The Agent Skills Directory

Dynamic Code Execution: The calibrate_grader.py script utilizes the exec() function to load and run Python files supplied by the user. This is a design choice to allow for flexible scoring logic during model calibration. The documentation includes a security note advising users to only use grader files they have reviewed.
Indirect Prompt Injection Surface: The skill features workflows for harvesting production traces and curating datasets for evaluation. As outlined in eval-datasets.md and observe.md, these processes involve ingesting external data which could theoretically contain instructions intended to influence the agent's behavior during automated testing.
Command Line Interface Execution: Scripts such as deploy_model.py and smoke_test_voice.py use the subprocess module to interact with the Azure CLI. These operations are necessary for authentication via token acquisition and for managing Azure AI Services resources.
External Dependency Resolution: The documentation and scripts reference the installation of official Microsoft extensions and common AI libraries like openai and azure-ai-projects. These are required for the skill to interact with the Azure environment and are resolved through standard package registries.

microsoft-foundry