What is the main benefit of the Showdown skill?

It allows you to explore multiple implementation paths simultaneously, ensuring the highest quality code while significantly reducing costs by using high-speed hosted LLMs for bulk generation.

Does it clean up the extra branches it creates?

Yes, once a winner is selected and the code is merged, the skill automatically removes the associated Git worktrees and deletes the branches of the losing runners.

How is the winner of a showdown determined?

A structured judging phase evaluates implementations based on passing test counts, design adherence, code quality (via fresh-eyes review), and token efficiency metrics.

Which LLMs are used for code generation?

The skill is optimized to use high-throughput hosted LLMs like Cerebras for initial code generation, while Claude is used for surgical fixes, reviews, and overall coordination.

How does the skill ensure implementations aren't identical?

Each runner is required to create its own independent implementation plan from the design document before generating code, which introduces natural variation in logic and architecture.

Speed-Run Showdown

Name: Speed-Run Showdown
Author: 2389-research

by2389-research

•

生産性とワークフロー

Orchestrates multiple parallel AI runners to generate and compete on independent code implementations from a single design document.

Speed-Run Showdown is a sophisticated workflow skill for Claude Code that maximizes code quality and token efficiency by spawning multiple independent runners to implement the same technical design. By requiring each runner to generate its own unique implementation plan, the skill surfaces diverse architectural approaches and logic patterns. It leverages high-speed hosted LLMs for bulk code generation, applies rigorous Test-Driven Development (TDD) and 'fresh-eyes' reviews, and utilizes a structured judging framework to select the most robust, efficient, and maintainable implementation for the final codebase.

主な機能

01Automated complexity assessment to determine the optimal number of competing agents

02Structured judging framework with standardized scoring worksheets and metrics

0323 GitHub stars

04Parallel execution of multiple independent implementation runners using Git worktrees

05Cost-efficient first-pass code generation via high-speed hosted LLMs like Cerebras

06Automated cleanup of losing branches and temporary worktree environments

ユースケース

01Exploring multiple architectural approaches for complex feature implementation

02Reducing token expenditure on large-scale boilerplate and algorithmic code generation

03Benchmarking different implementation strategies to find the most performant solution

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add 2389-research/claude-plugins showdown

For use in Claude.ai and ChatGPT

Download Skill