azure-ai-projects-ts
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Credential Management Patterns: The skill provides documentation and examples for retrieving authentication details from Azure project connections, including API keys. This functionality is essential for the SDK to interact with linked services like Azure OpenAI and AI Search. Users should follow the skill's own best practices to avoid logging these credentials in production environments.
- Agent Tool Capabilities: The skill defines agents that can be configured with powerful tools, such as a code interpreter, web search, and Model Context Protocol (MCP) servers. These tools enable the agent to perform complex tasks but require appropriate scoping of permissions to ensure they operate within the intended boundaries.
- Data Processing for Evaluations: The evaluation features ingest datasets and query-response pairs to measure model performance. Because this data influences evaluation metrics and uses prompt-based evaluators, it represents a surface where untrusted input could potentially affect results.
- Ingestion points: Evaluation data is ingested via inline arrays in
references/evaluations.mdand through dataset uploads inSKILL.md. - Boundary markers: No specific delimiters or "ignore" instructions are shown in the examples.
- Capability inventory: The skill enables model inference via Azure OpenAI, code execution through the code interpreter tool, and network access via search tools.
- Sanitization: The examples show direct interpolation of data into prompt templates for custom evaluators without explicit sanitization steps.
Audit Metadata