wechat-article-search
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data from third-party WeChat articles which could contain text instructions designed to manipulate downstream agent behavior.
- Ingestion points: HTML contents fetched from
https://weixin.sogou.com/weixininsidescripts/search_wechat.jsare parsed into plaintext article objects. - Boundary markers: No isolation wrappers or prompt formatting boundaries are specified to isolate the content fields before passing them back into the LLM context.
- Capability inventory: The skill possesses file writing capabilities via
fs.writeFileSync(when using the-oparameter) but does not execute external system processes or code evaluation. - Sanitization: Parsed fields such as
titleandsummaryare stripped of extra spaces using.trim(), but no content filtering or safety-level sanitization is performed on the parsed text content.
Audit Metadata