The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents the use of the dspy.ProgramOfThought module for math and parsing tasks. This module allows the language model to generate and execute code at runtime to ground its responses, which is a native and intended feature of the DSPy framework.
[INDIRECT_PROMPT_INJECTION]: The skill outlines the creation of evaluation metrics using LLM-as-judge techniques, creating a potential surface for indirect prompt injection. 1. Ingestion points: Input data evaluated by the metric functions as described in SKILL.md. 2. Boundary markers: The skill recommends returning boolean values during optimization to reduce the model's susceptibility to instructions within the data. 3. Capability inventory: The framework includes code execution via ProgramOfThought and external tool usage via ReAct. 4. Sanitization: The SOP suggests decomposing metrics into specific sub-judges to validate individual criteria like factuality and conciseness.
[SAFE]: The skill references reputable documentation and research repositories such as dspy.ai and arxiv.org. It provides operational guardrails for managing optimization costs and ensuring deployment reproducibility using FastAPI and MLflow.

agentsop-dspy