Analysis updated 2026-05-18
Get a plan reviewed by multiple AI coding agents before acting on it.
Have agents debate a technical decision like a framework migration for several rounds.
Audit what each agent actually said by reading the saved deliberation transcript.
Catch a risky plan early since disagreeing agents can flag ship with changes or rethink.
| kitze/council | aerdelan/housand-domaintoolmatrix | affaan-m/opencode | |
|---|---|---|---|
| Stars | 22 | 22 | 22 |
| Language | — | TypeScript | TypeScript |
| Last pushed | — | — | 2026-02-09 |
| Maintenance | — | — | Maintained |
| Setup difficulty | easy | easy | easy |
| Complexity | 2/5 | 2/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Needs at least one other agent CLI besides your main one already installed and logged in.
council is an agent skill, a set of instructions you install so that a coding AI agent follows a specific process instead of improvising. Its purpose is to stop a single AI agent from giving you a plan based only on its own judgment. Instead, before answering, the agent you are talking to must call on the other agent command line tools already installed on your machine, such as Claude Code, Codex, Cursor, Gemini, Grok, or OpenCode, and have them weigh in. The way it works is simple to describe. You tell your agent something like council, 2 turns, should we migrate this app from Express to Hono. The lead agent, whichever one you are using, drafts a plan and shows it separately to each of the other installed agents, without letting them see each other's answers. Each one replies once per turn with a verdict of ship it, ship with changes, or rethink, leading with its strongest objection. The lead agent then revises the plan based on that feedback and repeats the process for however many turns you asked for, finally showing you the full back and forth conversation along with the final plan. Under the hood, every step is a plain one shot command line call to each agent tool. There is no background daemon and no extra API keys beyond the agent CLIs you already have installed and signed into. Every agent that gets consulted is only allowed to read your repository, never modify it, and if a particular agent tool is missing or fails to respond, it is simply skipped and noted in the transcript rather than causing an error. To install it, you clone the single SKILL.md file into the folder where your agent looks for skills, either for your personal setup or for a specific project. You need at least one other agent CLI installed and logged in besides the one giving you the final answer, and having more of them installed makes the council bigger. Every deliberation, including the draft plans and each member's replies, is saved locally so you can review exactly what each agent said. The project is released under the MIT license.
An agent skill that makes your coding AI consult other installed agent CLIs and debate a plan for several turns before giving you a final answer.
MIT license: free to use, modify, and distribute, including commercially, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.