The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface because it is designed to process and review untrusted source code and repository data.
Ingestion points: Files within the local repository and .NET source code context that the agent is instructed to inspect (SKILL.md).
Boundary markers: The instructions do not explicitly require the use of delimiters or "ignore embedded instructions" warnings when processing the repository content.
Capability inventory: The agent is authorized to edit files and execute shell commands such as dotnet build, test, lint, and validation (SKILL.md).
Sanitization: No explicit sanitization or validation protocols are defined for handling instructions that might be embedded within the source files being reviewed.
[COMMAND_EXECUTION]: The skill workflow involves executing developer-centric shell commands, including project build and test execution via the .NET CLI (SKILL.md). While these are standard capabilities for a development assistant, they represent an execution surface that could be leveraged by a malicious actor if influenced via prompt injection.
[EXTERNAL_DOWNLOADS]: The included documentation provides instructions for installing .NET SDKs, tool templates, and NuGet packages from the official NuGet registry and Microsoft domains. It also references downloading AI models from the Ollama registry for local model development (references/official-docs/quickstarts/chat-local-model.md). These references target well-known and trusted services.

microsoft-extensions-ai