python-sdk
Warn
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The documentation includes multiple code examples in 'references/tool-builder.md' and 'references/agent-patterns.md' that use 'eval()' to execute logic based on arguments provided by the AI (e.g., 'eval(call.args["expression"])'). This promotes an insecure coding pattern where unsanitized input from a model can lead to arbitrary code execution.
- [COMMAND_EXECUTION]: The SDK documentation features an 'Internal Tools' API that allows developers to enable a 'code_execution(True)' capability. This feature enables agents to run code in their environment, which poses a high risk if the agent is targeted by prompt injection or processes malicious external data.
- [PROMPT_INJECTION]: The skill documents patterns for agents to ingest and process external content, which facilitates indirect prompt injection attacks.
- Ingestion points: Data enters the agent context through file attachments, web search outputs, and webhook responses.
- Boundary markers: The provided examples do not demonstrate the use of delimiters or instructions to ignore commands within the ingested text.
- Capability inventory: The SDK supports high-privilege capabilities including Python code execution and outbound network requests.
- Sanitization: The examples do not include any logic for sanitizing or validating external data before it is processed by the agent.
Audit Metadata