# Multi Agent Council

> Facilitate a requested multi-model council or structured debate, collecting independent evidence, challenging assumptions, and recording unresolved dissent before the user decides.

- Skill: `ktaey7/multi-agent-council` (Agent Skill, multi-file: 3 files)
- Install (CLI): `npx skillmds@latest add ktaey7/multi-agent-council`
- Raw SKILL.md: https://api.skillmd.com/api/skills/ktaey7/multi-agent-council/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: ktaey7 (https://skillmd.com/u/ktaey7)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/ktaey7/multi-agent-council

---


# Multi-Agent Council

Use the [council protocol](references/protocol.md) for the requested review.
Resolve reference paths relative to this installed SKILL.md, never the user's
project directory. The protocol and all required prompts ship with this skill.

Council defines how reviewers reason together. It does not install a runtime,
change model settings, or grant permission to execute changes.

Choose the available transport within the user's authorized scope:

- **Manual:** prepare independent prompts from the protocol for the user to
  paste into their chosen AI apps, then compare the returned reviews.
- **Crosscrew (optional):** if already installed and the user wants automated
  delegation, read [the Crosscrew guide](references/crosscrew.md). Use its current
  calling card and existing native CLI authentication.
- **Another existing tool:** use its documented invocation and permission
  contract. Do not invent executable names or claim unavailable integrations.

Do not run repository-relative `scripts/council-runner.sh` from an installed
skill. The repository's old direct-CLI runner is a legacy experiment, not this
skill's execution backend.

At least two external reviewers plus the host are required for a formal council.
If fewer are available, explain the limitation and offer a smaller cross-review
or self-review under that name; do not simulate independent models.