DSPy Skills Collection

A comprehensive collection of AI-powered skills for programming and optimizing LLM applications using the DSPy framework. These skills enable you to move from manual prompt engineering to systematic, programmatic LLM development.

Version Compatibility: All skills target DSPy 3.1.2 (released January 19, 2026). All code examples and APIs have been verified against this version.

🎯 What is DSPy?

DSPy is a declarative framework that lets you program language models instead of prompting them. It provides:

• Modular architecture - Compose LLM programs from reusable components
• Automatic optimization - Tune prompts, examples, and weights algorithmically
• Self-improving pipelines - Systems that get better with data

📚 Skills Index

Optimizers

Skill	Description	Best For
dspy-bootstrap-fewshot	Auto-generate few-shot examples	Quick optimization with ~10 examples
dspy-miprov2-optimizer	Bayesian instruction+demo optimization	200+ examples, comprehensive tuning
dspy-gepa-reflective	LLM reflection on execution traces	Agentic systems, complex workflows
dspy-simba-optimizer	Mini-batch Bayesian optimization	Custom feedback, budget-friendly
dspy-finetune-bootstrap	Fine-tune model weights	Production deployment, efficiency

Pipelines & Components

Skill	Description	Best For
dspy-rag-pipeline	RAG with ColBERTv2 retrieval	Knowledge-grounded generation
dspy-signature-designer	Design type-safe I/O specs	Clean, validated outputs
dspy-evaluation-suite	Metrics and evaluation	Quality assessment
dspy-haystack-integration	DSPy + Haystack pipelines	Existing Haystack projects

Agent Development

Skill	Description	Best For
dspy-react-agent-builder	Build ReAct agents with tools	Multi-step reasoning tasks

Output Validation

Skill	Description	Best For
dspy-output-refinement-constraints	Refine outputs with constraints	Format/content validation

Advanced Patterns

Skill	Description	Best For
dspy-advanced-module-composition	Ensemble and multi-chain patterns	Complex multi-module programs
dspy-custom-module-design	Build custom DSPy modules	Reusable production components

Debugging & Monitoring

Skill	Description	Best For
dspy-debugging-observability	MLflow tracing and monitoring	Production debugging, cost tracking

🎯 Choosing the Right Optimizer

Your Situation	Recommended Skill	Why
10-50 labeled examples	dspy-bootstrap-fewshot	Fast, cost-effective
200+ examples, need best performance	dspy-miprov2-optimizer	State-of-the-art results
Building agents with tools	dspy-gepa-reflective	Optimizes execution traces
Production deployment, cost reduction	dspy-finetune-bootstrap	Creates efficient fine-tuned models

🔄 Typical Workflow

1. Design → Signature Designer - Define inputs/outputs
2. Build → RAG Pipeline or Agent Builder - Create your DSPy program
3. Validate → Output Refinement - Add constraints
4. Optimize → Choose optimizer based on your data
5. Evaluate → Evaluation Suite - Measure improvements
6. Debug → Debugging & Observability - Monitor performance
7. Deploy → Fine-tune Bootstrap - Optional production optimization

📖 Documentation

• DSPy Framework Guide - Complete framework reference

🚀 Installation

# Install DSPy 3.1.2 or later
pip install dspy-ai>=3.1.2

Optional Dependencies

# For ColBERTv2 retrieval
pip install colbert-ai

# For Haystack integration
pip install haystack-ai

# For fine-tuning
pip install transformers datasets

💡 Quick Start

import dspy

# Configure LM
dspy.configure(lm=dspy.LM("openai/gpt-4o-mini"))

# Create a simple classifier
classify = dspy.Predict("text -> sentiment: bool")
result = classify(text="I love this product!")
print(result.sentiment)  # True

📁 Examples

See examples/code-snippets.py for production-ready code.

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

📜 License

MIT License - see LICENSE for details.

🙏 Acknowledgments

Built for the SkillsMP marketplace. Uses the DSPy framework by Stanford NLP.