Which discussion mode should I choose?

Use 'Deep' for high-stakes, unprecedented architectural decisions; 'Standard' for typical design trade-offs; and 'Lightweight' for quick sanity checks and idea validation.

How is the final result of the discussion delivered?

The Historian role synthesizes the entire debate into a final report, including an argument graph, a summary of evidence, and a record of why specific experts shifted their positions.

What are 'Dynamic Experts' in this skill?

The skill analyzes your specific problem and generates custom AI personas—such as a 'Database Pragmatist' or 'Cloud Architect'—with specific biases and expertise relevant to your unique topic.

How does Swarm Discussion differ from a standard Claude response?

Standard responses often seek quick consensus; Swarm Discussion deliberately introduces tension and forces agents to take opposing sides and 'steel-man' arguments to reveal hidden risks and blind spots.

Can I participate in the AI discussion?

Yes, the skill uses the AskUserQuestion interface at key milestones to allow you to confirm expert composition, steer the direction of the debate, or provide additional context.

Swarm Discussion

Name: Swarm Discussion
Author: Ischca

byIschca

0•

Collaboration Tools

Replicates Staff+ engineer decision-making through structured, multi-agent expert debates that expose blind spots and pressure-test complex technical solutions.

Swarm Discussion is a sophisticated multi-agent orchestration skill designed to solve high-stakes, unsolved technical problems by simulating a diverse team of experts. Unlike standard AI responses that often converge quickly on the first plausible answer, this skill uses a Structured Disagreement Protocol and a Tension Map to force agents into meaningful, adversarial debate. By assigning specialized roles—such as the Moderator, Contrarian, and Historian—alongside dynamically generated domain experts, the skill builds a comprehensive argument graph and tracks position shifts in real-time. This ensures every proposal is rigorously stress-tested and every trade-off is documented, mirroring the collaborative deep-dive thinking of senior engineering leadership teams.

Key Features

010 GitHub stars

02Adaptive discussion modes (Deep, Standard, Lightweight) to balance depth and resource usage

03Dynamic expert generation with specific biases, stakes, and blind spots based on topic

04Multi-agent team architecture utilizing Claude Code's Teammate and Messaging APIs

05Real-time argument graph construction to track citations, rebuttals, and position shifts

06Structured Disagreement Protocol including mandatory position declarations and steel-manning

Use Cases

01Evaluating technical trade-offs where no industry standard or clear 'right' answer exists

02Validating high-risk security or infrastructure strategies through structured adversarial debate

03Architectural decision-making for complex distributed systems or microservices

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ischca/swarm-discussion-skill swarm-discussion

For use in Claude.ai and ChatGPT

Download Skill