The Agent Skills Directory

[CREDENTIALS_UNSAFE]: A hardcoded Weibo session cookie ('SUB') is present in the headers of the fetch_weibo function within the Python collection script.
Evidence: File scripts/content/fetch_news.py contains "Cookie": "SUB=_2AkMWIuNSf8NxqwJRmP8dy2rhaoV2ygrEieKgfhKJJRMxHRl-yT9jqk86tRB6PaLNvQZR6zYUcYVT1zSjoSreQHidcUq7".
[COMMAND_EXECUTION]: The skill uses Node.js subprocess methods to execute local Python scripts and external skill scripts, which could be exploited if pathing or arguments are manipulated.
Evidence: scripts/content/smart-collect.mjs and scripts/content/collect-news.mjs use spawn and execSync to run python3 scripts.
[EXTERNAL_DOWNLOADS]: The collection module fetches data from over 20 external domains including social media and news aggregators, which is functional but increases the risk of ingesting malicious content.
Evidence: scripts/content/fetch_news.py targets domains like s.weibo.com, www.zhihu.com, api.juejin.cn, and www.toutiao.com.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it aggregates untrusted external data and interpolates it directly into instructions for the AI to generate articles.
Ingestion points: scripts/content/smart-collect.mjs and scripts/content/fetch_news.py (fetches news titles and content from 20+ sources).
Boundary markers: Absent. The article generation prompt in scripts/content/generate.mjs lacks delimiters to separate untrusted content from instructions.
Capability inventory: scripts/content/publish.mjs (capability to create and publish drafts to the WeChat Official Account platform via the wemp skill).
Sanitization: Absent. The external content is used directly in the prompt generation logic in scripts/content/generate.mjs.

wemp-operator