verify-refs
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill performs its primary functions using local, deterministic Python scripts rather than relying on generative model memory, which mitigates the risk of hallucinated or manipulated citation data.
- [DATA_EXFILTRATION]: Network activity is restricted to well-known, authoritative scientific metadata services (crossref.org, ncbi.nlm.nih.gov, openalex.org, and doi.org). These calls are essential for the tool's purpose and do not involve the transmission of sensitive local credentials or files.
- [SAFE]: The implementation includes specific logic (portable_source) and regression tests to prevent the leakage of absolute filesystem paths—which often contain local usernames—into the audit reports generated by the skill.
- [INDIRECT_PROMPT_INJECTION]: While the skill processes user-supplied manuscripts, it extracts structured metadata into a JSON format with a defined schema. The reporting helpers further sanitize this data when rendering Markdown tables, reducing the risk of injected content influencing downstream agent actions.
- [COMMAND_EXECUTION]: The use of subprocesses is confined to the test suite for validating CLI behavior and is not part of the runtime path for manuscript analysis.
Audit Metadata