Is there a limit to how many approaches I can test?

The skill recommends a limit of 2-5 variants to prevent resource exhaustion and ensure that the judging phase remains focused and manageable.

What happens to the code of the losing variants?

The skill includes an automated cleanup phase that removes the temporary Git worktrees and deletes the branches associated with the discarded implementations.

What is the 'Any%' approach in coding?

Inspired by speedrunning, Any% is a strategy focused on reaching a functional implementation via the fastest path possible, prioritizing parallel experimentation over upfront deliberation.

How does this skill help save tokens?

It utilizes specialized MCP tools to call hosted LLMs for bulk code generation, reserving Claude's primary context for high-level coordination and surgical fixes.

Speed-Run: Any% Architectural Exploration

Name: Speed-Run: Any% Architectural Exploration
Author: 2389-research

by2389-research

•

Productivity & Workflow

Explores and implements multiple architectural approaches in parallel using hosted LLMs to identify the optimal solution through real-world testing.

Speed-Run Any% is a high-velocity development skill for Claude Code designed to eliminate architectural indecision. Instead of lengthy deliberation, this skill dispatches multiple parallel agents to implement different design variants—such as SQL vs. NoSQL or competing framework choices—using hosted LLMs for token-efficient code generation. By leveraging Git worktrees and an automated judging framework, it allows developers to compare functional implementations side-by-side, selecting a winner based on performance and code quality while automatically pruning discarded approaches.

Key Features

01One-click cleanup of rejected branches and temporary worktrees

02Parallel execution of up to 5 architectural variants using isolated agents

03Standardized scoring and judging framework for objective comparison

04Automated Git worktree management for clean, concurrent development

05Token-efficient code generation via hosted LLM integration

0623 GitHub stars

Use Cases

01Comparing multiple database engines or state management strategies in real-time

02Prototyping a feature across different frameworks to evaluate DX and performance

03Rapidly exploring alternative API structures when the optimal path is unclear

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add 2389-research/claude-plugins any-percent

For use in Claude.ai and ChatGPT

Download Skill