What is the purpose of the Cookoff skill?

It allows multiple AI agents to implement the same design in parallel, creating diverse solutions so you can pick the best-performing and most robust code.

Why does each agent create its own implementation plan?

Independent planning ensures genuine variation in code structure and logic, preventing agents from simply copying the same approach.

How are the different implementations compared?

The skill uses a judging framework that evaluates test pass rates, design adherence, and a 'fresh-eyes' quality review to score each version.

Does Cookoff handle cleanup of failed implementations?

Yes, once a winner is selected, the skill handles the removal of temporary worktrees and branches for the losing implementations.

Cookoff Implementation Strategy

Name: Cookoff Implementation Strategy
Author: 2389-research

by2389-research

•

생산성 및 워크플로

Orchestrates multiple parallel AI agents to compete on implementing the same design doc to ensure high-quality and varied code solutions.

소개

The Cookoff skill manages the critical transition from design to implementation by spawning multiple parallel subagents to tackle the same feature. Unlike standard workflows, each agent is required to generate its own unique implementation plan from a shared design document, fostering architectural variation and surfacing different approaches. The skill assesses complexity to determine implementation depth, manages isolated git worktrees for each team, and utilizes a rigorous judging framework to select the winning implementation based on test results, design adherence, and security reviews.

주요 기능

Isolated Git Worktrees: Maintains clean environments for each implementation attempt to allow side-by-side comparison.
Independent Implementation Planning: Prevents identical code by requiring each agent to draft its own distinct plan.
Automated Complexity Mapping: Dynamically scales the number of agents based on feature scope and risk signals.
Parallel Agent Competition: Dispatches multiple agents simultaneously to implement a single feature design.
8 GitHub stars
Multi-Step Judging Framework: Employs a scoring system with 'fresh-eyes' reviews to identify the most robust solution.

사용 사례

Reducing single-agent bias by comparing multiple independent implementations of the same design.
Implementing high-stakes features where diverse architectural approaches need to be explored.
Accelerating development speed by running multiple implementation paths in parallel branches.

소개

주요 기능

Isolated Git Worktrees: Maintains clean environments for each implementation attempt to allow side-by-side comparison.
Independent Implementation Planning: Prevents identical code by requiring each agent to draft its own distinct plan.
Automated Complexity Mapping: Dynamically scales the number of agents based on feature scope and risk signals.
Parallel Agent Competition: Dispatches multiple agents simultaneously to implement a single feature design.
8 GitHub stars
Multi-Step Judging Framework: Employs a scoring system with 'fresh-eyes' reviews to identify the most robust solution.

사용 사례

Reducing single-agent bias by comparing multiple independent implementations of the same design.
Implementing high-stakes features where diverse architectural approaches need to be explored.
Accelerating development speed by running multiple implementation paths in parallel branches.