panel-debate（小组辩论）

Panel Discussion Skill

Generate diverse expert personas, facilitate structured debate, and synthesize actionable recommendations.

Arguments

Parse $ARGUMENTS:

• size:N → panel size (3-7, default: auto based on breadth)
• depth:X → quick (1 round), standard (2-3), deep (4+)
• style:X → collaborative, adversarial, academic
• Remaining text → topic

Workflow Progress

Copy and update this checklist as you proceed:

- [ ] Complexity scored (5-15)
- [ ] Experts generated (diversity score ≥60/85)
- [ ] Opening round complete
- [ ] Cross-examination complete
- [ ] Synthesis generated
- [ ] Final report produced

Flow

COMPLEXITY_CHECK → EXPERT_GENERATION → DISCUSSION → SYNTHESIS → REPORT
      ↓                                     ↑
[5-7: Warn+Offer direct answer]      [Multiple rounds]

Complexity Check

Score the topic on 5 dimensions (1-3 each):

Total = sum of 5 dimensions (range 5-15)

Expert Generation

Load detailed algorithms: Read references/expert-generation.md

Required archetypes (every panel):

1. Contrarian - challenges consensus, offers alternatives
2. Synthesizer - connects perspectives, finds common ground
3. Specialist - deep domain expertise, grounds discussion

Additional: Optimist, Skeptic, Pragmatist, Theorist

Panel size by breadth:

• Narrow (single tech): 3-4 experts
• Medium (cross-functional): 4-5 experts
• Broad (strategic): 5-7 experts

Diversity score ≥60 required (additive, max 85):

• All 3 required archetypes present: +20
• No single archetype >30% of panel: +10
• 4+ distinct archetypes: +10
• 2+ knowledge domains: +15
• Optimist + Skeptic both present: +15
• User/external perspective included: +15

Validation: After generating experts, compute diversity score. If <60, regenerate or add experts until threshold met. Do NOT proceed with score <60.

Discussion Phases

Load turn-taking mechanics: Read references/turn-taking.md

Phase 1: Opening (Thesis)

🎤 Expert (Role): "[Initial position, 3-5 sentences]"

Phase 2: Cross-Examination (Antithesis)

Experts challenge and build. Patterns: Question→Response, Claim→Counter-claim, Challenge→Defense.

Contrarian protection: Before any synthesis, ask: "Before we synthesize, [Contrarian], what are we missing?"

Phase 3: Synthesis

Load synthesis patterns: Read references/synthesis-patterns.md

📋 Round N Synthesis:
   • Agreement: [point]
   • Tension: [disagreement]
   • Open question: [needs exploration]

Convergence: End early if all agree or cycling. Extend if major tension unexplored.

Validation: After synthesis, verify no "it depends" statements appear without specifying the context factors it depends on. If found, rewrite with decision criteria.

Output Formats

Load formatting specs: Read references/output-formats.md

Panel Header

╭─ Panel Discussion: [Topic] ────────────────────────────────╮
│ Experts: [Name] ([Role]), [Name] ([Role]), [Name] ([Role]) │
╰────────────────────────────────────────────────────────────╯

User Menu

╭───────────────────────────────────────────────────────────╮
│ [1] Continue  [2] Follow-up  [3] Redirect  [4] Conclude   │
╰───────────────────────────────────────────────────────────╯

Final Report

╔═══════════════════════════════════════════════════════════╗
║                    PANEL CONCLUSIONS                       ║
╚═══════════════════════════════════════════════════════════╝

## Executive Summary
[1-2 sentences]

## Consensus Points
1. **[Point]**: [Details]

## Key Trade-offs
- Trade-off: [Description]
- Recommendation: [Action]
- Dissent: [If any]

## Actionable Recommendations
### Immediate (Week 1)
### Short-term (Month 1)

## Dissenting Views
### [Expert]: "[Direct quote]"

## Open Questions

Synthesis Rules

Labels: UNANIMOUS | STRONG | MAJORITY | CONTESTED | CONTEXT-DEPENDENT

Weight by confidence x domain relevance:

• High confidence + Core domain: 1.0
• Medium confidence + Adjacent domain: 0.49
• Low confidence + Outside domain: 0.16

Never say: "Both make valid points" or "It depends" without specifying decision factors.

User Commands