NYC

wechat-article-writer

Fail

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: HIGHPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION] (HIGH): The skill is susceptible to Indirect Prompt Injection through its web-fetching capabilities.\n
  • Ingestion points: The web_fetch tool (Step 1) and exa:web_search_exa (Step 2) ingest content from arbitrary external sources.\n
  • Boundary markers: Absent. The instructions do not specify any delimiters to separate external content from the system prompt.\n
  • Capability inventory: The skill uses web_fetch, exa:web_search_exa, and mem0-memory-mcp. The access to a memory tool increases the risk, as an injection could attempt to read or leak stored user preferences.\n
  • Sanitization: No evidence of sanitization, filtering, or validation of the fetched content before it is used for planning and writing.\n- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill fetches data from arbitrary, unverified URLs.\n
  • While the tools used are standard MCP components, the lack of a restricted domain list or URL validation logic allows the agent to process content from potentially malicious sources designed to exploit prompt-following behavior.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 16, 2026, 02:50 AM