agent-framework-azure-ai-py
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill provides patterns for agents to ingest data from external sources like web searches and documents. While intended, this creates a surface for potential indirect prompt injection.
- Ingestion points: Data entering through
HostedWebSearchToolandHostedFileSearchToolinreferences/tools.md. - Boundary markers: The documentation includes citation formats (e.g.,
【source†title】) inreferences/advanced.mdto help the agent distinguish between source content and generated responses. - Capability inventory: The framework supports Python code execution via
HostedCodeInterpreterTooland external API calls through MCP tools, as described inSKILL.mdandreferences/mcp.md. - Sanitization: The framework is designed to utilize isolated service-side environments for tool execution to mitigate risks.
- Dynamic Execution Capabilities: The skill demonstrates the use of the
HostedCodeInterpreterTool, which allows agents to generate and execute Python code. This capability is a core feature of the framework and is executed within the Azure AI service's secure and isolated infrastructure. - External Package References: The installation steps in
SKILL.mdreference theagent-frameworkandagent-framework-azure-aipackages. These are the official SDK components provided by the vendor for building agent applications.
Audit Metadata