figure-verifier

Pass

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external scientific sources, which provides a surface for potential indirect prompt injection attacks.
  • Ingestion points: The agent fetches publication abstracts via the CrossRef API and retrieves HTML or figure images from arXiv and PubMed Central (SKILL.md).
  • Boundary markers: The instructions do not define boundary markers or explicit safety guidelines to disregard instructions that might be embedded in the scientific text or metadata being processed.
  • Capability inventory: The skill leverages network access and visual analysis (Read tool) to process external data.
  • Sanitization: No sanitization, validation, or filtering of the retrieved research content is specified in the skill's logic.
  • [EXTERNAL_DOWNLOADS]: The skill fetches data from well-known and reputable scientific platforms.
  • Downloads metadata and publication abstracts from the CrossRef API.
  • Retrieves figure images and article HTML from arXiv and PubMed Central repositories.
  • [DATA_EXFILTRATION]: The skill handles user contact information to interact with external services.
  • Requests and utilizes the user's email address to satisfy the requirements of the Unpaywall API (api.unpaywall.org), which is an established prerequisite for accessing that service.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 10, 2026, 10:15 PM
Security Audit — agent-trust-hub — figure-verifier