chuhai-skills
Open-source Agent Skills that tear down any public Shopify store and check your own store before launch, with every claim labelled by its evidence.
Works in Claude Code, Codex, Cursor, Gemini CLI and any SKILL.md-compatible agent. Public data only, no API key, MIT.
npx skills add harrison-ming/chuhai-skills
More reviewed e-commerce skills on Moveset · 中文
Page 1 of a real run: deathwishcoffee.com, 2026-11-05 (report language follows the question; this one was asked in Chinese). That run could not fetch the full product feed (rate-limited), and the report says so instead of guessing.
Skills
| Skill | What it does | Status |
|---|
| shopify-store-teardown | Tear down any public Shopify store: identity and store age checks, full catalog / price band / launch cadence stats, theme / app / Markets detection, policy consistency | Available (experimental) |
| payments-readiness | Payments readiness check: compares public policies, contact details, payment icons and currency display with providers' published checklists, checks the payee / provider / payout chain, and estimates payment fees from your order value and volume (dated official rate snapshot) | In development (experimental) |
| competitor-change-radar | Competitor change radar: compare two snapshots of one competitor store for new / removed products, price and discount changes, homepage promos, theme / apps, policy edits, each traceable; the first run builds a baseline and names the next run date | In development (experimental) |
| profit-model | Profit and pricing model: contribution per order, profit after ads, break-even ROAS and max CPA, price floor / target, three scenarios, launch cash; dated Shopify / Stripe / PayPal fee table, every number traceable | Available (experimental) |
| tracking-audit | Ad tracking audit: which ad / analytics tags a site runs, duplicates, placeholder IDs, legacy code; admin checklist for purchases and server-side events; privacy and consent notes | Available (experimental) |
What you get
- A PDF report written in plain language for sellers: a one-page summary up front ("一页看懂": conclusions plus the 3 things most worth copying), question-led sections, and a colour badge on every key claim (verified / store's own claim / third-party / inference ...). Technical details and the evidence table live in the appendix.
- A product list (
商品清单.csv) of every public product with price, compare-at price, stock and launch date, ready for Excel.
Files are saved under your Documents → 出海拆解报告 folder (OneDrive Documents on Windows when present), one folder per run, opened automatically when done:
Documents/
└── 出海拆解报告/
└── allbirds.com 对标拆解 2026-09-30/
├── Allbirds 对标拆解报告.pdf
├── 商品清单.csv
└── 原始数据/ (raw pages and statistics)
Use --out-root <dir> or the CHUHAI_OUTPUT_DIR environment variable to save elsewhere. Without Edge or Chrome the report is an .html file you can print to PDF from the browser.
Install
Claude Code
/plugin marketplace add harrison-ming/chuhai-skills
/plugin install shopify-store-teardown@chuhai-skills
Claude.ai / Claude Desktop: download shopify-store-teardown-<version>.zip from Releases and upload it on the Skills page in settings (code execution must be on). The personal-plan sandbox cannot reach arbitrary websites, so the skill switches to the web-fetch tool automatically.
Other SKILL.md-compatible tools: copy skills/shopify-store-teardown/ into the tool's skills directory, or run npx skills add harrison-ming/chuhai-skills.
Coze and other cloud platforms: upload the zip from Releases.
Requirements
Python 3.8+, standard library only (nothing to pip install). macOS usually has it; most Windows PCs do not, and there python / python3 only print "Python was not found; run without arguments to install from the Microsoft Store". The skill checks first and walks you through installing it, either:
- Type
python in a Command Prompt, which opens the Microsoft Store page for Python, and click Get; or
- Download the installer from python.org and tick "Add python.exe to PATH" on the first screen.
Restart your AI assistant or terminal afterwards. On Windows the command is python or py -3, not python3. Without Python the skill still works in chat, but produces no PDF or product list.
PDF output uses the Edge or Chrome already on your computer. Works best with network access; without it the skill falls back to your agent's web-fetch tool or to pasted content.
Known limitations
- From mainland China, some off-site sources (such as the web archive at web.archive.org) may be unreachable. The report marks them as not found; other conclusions are unaffected.
- When the assistant runs as a background service (for example remotely or as a scheduled task), the browser may be unable to print a PDF. You get an HTML report instead; open it in a browser and print to PDF if needed.
- On a freshly installed Windows, the first HTTPS request to some sites can fail certificate checks until Windows downloads the root certificate. The collector triggers that download once and retries, with verification always on; if it still fails, open the site once in a browser or try again a few minutes later.
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗
