citation-faithfulness

Pass

Audited by Gen Agent Trust Hub on Jul 26, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill operates entirely through the agent's built-in capabilities (Read, WebSearch, WebFetch) and does not bundle or execute external scripts.
  • [SAFE]: Data handling is restricted to the local environment and well-known, trusted academic repositories (arXiv, publisher landing pages). While an optional AMiner API key can be used for enrichment, the instructions explicitly forbid printing the token and treat it as a sensitive environment variable.
  • [SAFE]: The logic includes extensive guardrails against 'hallucination' or false accusations, mandating an 'UNVERIFIABLE' verdict rather than a negative one if full text cannot be retrieved.
  • [SAFE]: No obfuscation, persistence mechanisms, or privilege escalation patterns were found. The use of Write to output results is a standard feature for generating reports.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 26, 2026, 12:18 AM
Security Audit — agent-trust-hub — citation-faithfulness