The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were identified. The skill consists of documentation and educational code examples that utilize the DSPy library for structural prompt handling.
[PROMPT_INJECTION]: The skill describes an architecture for processing untrusted data through LLMs, but implements robust defensive measures via its delimiter system.
Ingestion points: Untrusted data enters via signature inputs such as ticket_text or notes in the examples.md file.
Boundary markers: The adapter uses a strict field delimiter protocol ([[ ## field_name ## ]] and [[ ## completed ## ]]) to separate user data from instructions.
Capability inventory: No executable scripts or exploitable system capabilities (e.g., file writes, subprocess calls) are included in the skill.
Sanitization: The adapter performs structured parsing and type validation on all LM responses.

dspy-chatadapter