citation-and-reference-validator

Pass

Audited by Gen Agent Trust Hub on Jul 27, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a bundled Python script (validate_references.py) to automate the verification of academic metadata, providing structured reports on reference validity.
  • [EXTERNAL_DOWNLOADS]: The validation script performs network requests to well-known academic services, specifically the Crossref and OpenAlex APIs, to fetch authoritative metadata for papers and DOIs.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from external academic databases and user-supplied reference files, creating a potential surface for indirect prompt injection.
  • Ingestion points: Academic metadata and abstracts fetched from Crossref/OpenAlex APIs and local references.json file.
  • Boundary markers: No explicit delimiters are used to wrap external content.
  • Capability inventory: Execution of a local Python script via shell commands.
  • Sanitization: Employs standard JSON parsing and regex-based normalization of titles.
  • [SAFE]: The skill documentation includes strong adversarial guidelines focused on preventing reference hallucination and ensuring contextual alignment, demonstrating a high focus on data integrity.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 27, 2026, 09:42 AM
Security Audit — agent-trust-hub — citation-and-reference-validator