exa-search
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The instruction files references/web-search.md and references/web-extract.md provide command templates that interpolate user-provided variables ($ARGUMENTS) directly into shell commands via uv run. This requires the agent to properly escape input to prevent command injection.
- [INDIRECT_PROMPT_INJECTION]: The skill fetches and processes arbitrary content from the web and scholarly sources, making it vulnerable to malicious instructions embedded in that content. Mandatory Evidence Chain: 1. Ingestion points: scripts/exa_search.py and scripts/exa_extract.py (web fetching). 2. Boundary markers: Absent from the prompt instructions. 3. Capability inventory: Python script execution, local file writing, and network requests via the Exa API. 4. Sanitization: None performed on the fetched external data before processing.
- [DYNAMIC_EXECUTION]: The skill uses uv run to execute local Python scripts based on command templates that are dynamically populated with user arguments at runtime.
- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation and execution of the exa-py and python-dotenv packages from the official Python Package Index (PyPI). These are well-known libraries required for the skill's primary search functionality.
Audit Metadata