past-research-search
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill uses Python scripts embedded in shell commands to manage a local research log. These scripts are static templates used for standard file I/O operations.
- [INDIRECT_PROMPT_INJECTION]: The skill maintains a local log of user-provided research queries, creating an ingestion surface for previously processed data. 1. Ingestion points: User research queries stored in .tavily/runs.jsonl. 2. Boundary markers: The data is stored in structured JSONL format. 3. Capability inventory: Local file access and execution of the tvly CLI tool. 4. Sanitization: The Python script uses json.dumps for serialization, ensuring queries are handled as data rather than instructions.
Audit Metadata