What is a design tournament in Claude Code?

A design tournament is a structured workflow where multiple AI agents independently create design proposals for a specific problem, followed by a ranking phase where separate agents evaluate and score them to find the best technical approach.

Can I merge ideas from different proposals?

Yes, the final output includes a 'Cherry-pick list' which identifies specific strengths and unique insights from runners-up to be incorporated into the winning design.

How many agents are involved in the tournament process?

By default, the skill utilizes 5 explorer agents to generate diverse designs and 2 ranking agents to evaluate them, though you can specify a different number of explorers (minimum 3).

Does the skill have access to my existing code?

Yes, the explorer agents can use tools like Glob, Grep, and Read to gain context from your codebase, ensuring that their proposals are feasible and integrated with your current system.

How are the final design scores calculated?

Scores are based on a weighted average of criteria including Correctness (25%), Simplicity (20%), Robustness (20%), Extensibility (15%), Specificity (10%), and Trade-off Awareness (10%).

Design Tournament

Name: Design Tournament
Author: ahrav

byahrav

•

Productividad y Flujo de Trabajo

Orchestrates a multi-agent "wisdom of crowds" process to generate, evaluate, and rank diverse software design proposals.

The Design Tournament skill automates a sophisticated two-phase architectural review process within Claude Code. It first launches multiple independent subagents to explore a problem space from different angles without cross-contamination, ensuring a wide range of technical perspectives. In the second phase, two independent ranking agents evaluate these proposals against rigorous criteria—such as simplicity, robustness, and correctness—to produce a confidence-weighted stack ranking and a synthesized final recommendation. This approach is ideal for critical technical decisions where reducing blind spots and exploring trade-offs is essential for high-quality software engineering.

Características Principales

01Confidence-weighted stack rankings to identify clear winners or close ties

021 GitHub stars

03Comprehensive markdown reporting with full summaries and collapsed technical details

04Parallel agent execution for diverse, non-contaminated architectural exploration

05Independent ranking phase using standardized evaluation criteria and weighted scoring

06Automated synthesis of results including 'cherry-picking' the best ideas from all agents

Casos de Uso

01Breaking down large software requirements into optimal task decompositions through parallel analysis

02Evaluating competing architectural patterns for new features to minimize technical debt

03Designing complex data structures or internal system APIs where multiple valid approaches exist

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ahrav/gossip-rs design-tournament

For use in Claude.ai and ChatGPT