Agent Council

Agent Council is a skill for agentic coding tools (like Claude Code) that lets you convene a multi-agent deliberation session. It allows you to summon distinct engineering personas to brainstorm, debate, or collaborate on a task. Each persona has genuine technical biases, experience levels, and collaboration styles that create productive tension — just like a real engineering team.

Installation

You can install the Agent Council skill intuitively via our CLI installer using npx:

npx agent-council

The intuitive installer will ask where you'd like to place the skill:

• Local Project (.claude/skills or .agents/skills)
• Global Installation (~/.claude/skills or ~/.agents/skills)

Usage

Once installed, simply type /council followed by your prompt or command in your AI terminal:

/council [mode] "<topic>" [options]

Modes

| Mode | Description | Default Rounds | |------|-------------|----------------| | brainstorm (default) | Additive — agents build on each other's ideas | 1 | | debate | Adversarial — agents challenge each other's positions | 2 | | collaborate | Productive — agents work on complementary sub-tasks in parallel | 1 |

Options

All options can be provided in natural language:

• with <persona1>, <persona2>, ... — specify which personas to include
• <N> rounds — number of interaction rounds
• output <format> — hint for the requested output format ("plan", "code", "recommendation", "diagram")
• Custom personas can be defined inline: with a "Graph Database Expert" who loves Neo4j

Examples

/council "How should we architect the notification system?"
/council debate "monolith vs microservices for our scale" with architect, pragmatic_generalist, backend_engineer
/council collaborate "implement the auth system" with backend_engineer, frontend_engineer, cybersec_engineer
/council brainstorm "improve our CI/CD pipeline" 2 rounds with devops_engineer, pragmatic_generalist, architect

See Agents.md for a full list of available personas.