A Claude Code skill that runs a structured multi-agent debate on a topic. Specialized agents take positions, respond to each other across rounds, and surface insights through intellectual friction. Two modes:
- DEBATE — 3-round structured debate (positions → responses → synthesis), full transcript
- QUICK — single-round perspective check, fast
Best decisions don't come from one model thinking harder. They come from several models with different priors taking honest positions and having to defend them against each other. Council makes that explicit and visible — you read the transcript, not just the verdict.
Copy the skill folder into your Claude Code skills directory:
git clone https://github.com/goneil78-coder/council_skill.git ~/.claude/skills/CouncilOr, if you keep skills somewhere else, drop the folder there and Claude Code will pick it up.
That's it. No external dependencies. The skill spawns agents using Claude Code's built-in Agent tool.
Trigger by mentioning council in the prompt:
Council: Should we use WebSockets or SSE for the live transcript feature?
Quick council check: Is this API design reasonable?
Council with security: Evaluate this auth approach
The skill picks the workflow based on phrasing. Quick keyword → 1-round QUICK. Otherwise → 3-round DEBATE.
| File | What it does |
|---|---|
SKILL.md |
Entry point Claude Code reads. Routes to a workflow. |
CouncilMembers.md |
Default agent roles (architect, designer, engineer, researcher) and their voice/personality |
RoundStructure.md |
The 3-round debate structure |
OutputFormat.md |
Transcript format templates |
Workflows/Debate.md |
Full debate workflow |
Workflows/Quick.md |
Fast single-round workflow |
The skill works fine on a single default model. If you have a model-tier router (something that returns haiku/sonnet/opus per task), SKILL.md shows where to call it. Cheap models for fast positions, heavier models for synthesis. Skip the section if you don't have one — the skill doesn't need it.
Want a security expert in the council? Add them to CouncilMembers.md and reference them in your prompt (Council with security: ...). Same for any other role — domain expert, product manager, your own caricatures of people whose judgment you trust.
The default four are a starting point, not a constraint.
This is a port of Daniel Miessler's Council skill from his Personal AI Infrastructure (PAI) v2.5. The structure, round logic, agent personalities, and core mechanics are his design. This repo packages it for friends who want to drop it into their own Claude Code setup without copying the whole PAI system.
- Original: danielmiessler/PAI
- License: MIT (see
LICENSE)
If you find this useful, the PAI repo has a lot more.
MIT — see LICENSE.