What kind of output can I expect from the council?

You will receive a structured verdict (ranging from Proceed to Do Not Proceed), a risk assessment matrix, a summary of consensus/dissensus points, and a steelmanned version of the strongest counter-argument.

What are the different roles in the AI Council?

The council can include a Moderator, Visionary, Critic, Pragmatist, Innovator, Ethicist, and specialized Domain Experts depending on the configuration.

How long does a debate typically last?

By default, the council runs for 2 rounds of cross-examination, but this can be adjusted between 1 and 4 rounds using the -r flag.

Does this skill require specific Claude Code settings?

Standard mode requires Agent Teams to be enabled via the CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable. However, an Economy Mode (-e) is available for sequential execution.

What is the difference between workflow-debate and workflow-brainstorm?

Brainstorming is for divergent idea generation, while debate is for convergent evaluation. Use debate when you have a specific plan that needs to be challenged and stress-tested.

AI Council & Workflow Debate

Name: AI Council & Workflow Debate
Author: Devattom

byDevattom

0•

生产力与工作流

Orchestrates structured adversarial debates between specialized AI personas to rigorously stress-test ideas, architectures, and strategic decisions.

The Workflow Debate skill leverages the 'AI Council' framework to provide multi-perspective analysis of complex proposals. Unlike standard brainstorming, it utilizes structured adversarialism where specialized agents—including a Visionary, Critic, Pragmatist, and Ethicist—engage in direct cross-examination and peer messaging. This process surfaces blind spots, identifies hidden risks, and culminates in a nuanced verdict (PROCEED, RECONSIDER, or DO NOT PROCEED) supported by a risk assessment matrix, helping developers and leads make more informed technical and strategic choices.

主要功能

01Multi-persona adversarial debate (Visionary, Critic, Pragmatist, Innovator, Ethicist, Expert)

02Comprehensive synthesis including risk matrices and steelmanned counter-arguments

030 GitHub stars

04Economy mode for sequential processing when Agent Teams are not enabled

05Configurable intensity with adjustable debate rounds (1-4) and councillor counts (3-6)

06Direct peer-to-peer agent messaging for authentic cross-examination

使用场景

01Stress-testing security protocols and PII handling strategies before implementation

02Evaluating major architectural shifts like migrating from monolith to microservices

03Conducting rigorous 'pre-mortems' on high-stakes strategic or technical plans

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add devattom/.claude workflow-debate

For use in Claude.ai and ChatGPT