Agent Skills › nicknisi/claude-plugins › deliberate

deliberate

GitHub

通过四轮多智能体辩论（独立分析、相互批判、修正立场、最终综合）对复杂议题进行深度探讨，旨在通过同行评审机制发现盲点、达成共识并评估置信度。

plugins/consultant/skills/deliberate/SKILL.md nicknisi/claude-plugins

Trigger Scenarios

需要多角度深入分析的技术选型决策存在争议或需权衡利弊的架构设计问题希望获得经过批判性思维优化的复杂建议

Install

npx skills add nicknisi/claude-plugins --skill deliberate -g -y

More Options

Non-standard path

npx skills add https://github.com/nicknisi/claude-plugins/tree/main/plugins/consultant/skills/deliberate -g -y

Use without installing

npx skills use nicknisi/claude-plugins@deliberate

指定 Agent (Claude Code)

npx skills add nicknisi/claude-plugins --skill deliberate -a claude-code -g -y

安装 repo 全部 skill

npx skills add nicknisi/claude-plugins --all -g -y

预览 repo 内 skill

npx skills add nicknisi/claude-plugins --list

SKILL.md

Frontmatter

{
    "name": "deliberate",
    "description": "Have AI agents debate a topic through multiple rounds of critique and refinement",
    "disable-model-invocation": true
}

Deliberation Mode

Agents critique each other's answers and refine through peer review.

Usage

/consultant:deliberate Should we use REST or GraphQL for our API?
/consultant:deliberate What's the best state management approach for this app?

How It Works

Deliberation runs in 4 rounds:

Round 1: Initial Perspectives

Launch all 5 agents in parallel
Each agent provides initial analysis independently

Round 2: Critique & Challenge

Share all Round 1 responses with all agents
Each agent reviews others' answers, points out errors/gaps/strengths

Round 3: Refinement

Agents revise positions based on feedback
Identify emerging consensus

Round 4: Final Synthesis

Analyze convergence vs divergence
Highlight consensus points
Present unresolved disagreements with reasoning
Rate confidence based on agent agreement

Execution

Round 1: Launch 5 agents in parallel using Task tool:

consultant:codex-researcher
consultant:gemini-researcher
consultant:grok-researcher
consultant:perplexity-researcher
consultant:claude-researcher

Prompt each: "Provide your initial analysis on: $ARGUMENTS"

Round 2: Launch 5 agents again with all Round 1 responses. Prompt: "Review these perspectives and critique them. Point out errors, gaps, and strong points."

Round 3: Launch 5 agents with all prior responses. Prompt: "Based on the critiques, refine your position. Acknowledge valid points, defend or modify your stance."

Round 4: Synthesize in the main session — identify convergence, note disagreements, provide confidence rating.