The Agent Skills Directory

Architectural Best Practices and Secure Configurations: The instruction set mandates specific security and resilience configurations for AI/ML tasks.
Enforces loading API keys and credentials solely from secure sources such as Environment Variables, .NET User-Secrets, or Azure Key Vault, strictly prohibiting hardcoded token strings.
Instructs using standard platform abstractions (Microsoft.Extensions.AI, Microsoft.Extensions.VectorData.Abstractions) and registering services via Dependency Injection (DI) rather than direct instance management inside business logic.
Directs developers to integrate error handling and transient fault mitigation tools like RetryingChatClient or Polly retry loops.
Native Memory Management Controls: For production inference workloads using ONNX Runtime, the instructions direct developers to instantiate a singleton InferenceSession to avoid file-system thrashing, and explicitly require the prompt disposal of native-memory-backed results (IDisposable pattern).
Data Handling Boundaries: The RAG and ingestion guidelines require structural mechanisms such as semantic chunking boundaries, minimum relevance thresholds to filter noise, and embedding cache implementation strategies.

technology-selection