python-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DYNAMIC_EXECUTION]: Documentation in references/tool-builder.md and references/agent-patterns.md includes examples using the eval() function to process tool arguments. This pattern is unsafe as it allows for arbitrary code execution if the input is not strictly validated.
- [COMMAND_EXECUTION]: The SDK provides a built-in code_execution tool that allows agents to run arbitrary scripts. While a feature of the library, it increases the risk profile of agents built with this SDK.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a framework for agents that ingest data from external sources (web search, file uploads, API responses) and have access to powerful tools. The documentation lacks clear guidance on sanitization or boundary markers to prevent the agent from following instructions embedded in untrusted data. (1) Ingestion points: agent messages, file content, and search results; (2) Boundary markers: Not specified in examples; (3) Capability inventory: code execution and webhooks; (4) Sanitization: Examples demonstrate direct execution of tool arguments.
- [EXTERNAL_DOWNLOADS]: The skill references several Python dependencies for installation, including inferencesh, requests, aiohttp, tqdm, and fastapi.
Audit Metadata