The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides detailed instructions and code examples for using the CliWrap library to execute external command-line processes from .NET applications. The documentation appropriately emphasizes security best practices, such as using array-based arguments to ensure automatic escaping of spaces and special characters, which helps prevent command injection vulnerabilities.
[DATA_EXFILTRATION]: The skill documents the Model Context Protocol (MCP), which allows AI agents to interact with local resources like the file system and databases. The provided examples and rules include critical security guidance, such as validating all tool inputs, restricting database queries to read-only SELECT statements, and sanitizing string inputs to prevent injection attacks.
[PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data through file reading, database queries, and multi-agent communication protocols. While this creates a surface for indirect prompt injection, the skill mitigates these risks by recommending strict input validation, the use of explicit boundary markers, and structured response formats (like JSON) for tool outputs.
Ingestion points: MCP tools (ReadFile, QueryDatabase), Semantic Kernel chat history, and A2A protocol task messages.
Boundary markers: The documentation recommends using .EscapeMarkup() for terminal rendering and providing detailed [Description] attributes for tool discovery to reduce model confusion.
Capability inventory: Support for executing shell commands via CliWrap and performing file system operations via MCP tool definitions.
Sanitization: Explicit guidance to validate and sanitize all tool inputs and to use strongly-typed generics and options to avoid manual string parsing errors.

dotnet