research-guardian
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The skill performs network searches to verify research citations. According to the documentation, these operations target established scholarly platforms such as PubMed, Google Scholar, and Semantic Scholar.
- [REMOTE_CODE_EXECUTION]: The system is designed to execute statistical verification scripts and conduct sanity checks on research code to ensure accuracy. This functionality is part of the skill's primary purpose of research validation.
- [COMMAND_EXECUTION]: The skill includes a Python script (runner.py) that manages the pipeline execution, checkpointing, and report generation using local file operations in /tmp/research-guardian.
- [PROMPT_INJECTION]: The skill processes untrusted research papers which could contain adversarial content. Ingestion points: runner.py reads user-provided files into the agent context. Boundary markers: The Paper Parser system (paper-parser.md) segments input into sections before analysis. Capability inventory: Scholarly network search and script execution for verification. Sanitization: Employs anti-hallucination rules and normalization schemas to prevent the evaluator's output from being influenced by embedded data content.
Audit Metadata