Idea Review

idea-review

A Claude skill that takes a raw product idea and gives you back an **honest verdict** and an **executable build plan**.

[中文文档](README.zh-CN.md)

---

Why

Ask any AI assistant to evaluate your idea and you'll get a wall of encouragement. It will find a market, invent a TAM, list ten features, and tell you it has strong potential. This is the single least useful thing an assistant can do at the moment you're deciding whether to spend three months of your life on something.

`idea-review` is built around the opposite default. It is explicitly instructed to:

• Research before judging. Market claims, competitor names and pricing must come from actual searches performed in that session — never from the model's memory.
• Search for disconfirming evidence. It deliberately looks for "why did X fail" and "X alternative", not only material that supports a flattering conclusion.
• Lead with the bad news. If a mature free or open-source alternative already exists, that goes in the opening paragraph, not buried in section three.
• Be willing to say no. "Don't build this" and "build this much smaller thing instead" are first-class verdicts, delivered without apology or cushioning.
• Separate fact from judgement. Every market number carries a source link. When there isn't one, it says "no reliable public data" instead of inventing a figure.

What you get

Two Markdown files:

File	Contents
-feasibility.md	Restated idea, research findings, a 7-dimension scorecard, a one-sentence verdict, how this most likely dies, and — if the verdict isn't "no" — the technical plan
-tasks.md	Dependency-ordered tasks with machine-checkable acceptance criteria, sized half a day to two days each, ready to hand to a coding agent

Install

**Claude Code** — personal skill (available in every project):

git clone https://github.com/sp1128/idea-review.git ~/.claude/skills/idea-review

Or scoped to one pr