Skip to content

microsoft/model-council

v3.0.0MIT

Multi-model council for code review and planning — model-pinned subagents independently review or plan, then findings are synthesized by consensus.

model-council

Multi-model council for code review and planning. Independent model-pinned agents inspect the same change set or plan the same task, then findings are synthesized by consensus — highlighting where models agree, where they disagree, and optionally running a debate to resolve contested findings.

Skills

SkillDescription
council-reviewOrchestrate a multi-model council review: fan out to pinned reviewers, synthesize by consensus vs. contested findings, and optionally cross-review to resolve disagreements
council-planMulti-model council planning: each agent independently proposes an implementation plan, then proposals are synthesized into a consensus plan with alternatives and risks

How It Works

  1. Independent reviews — each model-pinned subagent receives the same scope and reviews in parallel without seeing each other's findings.
  2. Consensus synthesis — findings are classified as consensus (multiple models agree), single-reviewer (one model, strong evidence), or contested (models disagree).
  3. Cross-review debate (optional) — when contested findings exist or the user asks to "discuss" / "debate", the anonymized synthesis is shared back to each agent for a second-look deliberation round.

Notes

  • Council subagents are dispatched by the council-review and council-plan skills using the model parameter — no custom agent files required.
  • Copilot model labels can change over time. You might need to update the model names in the skill files to match the names in your local model picker.

Plugin Structure

model-council/
├── plugin.json
├── README.md
└── skills/
    ├── council-plan/
    │   └── SKILL.md
    └── council-review/
        └── SKILL.md