Welcome to the AI Makerspace workshop on UPSkill - HuggingFace's CLI tool for generating and evaluating reusable agent skills.
UPSkill enables you to create reusable "skills" with powerful teacher models (expensive/slow) that student models (cheap/fast) can use to perform harder tasks reliably. Instead of always relying on expensive frontier models, you can:
- Generate skills using a teacher model (e.g., Claude Opus, GPT-4)
- Package instructions into reusable SKILL.md files
- Evaluate performance across multiple student models
- Automatically refine skills until they meet quality thresholds
This approach lets you "distill" the reasoning of expensive models into portable, reusable instructions that dramatically improve the performance of cheaper models.
This repository contains materials for learning UPSkill through hands-on examples. You'll generate real skills, evaluate them against multiple models, and understand when and how to use this powerful tool in production.
- The Teacher/Student Paradigm - How expensive models can teach cheaper ones
- Skill Generation - Creating skills from simple task descriptions
- Skill Structure - Understanding SKILL.md files and supporting documents
- Evaluation & Metrics - Measuring skill lift and success rates
- Multi-Model Benchmarking - Comparing performance across providers
- Automatic Refinement - Iterating until quality thresholds are met
- Python API Integration - Using UPSkill programmatically in your applications
In production AI systems, you often face a tradeoff:
| Approach | Cost | Latency | Quality |
|---|---|---|---|
| Always use frontier models | High | High | High |
| Always use cheaper models | Low | Low | Variable |
| UPSkill approach | Low | Low | High |
UPSkill lets you invest once in generating high-quality skills, then reuse them infinitely with cheaper models. This is particularly valuable for:
- Repetitive tasks (commit messages, code review, data extraction)
- Cost-sensitive applications (high-volume processing)
- Latency-sensitive applications (real-time systems)
- Consistent outputs (standardized formats, compliance)
git clone <repository-url>
cd "UPSkill Event"uv syncCreate a .env file with your API keys:
ANTHROPIC_API_KEY=your-anthropic-key-here
OPENAI_API_KEY=your-openai-key-here # Optional, for multi-model evaluationuv run jupyter notebook| Section | Topic | Description |
|---|---|---|
| 1 | Introduction | Understanding UPSkill and the teacher/student paradigm |
| 2 | Setup | API configuration and installation verification |
| 3 | First Skill | Generate a skill for writing git commit messages |
| 4 | Evaluation | Run evaluations and understand metrics |
| 5 | Complex Skills | Generate skills from detailed task descriptions |
| 6 | Refinement | Automatic skill improvement based on failures |
| 9 | Conclusion | Key takeaways and resources |
After completing this workshop, you will be able to:
- Generate reusable skills using UPSkill CLI and Python API
- Evaluate skill effectiveness across multiple models
- Understand when UPSkill is the right solution for your use case
- Integrate skill-enhanced prompting into production applications
- Create custom skills with supporting documentation
| File | Description |
|---|---|
01_UPSkill_Tutorial.ipynb |
Main workshop notebook |
pyproject.toml |
Python project configuration |
.env |
API keys (create this yourself) |
README.md |
This file |
Built with care by AI Makerspace