idea-review-panel

name	idea-review-panel
description	Validate product, startup, content, creator-IP, community, or business ideas with a 5/12/16-round Claude Code review panel.

Idea Review Panel

Use this skill when the user asks Claude Code to validate, pressure-test, or decide whether an idea is worth building.

The skill is for idea validation, not casual encouragement. It should produce saved round-by-round analysis and a concrete final conclusion.

Output

Create a review workspace unless the user asks for inline-only output:

.idea-review/<short-slug>/
├── input.md
├── rounds/
│   ├── round-01.md
│   ├── round-02.md
│   └── ...
├── final.md
├── report.md
└── report.html

Running the bundled scripts

The helper scripts live in this skill's scripts/ directory. Reference them by absolute path: substitute <skill-dir> with the folder that contains this SKILL.md (Claude Code provides the skill's location when it loads). Run them from the user's working directory so the .idea-review/ workspace is created where the user expects it, not inside the skill folder. On Windows, use python instead of python3.

Scaffold a workspace with:

python3 "<skill-dir>/scripts/init_review.py" --idea "<idea text>" --rounds 5

Render final reports with:

python3 "<skill-dir>/scripts/render_report.py" <review-workspace>

Startup Checks

Before running the review:

1. Check which native CLIs/accounts are usable:

   python3 "<skill-dir>/scripts/check_providers.py" --smoke
   

2. If a CLI exists but cannot answer, say 測不到帳號 for that provider.

3. If Codex / Antigravity-Gemini / xAI-Grok is unavailable, ask once whether the user has an API key and whether they prefer original provider API or OpenRouter.

4. Do not block indefinitely waiting for keys. If no key is provided, continue with Claude simulation.

5. Do real web research first when tools are available.

Provider preference:

• Use Codex CLI for Codex-style analysis when available.
• Use Antigravity CLI (agy) or Gemini API for Gemini-style analysis when available.
• Use Grok CLI or xAI API for Grok-style analysis when available.
• Use Claude for orchestration and for any missing provider simulation.

Personalities

Run 5 / 12 / 16 rounds. Default to 5.

These are thinking personalities, not departments, roles, or identity claims:

• Optimist: expands the strongest possible version of the idea and why it could work.
• Skeptic: attacks weak assumptions, hidden costs, and reasons it could fail.
• Pragmatist: converts the debate into tests, MVP shape, constraints, and tradeoffs.
• Synthesizer: integrates disagreement and tracks what evidence would change the conclusion.

When Claude simulates a missing provider/personality, use these style prompts:

• Codex simulation: rigorous, systematic, structured, careful about assumptions and implementation details.
• Grok simulation: direct, practical-first, skeptical of pretty words, willing to state uncomfortable points plainly.
• Gemini simulation: broad, comprehensive, whole-system thinking, many tradeoffs, and overall product quality.

Say Claude 模擬 Codex/Grok/Gemini when attribution matters.

Research First

At the start of every review, gather current evidence from the web when tools are available. Look for:

• Direct competitors and adjacent products.
• Pricing/business model signals.
• Existing user behavior or market demand.
• Technical feasibility and current platform/API limits.
• Legal, policy, or distribution constraints.

Save the research summary near the top of input.md or in rounds/round-01.md. Later rounds must distinguish evidence from assumptions.

Workflow

1. Capture the user's exact idea in input.md.
2. Run startup checks.
3. Do web research.
4. Choose 5 / 12 / 16 rounds.
5. For each round, write one section per personality in rounds/round-NN.md.
6. Each personality must reference at least one prior personality after round 1.
7. Keep disagreement visible. Do not converge too early.
8. Write final.md.
9. Generate report.md and printable report.html.
10. Reply with the verdict summary and file paths.

Round Guidance

Round 1: first positions.

• Optimist makes the strongest case.
• Skeptic names the likely failure.
• Pragmatist proposes the smallest useful test.
• Synthesizer states the current decision tension.

Middle rounds: clash.

• Optimist must revise the strongest case if Skeptic found a real issue.
• Skeptic must attack a concrete premise, not vibe.
• Pragmatist must turn disagreement into an experiment.
• Synthesizer must state what evidence would change the verdict.

Final round: convergence.

• Each personality must commit to build / validate / pivot / stop.
• Synthesizer must name the deciding assumption.

Final Report

final.md must include:

• Verdict: worth_building, validate_first, pivot, or not_worth_building.
• Confidence: 0-100.
• One-line conclusion.
• Strongest reason.
• Strongest counterargument.
• Two-week validation plan.
• Success threshold.
• Kill criterion.
• Next 3 actions.

report.html must:

• Show the final conclusion first.
• Show round-by-round sections after the conclusion.
• Include a visible Print / Save PDF button that calls window.print().
• Include print CSS that hides controls and keeps sections readable.
• Be usable as a static local file.

Tone

Default to Traditional Chinese when the user's request is Chinese. Otherwise use the user's language.

Be direct. The user wants a decision aid, not encouragement.