Expert debate

Skill wenboxia/expert-debate/skills/expert-debate

Adversarial brainstorm skill — two AI models generate proposals as conflicting expert roles. Part of the Liangyi skill family.

Install

npx -y skills add wenboxia/expert-debate --skill expert-debate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Adversarial expert brainstorm using conflicting role personas across two AI models. Use when brainstorming a product idea, evaluating a technical direction, or when the user wants multiple expert perspectives on any decision. Triggers on "brainstorm", "expert debate", "role debate", "get different perspectives", "challenge this idea".

SKILL.md

5.6 KB, as published. Nobody here has run it

Expert Debate

Generate structured adversarial brainstorms by assigning conflicting expert personas to two AI models. Each expert brings real professional biases, not generic feedback.

Companion skill: Best Minds

This skill pairs powerfully with Best Minds (npx skills add Ceeon/best-minds). Best Minds activates the "simulator thinking" pattern — instead of asking "what do you think?", ask "who in the world knows this best? What would THEY say?" If Best Minds is installed, this skill will use its trigger pattern to deepen the expert persona activation.

Model compatibility

This skill runs in Claude Code (Model A = Claude). For Model B, Gemini is recommended, but any capable LLM works — GPT, Qwen, Doubao, DeepSeek, etc. Use a different provider so training biases don't overlap. "Gemini" below is shorthand for your chosen Model B.

Quick start

/expert-debate "Build a payment retry system with smart scheduling"

How it works

You describe your idea or proposal
This skill helps you pick two expert roles with structural tension
Claude generates a full proposal as Role A (with deliberate professional bias)
A ready-to-copy prompt is generated for Gemini as Role B (independent, no cross-contamination)
You compare both proposals, logging each decision point

Step 1: Select expert role pair

Read the role reference:

cat references/expert-roles.md

Help the user choose two roles. Selection rules:

Tension test: If both roles would give the same top-3 recommendations, insufficient tension. Pick another pair.
Relevance test: Both roles must have legitimate authority over the domain.
Coverage test: Between the two, most critical dimensions (technical, market, user, financial) should be covered.

If the user's idea doesn't clearly fit a category, use a universal high-tension pair (e.g., Investor vs Senior Engineer, or User Advocate vs Competitive Analyst).

Step 2: Generate proposal as Role A

Adopt the selected expert persona fully. If Best Minds skill is installed, activate it with: "Who in the world is the best [Role A]? Think as that person."

Instructions:

【Best Minds mode activated】

You are now [Role A]. You have 20 years of experience in this domain.
Your thinking, priorities, and risk tolerance must match this role exactly.

Rules:
- When your role's judgment conflicts with "generic PM advice", follow your role
- Mark such conflicts explicitly: "As a [Role A], I prioritize X over Y because..."
- Preserve your role's blind spots — do not self-correct them
- Ask the user clarifying questions when genuinely uncertain

Generate a complete proposal:

Core value proposition (from this role's lens)
Feasibility assessment (only dimensions this role prioritizes)
Top 3 concerns (what keeps this role up at night)
Alternative approaches (at least 2)
Recommended final shape

Step 3: Generate Gemini prompt

Create a ready-to-copy prompt for a separate, clean Gemini session.

The prompt must:

Assign Role B persona with the same depth as Role A
Include the user's refined idea
Include the Skills registry (read from references/skills-registry.md if configured)
NOT include any of Claude's Role A output
End with: "Give me your complete, independent proposal."

Save to prompts/gemini-expert-debate.md.

After generating, automatically copy to clipboard:

cat prompts/gemini-expert-debate.md | pbcopy 2>/dev/null || cat prompts/gemini-expert-debate.md | xclip -selection clipboard 2>/dev/null || echo "Prompt saved to prompts/gemini-expert-debate.md — please copy manually"

Tell the user: "Prompt copied to clipboard. Paste it into a new Gemini session."

Step 4: Compare and log decisions

After the user provides Gemini's output:

Present a side-by-side comparison of key divergence points
For each divergence, ask: "Which direction do you lean, and why?"
Log to decision-log.md:

### Expert Debate: [topic]
**Date:** [auto]
**Role A ([name]):** [core position]
**Role B ([name]):** [core position]
**Divergence points:**
1. [topic]: A says X, B says Y
2. [topic]: A says X, B says Y

**Decision 1:** Chose [A/B/hybrid] because [human reasoning]
**Decision 2:** Chose [A/B/hybrid] because [human reasoning]
**What AI couldn't weigh:** [the human context that drove the choice]

Output files

File	Purpose
`prompts/gemini-expert-debate.md`	Ready-to-paste Gemini prompt
`decision-log.md`	Human decision record (append mode)

Tips

For quick decisions, skip Gemini and just run Role A + Role B both in Claude (ask Claude to switch roles mid-conversation). Less rigorous but faster.
The best debates come from roles that both have legitimate authority but optimize for different metrics. "Security lead vs UX designer" is productive. "Chef vs astronaut" is not.
You can re-run with different role pairs on the same idea for additional angles.