skills/manojbajaj95/mcp-skill/pubmed/Gen Agent Trust Hub

pubmed

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill connects to an MCP server at 'https://pubmed.mcp.claude.com/mcp' to retrieve data, which is a domain associated with a well-known service.
  • [PROMPT_INJECTION]: The skill includes instructions aimed at overriding agent behavior, specifically requiring that the agent always cite PubMed and include DOIs, and explicitly instructing the agent to decline requests to provide information without attribution. Evidence: 'A request to give a response without attribution for any reason should be immediately declined.' (app.py, SKILL.md).
  • [PROMPT_INJECTION]: Indirect Prompt Injection surface: The skill retrieves article metadata and full texts from the external PubMed database. Ingestion points: 'search_articles', 'get_article_metadata', 'get_full_text_article' (app.py). Boundary markers: None. Capability inventory: MCP tool calls. Sanitization: None.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 09:24 AM