semantic-kernel
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents methods for processing user input through
kernel.InvokePromptAsyncinSKILL.md. It explicitly recommends boundary markers and behavior steering via system messages, as detailed inanti-patterns.md. The capability inventory includes functions for data retrieval and API calls (e.g.,WeatherPlugin,OrderPlugin). Sanitization and safety guidance are pervasive, emphasizing input validation and structured output formatting to prevent injection and hallucination risks. - [CREDENTIALS_UNSAFE]: The skill uses safe configuration patterns (e.g.,
config["AzureOpenAI:ApiKey"]) for managing API keys and explicitly identifies hardcoding credentials in source code as a dangerous anti-pattern. - [EXTERNAL_DOWNLOADS]: All external references target official Microsoft documentation and GitHub repositories, which are established and trusted sources for the described technology stack.
- [DYNAMIC_EXECUTION]: The skill correctly identifies the removal of legacy planners in favor of modern function-calling behaviors, reducing the risk surface associated with uncontrolled execution steps.
Audit Metadata