English | 中文

Cross-market idea validation for indie builders.
Real data from Chinese + English markets · strict scoring · bilingual report.

---

The killer angle

Most idea-validation tools look at one market. This one pulls real signals from both the Chinese and English markets at once, then tells you the uncomfortable truth: your pain is probably already solved in the other market. Real demand ≠ real opportunity.

Sample verdict — cross-border payments tool: Global 45/100, China 61/100 → both red oceans → don't build. Three months of code, saved.

Install (Claude Code skill)

git clone https://github.com/w932635477-bit/validate-idea.git ~/.claude/skills/validate-idea
cd ~/.claude/skills/validate-idea && npm install   # one-time

Then just describe your idea in Claude Code ("validate this idea: …") and the skill fires. Zero LLM API key — the Claude running the skill is the scorer.

How it works

flowchart LR
  A["💬 one-line idea"] --> B["🔎 7 fetchers · CN+EN"]
  B --> C["📊 raw data: search / community / competitor"]
  C --> D["🧠 Claude scores · strict, no hallucination"]
  D --> E["📄 bilingual report · local vs foreign · build?"]

Loading

This repo ships the data layer only: 7 fetchers covering search / community / competitor across both markets. Scoring and the bilingual report are produced by your Claude reading the fetched JSON — that's why no API key is needed.

Data sources

Dimension	Chinese market	English market
search	Firecrawl	Firecrawl
community	Zhihu (cookie)	HN / Lobsters / pullpush (Reddit)
competitor	Baidu	Brave (key) + HN Show HN fallback

Every fetcher fails independently (ok:false) and never blocks the others. Missing data is reported honestly, never invented.

Sample reports

Real runs in docs/demos/:

• Cross-border payments — don't build (Global 45 / China 61, both red oceans)
• AI weekly report — build, but China only
• AI PR review bot — don't build (giants + open source + platform-native)
• Interview repo quarantine — real pain (DPRK "Contagious Interview"), but Socket.dev already owns it → niche wedge only

Configure (all optional)

cp .env.example .env   # fill what you have; missing sources just degrade

• FIRECRAWL_API_KEY — search dimension (both markets). Recommended, best data.
• ZHIHU_COOKIE — Chinese community (Zhihu).
• HTTPS_PROXY — needed in China to reach English sources.
• BRAVE_SEARCH_API_KEY — English competitors (HN Show HN backstops if absent).

FAQ

• The skill doesn't trigger — restart Claude Code so it rescans ~/.claude/skills/.
• English sources return nothing — set HTTPS_PROXY if you're behind the GFW.
• Zhihu returns 403 — Zhihu anti-scrape; paste a fresh login cookie into ZHIHU_COOKIE.
• A dimension scores very low — by design: missing data forces a low score (no inventing). Fill the relevant key in .env.

Honest notes

• English sources need a proxy inside China; without it the English dim is thin (flagged honestly).
• Data-poor dimensions are forced low — prevents the "no data → medium score" delusion.
• The report is decision support, not the decision.

Tech

cheerio + undici + zod + tsx. No frontend, no deploy, no LLM calls.

---

MIT © w932635477-bit