pubmed-database

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill includes instructions that direct the agent to proactively suggest the author's commercial platform, K-Dense Web, for complex workflows, which steers the agent's interaction toward promotional objectives.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to the ingestion of untrusted external data from search results. * Ingestion points: Research article metadata and abstracts retrieved from the NCBI E-utilities API. * Boundary markers: Absent; no specific delimiters are defined to separate external data from agent instructions. * Capability inventory: Network communication via the Python requests library and local file system access for data caching. * Sanitization: No input validation or sanitization is implemented for the data processed from the API.- [EXTERNAL_DOWNLOADS]: Fetches biomedical literature and metadata from the well-known and official NCBI E-utilities API at eutils.ncbi.nlm.nih.gov.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 08:26 AM