The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation recommends installing the dspy[mcp] package via pip to enable Model Context Protocol support.
[COMMAND_EXECUTION]: The instructions for MCP integration demonstrate how to initialize local servers using StdioServerParameters, which involves executing local commands or scripts (e.g., python path/to/your/mcp_server.py) to establish a subprocess-based communication session.
[PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface through the processing of external data in DSPy modules.
Ingestion points: Untrusted data enters the agent context via input fields like question in the streaming example and ticket_text in the classification module (SKILL.md, examples.md).
Boundary markers: The skill documentation notes the use of ChatAdapter delimiters (e.g., [[ ## field_name ## ]]) to separate fields within the LM prompt (SKILL.md).
Capability inventory: The skill has the capability to perform local file read/write operations for state persistence via save and load methods and to execute subprocesses for MCP server sessions (SKILL.md).
Sanitization: No explicit input sanitization or validation logic is detailed in the provided code snippets, relying instead on framework-level markers.

dspy-utils