paper-analysis

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted data from external research papers and URLs, which creates an inherent surface for indirect prompt injection.
  • Ingestion points: The skill fetches content from URLs via web_fetch, arXiv/DOI repositories, and processes user-provided PDF files.
  • Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions when processing fetched paper content.
  • Capability inventory: The skill is limited to text analysis, citation extraction, and summarization; it lacks capabilities for shell execution, file system modification, or non-whitelisted network-write operations.
  • Sanitization: No specific input validation or sanitization mechanisms are defined for the external text processed by the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 11:31 PM
Security Audit — agent-trust-hub — paper-analysis