verify-refs

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing untrusted user manuscripts and bibliographies to audit citations.
  • Ingestion points: The skill reads manuscript content from multiple file formats, including Markdown, DOCX, BibTeX, and TSV files, in the verify_refs.py and check_claim_fidelity.py scripts.
  • Boundary markers: The processing logic does not implement explicit boundary markers or instructions to isolate ingested manuscript text from the agent's control flow.
  • Capability inventory: The skill is capable of writing audit reports (JSON and Markdown) to the project directory and performing network lookups via bibliographic APIs.
  • Sanitization: Text from manuscripts is normalized and stripped of markdown noise before processing, and outputs are handled via standard JSON serialization to prevent structural injection.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 12:24 PM
Security Audit — agent-trust-hub — verify-refs