roundtable

name	roundtable
description	Virtual expert roundtable. Multiple thinkers analyze a document with different frameworks, then synthesize consensus/disagreements/blind spots.
argument-hint	[file_path] [--experts rumelt,meadows,taleb,grove] [--rounds 1]
allowed-tools	["Bash","Read","Write","Glob","Grep","Agent"]
author	Maki
version	1.0.0
tags	["analysis","decision-making","strategy","multi-perspective"]
required_env	[]
required_bins	[]
platforms	["macos","linux","windows"]
safety_level	safe

Virtual Expert Roundtable

Analyze a document through multiple analytical frameworks, producing a structured report of consensus, disagreements, blind spots, and emergent insights.

Architecture: Supervisor + N Agents + Synthesizer

Input Document
  │
  ▼
Stage 1: Independent Expert Analysis (parallel)
  ├── Expert A ──→ Structured analysis + confidence score
  ├── Expert B ──→ Structured analysis + confidence score
  ├── Expert C ──→ Structured analysis + confidence score
  └── Expert D ──→ Structured analysis + confidence score
  │
  ▼
Stage 2: Cross-Critique (optional, --rounds N)
  Each expert sees others' conclusions, points out disagreements and self-critiques
  │
  ▼
Stage 3: Roundtable Synthesis (Moderator)
  Claude as Moderator integrates all perspectives
  │
  ▼
Stage 4: Blind Spot Detection + Emergent Insight Extraction
  Dedicated prompts find hidden assumptions and cross-expert inferences
  │
  ▼
Output: Structured Markdown Report

Flow

Step 0: Parse Input

Read arguments:

• file_path: Document to analyze (required, or take from conversation context)
• --experts: Expert combination (default rumelt,meadows,taleb,grove)
• --rounds: Cross-critique rounds (default 0; set 1-2 for deeper but slower analysis)

Read the file. If over 3000 words, auto-summarize to under 3000 words for analysis input.

Step 1: Independent Expert Analysis

Select experts from the roster (see below), launch analysis requests in parallel using Agent subagents.

Each expert's prompt structure:

You are {expert_name}, {expert_identity}.
Your analytical framework is: {framework_description}.

Analyze the following document using your framework. Reply in strict JSON:

Document:
---
{document_content}
---

Reply format:
{
  "key_observations": ["Observation 1", "Observation 2", ...],
  "risks_and_concerns": ["Risk 1", "Risk 2", ...],
  "recommendations": ["Recommendation 1", "Recommendation 2", ...],
  "confidence": <1-5 integer, confidence in your analysis>,
  "one_line_verdict": "One-sentence summary judgment"
}

Step 2: Cross-Critique (optional)

If --rounds >= 1, run cross-critique:

You are {expert_name}. Below are other experts' analyses of the same document:

{other_experts_summaries}

Please:
1. Identify your top 3 disagreements with other experts
2. List where you might be wrong (self-critique)
3. After seeing others' reasoning, update your conclusions and confidence score

Reply JSON:
{
  "top_disagreements": ["Disagreement 1", ...],
  "self_critique": ["Possible error 1", ...],
  "updated_recommendations": ["Updated recommendation 1", ...],
  "updated_confidence": <1-5>
}