physics-literature-search

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from scientific repositories like arXiv, INSPIRE-HEP, NASA ADS, and Web of Science.
  • Ingestion points: The skill retrieves paper titles, author lists, and abstracts from Daily RSS feeds and APIs (defined in Step 2 of the Process section in SKILL.md).
  • Boundary markers: There are no explicit instructions or delimiters used to ensure the agent ignores malicious instructions that could be hidden within paper abstracts.
  • Capability inventory: The skill is focused on tracking, scoring, and generating Markdown digests. It does not explicitly request high-privilege tools, but the ingestion of arbitrary web content into the agent context creates a surface for prompt injection.
  • Sanitization: The instructions do not define any filtering or sanitization steps for the ingested metadata before it is processed for scoring and summarization.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 02:59 AM
Security Audit — agent-trust-hub — physics-literature-search