dspy-signature-designer_skill

Overview

This skill helps design clear, type-safe DSPy signatures to define what a DSPy module expects and returns. It guides creation of inline and class-based signatures, adding InputField/OutputField definitions, type hints, and optional Pydantic models for structured validation.

How this skill works

It inspects a task description and lists of input/output fields and type constraints, then generates a dspy.Signature class (or inline signature) with InputField and OutputField annotations. The skill can include Literal constraints, Optional types, collection hints, Pydantic models for nested structures, and validation constraints like min/max or ge/le. It also produces docstrings and field descriptions to guide model behavior.

When to use it

• Defining a new DSPy module or wrapping an LLM task with a signature
• When outputs must be structured, validated, or type-safe
• Designing multi-field responses or complex input/output relationships
• Adding Pydantic models to represent nested or record-like outputs
• Converting informal task descriptions into explicit DSPy inputs/outputs

Best practices

• Write a descriptive class docstring — it becomes the task instruction
• Use dspy.InputField.desc and OutputField.desc to guide the model
• Constrain categorical outputs with typing.Literal for safety
• Provide sensible defaults for optional inputs; validate outputs in forward()
• Use Pydantic models for complex nested types and post-run validation

Example use cases

• Summarization signature: document -> list[str] summary + word_count
• Entity extraction: text -> List[Entity] where Entity is a Pydantic model
• Sentiment analysis: text -> sentiment (Literal) + score + aspects + reasoning
• RAG answer module: context:list[str], question -> answer, confidence, source_passage
• Multi-label classifier returning categories list and a primary_category