legal-red-team
Pass
Audited by Gen Agent Trust Hub on Apr 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: Technical analysis of the skill's components confirms that it is a legitimate defensive toolset for legal document verification. The provided Python scripts,
legal_fact_checker.pyandlegal_quality_scorer.py, are benign utilities that use standard Python libraries (re,json,argparse) to perform static pattern matching on text. They do not contain any code for network communication, unauthorized file access, or credential harvesting. - [COMMAND_EXECUTION]: The skill utilizes local Python scripts to automate the identification of errors. These scripts are invoked with standard arguments and do not interact with the system shell in a way that would allow for command injection or privilege escalation. The code review confirms that the scripts only process the data for the purpose of scoring and flagging.
- [PROMPT_INJECTION]: The skill's workflow involves processing untrusted legal documents, which constitutes a surface for indirect prompt injection. Ingestion points: Document text processed through scripts via the
--inputargument. Boundary markers: Absent; the scripts analyze raw text input. Capability inventory: Local file reading and writing (via script outputs) and web searching (as specified in the methodology for human/agent verification). Sanitization: Absent. However, this surface is evaluated as safe because the skill is designed for adversarial fact-checking, and the scripts themselves do not execute instructions embedded in the data.
Audit Metadata