MAKORA SKILLS¶

Develop and optimize GPU kernels directly from Claude Code.

Makora Skills is a Claude Code plugin that gives you access to GPU kernel evaluation, optimization, generation, and documentation search — all from within your Claude Code session.

Overview¶

With Makora Skills installed, you can:

Evaluate your GPU code against a reference implementation
Optimize a CUDA/Triton kernel iteratively to beat the reference
Generate an optimized Triton kernel from a problem description
Search for GPU optimization snippets, techniques, and documentation

Quick Start¶

1. Confirm Makora is already installed¶

Check Makora CLI version and logged in user info:

makora info

This shows your username, Makora version, and current environment variable settings. If this command fails or you're not logged in, follow the steps from Getting Started

2. Install the Claude Code Plugin¶

makora install claude

This automatically:

Clears the Claude Code plugin cache for the makora marketplace
Installs the latest marketplace configuration
Installs the Makora plugin with all skills and commands

3. Start Using It¶

Ask Claude to evaluate or optimize your code using Makora, or use the slash commands directly. See the Commands section for the full list.

Commands¶

Makora Skills provides slash commands you can use directly in Claude Code.

Available Commands¶

Command	Description
`/makora-plugin:evaluate`	Evaluate your GPU code against a reference implementation
`/makora-plugin:optimize`	Iteratively optimize a CUDA/Triton kernel to beat the reference
`/makora-plugin:generate`	Generate an optimized Triton kernel from a problem description
`/makora-plugin:search-snippets`	Search for GPU optimization snippets and techniques
`/makora-plugin:search-docs`	Search for relevant GPU documentation

Usage¶

You can use commands in two ways:

1. Slash Commands¶

Type the command directly in Claude Code:

/makora-plugin:evaluate

Claude will guide you through providing the necessary inputs (file paths, problem descriptions, etc.).

2. Natural Language¶

Just ask Claude to perform the action. Mention specific file paths to reference code or an existing solution:

Evaluate my kernel in solution.py against the reference in problem.py using Makora.

Optimize the kernel in solution.py for H100.

Search for GPU optimization techniques for matrix multiplication.

Evaluate¶

Evaluates your GPU code against a reference implementation. Provides performance metrics and correctness checks.

/makora-plugin:evaluate

Optimize¶

Iteratively optimizes a CUDA/Triton kernel to beat a reference implementation. Runs multiple rounds of evaluation and improvement.

/makora-plugin:optimize

Generate¶

Generates an optimized Triton kernel from a problem description. Creates a starting solution you can then iterate on.

/makora-plugin:generate

Search Snippets¶

Searches for GPU optimization snippets and techniques from curated sources.

/makora-plugin:search-snippets

Search Docs¶

Searches for relevant GPU programming documentation and API references.

/makora-plugin:search-docs

Examples¶

The examples/ directory in the makora-skills repository contains sample GPU kernel optimization problems and a script to run them with Claude Code.

Running an Example¶

Use the provided script to generate a solution for any example problem:

./examples/run-example.sh kernelbench-problem-1

This script will:

Load the problem definition from examples/kernelbench-problem-1/problem.py
Invoke the Claude Code plugin to generate an optimized solution
Output the solution and evaluation results

Example Problems¶

kernelbench-problem-1¶

A GPU kernel optimization problem from the KernelBench benchmark suite.

examples/kernelbench-problem-1/
  problem.py    # Problem definition and reference implementation

kernelbench-problem-2¶

Another KernelBench problem with different optimization characteristics.

examples/kernelbench-problem-2/
  problem.py    # Problem definition and reference implementation

Creating Your Own Problems¶

To create a new problem, go to the Problem Format section.