Tribunal Multi-Agent Debate FAQs

Question 1

How does the user interact with the debate results?

Accepted Answer

The orchestrator presents a summary of all agent rankings, trade-offs, and blocking concerns. The user then makes the final decision to proceed with the recommended approach, pick an alternative, or request an additional round of debate.

Question 2

When does the debate phase occur in the development workflow?

Accepted Answer

The debate typically triggers automatically after the research phase and must be completed before the design review gate. This ensures that the implementation plan has been vetted by multiple perspectives before coding begins.

Question 3

What is the Tribunal Debate skill?

Accepted Answer

It is a specialized Claude Code skill that facilitates a structured, multi-agent brainstorming session where different AI personas—such as an Architect, Security Expert, or CTO—propose and critique technical solutions to ensure the best approach is chosen.

Question 4

What happens if one of the AI agents fails during the debate?

Accepted Answer

The system is designed for resilience; it will proceed with the remaining successful agents as long as a minimum of two agents are functional, ensuring the workflow isn't stalled by a single agent error.

Question 5

Can I customize which agents participate in the debate?

Accepted Answer

Yes. You can configure the specific agent roles (e.g., [architect, security, mobile_expert]) and the number of debate rounds within a tribunal.yaml file in your project root.

Tribunal Multi-Agent Debate

Tribunal Multi-Agent Debate

主な機能

ユースケース

主な機能

ユースケース