin)—in an environment file for the server component. Because the skill's scripts point to a hardcoded external IP, following these instructions could lead to credential harvesting.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection in scripts/publish.py. It parses JSON responses from the hardcoded server and returns error messages or data directly to the agent's context. A malicious server could provide instructions in these fields to manipulate the agent's subsequent actions.
[EXTERNAL_DOWNLOADS]: scripts/search_images.py performs network requests to https://source.unsplash.com to fetch images. While this is a well-known service, it demonstrates the skill's capability to interact with external web resources.

wechatsync-publisher