legal-search
Pass
Audited by Gen Agent Trust Hub on Apr 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill retrieves legal case law and legislative data from well-known and official public repositories, including CourtListener (US Federal and State opinions), Harvard Case Law Access Project (historical US cases), and the EUR-Lex portal (official EU legislation and case law).
- [COMMAND_EXECUTION]: Employs curl to fetch API responses and python3 to parse and format the resulting JSON data. The Python logic is embedded directly within the skill and performs standard extraction of fields like case names, dates, and citations from the API responses.
- [REMOTE_CODE_EXECUTION]: While automated scans flagged the piping of remote data to a Python interpreter, analysis confirms these operations target well-known and reputable legal information services. The Python scripts are local to the skill and serve as data formatters rather than executing arbitrary code from the remote source.
- [PROMPT_INJECTION]: The skill processes untrusted data from external legal APIs, representing a potential indirect prompt injection surface.
- Ingestion points: Data retrieved from CourtListener, Harvard Case Law, and EUR-Lex APIs (SKILL.md).
- Boundary markers: None identified; fields from API responses are printed directly to the console.
- Capability inventory: Network access (curl), file writing (via curl -o), and local script execution (python3).
- Sanitization: None observed; the scripts print data directly from the JSON responses without escaping.
Audit Metadata