The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of Retrieval-Augmented Generation (RAG) and agentic applications that ingest untrusted data from external sources and have the capability to execute actions based on that data.
Ingestion points: Data enters the agent context through Azure AI Search indexes (foundry_knowledge_index_list) and Bing web search results (BingGroundingToolDefinition) as described in language/python.md.
Boundary markers: Instructions provided in language/python.md include defensive prompts such as "If you cannot find the answer in the knowledge base, say 'I don't know'" and "Never answer from your own knowledge
only use the knowledge base."
Capability inventory: Agents are equipped with FunctionTool for executing custom Python functions and BingGroundingToolDefinition for web exploration, creating a potential for automated action based on malicious data.
Sanitization: The provided code samples do not implement explicit sanitization, filtering, or validation of the retrieved external content before it is processed by the language model.
[EXTERNAL_DOWNLOADS]: The guide includes instructions to install official and well-known Python libraries.
Packages: Fetches the following libraries from standard registries: azure-ai-projects, azure-identity, azure-ai-inference, openai, azure-ai-evaluation, and python-dotenv.
[COMMAND_EXECUTION]: The skill provides numerous Azure CLI commands to help users manage their infrastructure.
Commands: Includes commands like az resource list, az cognitiveservices account deployment create, and az role assignment create to configure services and permissions within the user's Azure environment.

microsoft-foundry