Skills

A skill is a set of instructions that gets installed into your AI coding assistant, teaching it how to use Weco end-to-end - from setting up optimizations to interpreting results. Once installed, just describe what you want to optimize in plain language and your assistant handles the rest.

Why use a skill?

Instead of remembering CLI flags and writing evaluation scripts from scratch, a skill lets you work naturally:

• Natural language - just say "make this function faster" or "optimize accuracy"
• Automated setup - your assistant inspects your code, writes the evaluation, and configures the run
• Live monitoring - the assistant watches the optimization, installs missing dependencies, and reports progress
• Results explained - you get a clear summary of what changed, why it worked, and the real-world impact

Claude Code

weco setup claude-code

~/.claude/skills/weco/
├── CLAUDE.md          # Trigger snippet (Claude reads this automatically)
├── SKILL.md           # Full optimization workflow and instructions
├── references/        # Advanced documentation (benchmarking, evaluation, etc.)
└── assets/            # Template evaluation scripts

Use Weco to make this function faster.

Optimize module.py for accuracy using evaluate.py as the benchmark.

Reduce the latency of the API handler.

Cursor

weco setup cursor

~/.cursor/
├── rules/
│   └── weco.mdc           # Always-on trigger rule
└── skills/
    └── weco/
        ├── SKILL.md       # Full optimization workflow and instructions
        ├── references/    # Advanced documentation
        └── assets/        # Template evaluation scripts

Use Weco to make this function faster.

Optimize module.py for accuracy using evaluate.py as the benchmark.

Reduce the latency of the API handler.

How it works

When you ask your assistant to optimize something, the skill guides it through a structured workflow:

1. Analyze - The assistant inspects your code and infers what metric to optimize (speed, accuracy, memory, etc.)
2. Baseline - It runs your code to measure current performance and makes the impact tangible
3. Evaluate - It writes (or reuses) an evaluation script that measures the metric
4. Optimize - It runs weco run with the right flags, monitors progress, and handles errors automatically
5. Report - It presents results with a clear explanation of what changed, why it worked, and the real-world impact
6. Apply - It asks your permission before modifying any project files

Two modes

When you start an optimization, the skill will ask you to choose a mode:

Example: Hello, World!

Here's a complete walkthrough using the skill with the Quickstart example:

git clone https://github.com/WecoAI/weco-cli.git
cd weco-cli/examples/hello-world/
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

claude

cursor .

Use the Weco skill to make the forward pass faster.

For the full step-by-step walkthrough (including CLI usage), see the Quickstart guide.

What can skills optimize?

Skills work with anything the Weco CLI supports:

• GPU kernels - CUDA, Triton, PyTorch
• ML models - Model development, training pipelines, inference
• Prompts - LLM prompt engineering, agent behaviors
• General code - Any language (Python, Rust, C++, JavaScript, Go, etc.), any metric you can measure

Troubleshooting

What's next?

• Quickstart - Run your first optimization step by step
• Set Up Your Own Optimization - Apply Weco to your own code
• CLI Reference - All CLI commands and options
• Writing Good Evaluation Scripts - Get better results with well-designed benchmarks
• Examples - See Weco optimize CUDA kernels, ML models, and more