A Principal-PM-level product-teardown skill. Point it at any product (Linear, Notion, Cursor, Perplexity…) and it reverse-engineers the thing as a system — loops, strategy, moat, opportunities — then renders the analysis as a bilingual (EN + ZH), print-ready HTML report with an embedded product-screenshot gallery. Not a "5 things I like about this app" review. A structural read of why the product wins, where it leaks, and what the next move should be.

How it works — just 2 steps

Invoke the skill with anything like "teardown Linear" / "break down Notion" / "analyze Cursor" and it runs the same flow every time:

1. Chat response — full teardown in 11 sections (Snapshot → Users & JTBD → Core Loop → Architecture → UX Quality → Business Model → Growth → AI Readiness → Friction → Opportunities → Final PM Verdict).
2. HTML report auto-generates — two cross-linked files (EN + ZH, toggle in the bottom-right), saved to ~/Desktop/Claude skills/product-teardown-<slug>-{en,zh}-<yyyymm>.html. Print-optimized (Cmd+P → clean PDF).

Teardown Report · the 11-section spine

Every run ends with two single-file HTML reports at ~/Desktop/Claude skills/. Fixed 11-section spine, plus a product-tour gallery:

Reports are print-optimized (Cmd+P → clean PDF) and cross-linked EN ↔ ZH via a bottom-right language switcher.

Example output · Linear teardown

product-teardown-linear-en-202607.html + -zh- counterpart · cross-linked EN / ZH switcher · 6-image product tour · 11-section deep dive.

TL;DR + key metrics	Users & JTBD	Core Product Loop
Competitor Landscape	AI / Future Readiness	Metrics · North Star + Guardrails

Five design principles

1. Structure over opinion. Every section answers a specific question. No "here's what I think about this app."
2. Loops, not features. The Core Loop section is the spine of the teardown — features are downstream of the loop.
3. AI placement is a spectrum, not a badge. Assistive / Embedded / Autonomous — pick one and defend it with evidence.
4. Screenshots must come from the product itself. Pull from og:image marketing assets (curl -s <url> | grep og:image), not random blog posts.
5. Bilingual by default. Every teardown ships EN + ZH in the same run, cross-linked. Not translated later.

Repo contents

product-teardown-skill/
├── product-teardown-prompt.md              # Portable skill prompt (paste into any tool)
├── templates/
│   ├── product-teardown-template-en.html   # English report template
│   └── product-teardown-template-zh.html   # Chinese report template (same placeholders)
└── scripts/
    ├── fill_linear.py                      # Reference implementation — renders both templates
    └── translate_template_zh.py            # One-shot: derives the ZH template from EN

• Aesthetic: cream paper + serif display + yellow accent. Print-optimized.
• Placeholders: {{ALL_CAPS_KEYS}}. Fill via str.replace with keys sorted longest-first so prefixes don't collide.
• AI meter: put active on exactly one of {{ACTIVE_IF_ASSISTIVE}} / {{ACTIVE_IF_EMBEDDED}} / {{ACTIVE_IF_AUTONOMOUS}}; empty on the other two.
• Language switcher: {{LANG_EN_HREF}} / {{LANG_ZH_HREF}} = target filenames; {{ACTIVE_IF_EN}} / {{ACTIVE_IF_ZH}} = active on the file's own language.
• Brand footer: contains TEARDOWN. mark + LinkedIn / X / Xiaohongshu links. Do not remove.

Use as a VS Code slash command

Copy product-teardown-prompt.md into your VS Code prompts folder as product-teardown.prompt.md and invoke via /product-teardown <product name>.

macOS: ~/Library/Application Support/Code/User/prompts/

How it thinks

• A product is a system, not a screen. The screen is what you see; the loop is what makes it work. The teardown always finds the loop first, then works outward to surfaces, business model, and moat.
• JTBD beats persona. "Who uses it" is less useful than "what did they hire it to replace." The interesting answer is almost never the obvious one.
• AI readiness is a placement problem, not a checkbox. Every AI feature sits somewhere on Assistive → Embedded → Autonomous. Naming the point is the analysis.
• Opportunities must be actionable. No "they should add AI." Every opportunity in §10 is a concrete wedge you could ship next quarter.

License

MIT — fork it, remix it, ship your own version.

Created by Dreameryanyan · LinkedIn · X · Xiaohongshu