dspy

Compile typed signatures and modules against defined metrics to optimize language model prompt programs.

40|6|Updated Jul 11, 2026

Search Tags:#prompt engineering #python #optimization #llm #compiler #dspy

Ready to run

npx skills add https://github.com/magnus919/agent-skills --skill dspy-magnus919

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: dspy
Source: https://github.com/magnus919/agent-skills/tree/main/dspy
Command: npx skills add https://github.com/magnus919/agent-skills --skill dspy-magnus919

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires dspy, and includes scripts (resource) and references (resource) components.

What problem does it solve?

This skill solves the instability and manual labor of traditional prompt engineering by treating prompt programs as compilable code that can be automatically optimized against specific metrics.

Core Features & Use Cases

Programmatic Optimization: Automatically tunes instructions and few-shot examples using Bayesian or evolutionary search.
Modular Architecture: Build complex pipelines using typed signatures and Python control flow rather than brittle string concatenation.
Use Case: When you have a multi-step reasoning task that is failing, use this skill to compile your program against a validation dataset to find the optimal prompt instructions and few-shot examples that maximize accuracy.

Quick Start

Use the dspy skill to compile a ChainOfThought program against the provided training dataset to optimize its performance.

Frequently Asked Questions about dspy

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automatically optimize LLM prompt programs against specific metrics?▼

You can optimize LLM prompt programs by compiling typed signatures and modules against defined metrics using the dspy framework. This treats prompts as compilable code, automatically tuning instructions and few-shot examples via Bayesian or evolutionary search.

What is the best way to fix a failing multi-step reasoning pipeline without manual prompt tuning?▼

The best way to fix a failing multi-step reasoning pipeline is to compile your program against a validation dataset. This systematically finds the optimal prompt instructions and few-shot examples to maximize accuracy, eliminating manual string tuning.

Does dspy work with Python for building complex agentic tool-use workflows?▼

Yes, dspy works with Python to build complex agentic tool-use workflows. It uses a modular architecture with typed signatures and Python control flow, replacing brittle string concatenation with structured, compilable program logic.

How do I compile a ChainOfThought program to find the optimal prompt instructions?▼

To compile a ChainOfThought program, compile it against a provided training dataset using the dspy framework. This manages program structure, optimizer selection, and evaluation loops to automatically discover the best prompt instructions.

When should I use programmatic prompt optimization instead of manual string concatenation?▼

You should use programmatic prompt optimization when you face instability and manual labor in traditional prompt engineering. It is ideal for multi-step reasoning tasks, allowing you to systematically compile and evaluate typed modules against validation data.

Can I use Bayesian optimization to tune few-shot examples for language model pipelines?▼

Yes, you can use Bayesian or evolutionary search to tune few-shot examples and instructions for language model pipelines. The dspy framework manages these optimization loops to maximize your program's accuracy against defined metrics.