SaaS Platform Teardown Kit

One SaaS URL in, a due-diligence-grade teardown out. A Claude Code skill.

What you get · Quick start · How it works · Kit anatomy · Examples · FAQ · Roadmap · Contributing

Give it a landing page and it produces a structured, multi-file Markdown teardown of the whole product: what it does, every feature and user journey on every platform (web / desktop / mobile), how those platforms integrate — and the business logic behind those choices — plus hardware coverage, business-scale/revenue signals, and concrete release artifacts per platform.

Every claim is sourced, dated, and confidence-tagged, so the report tells you not just what the product looks like but how much of that you should believe.

✨ What you get

Nine linked report files per teardown:

<product>-teardown/
├── 00-INDEX.md                    # executive summary + access-methods summary
├── 01-overview-and-recon.md       # + complete verified link inventory
├── 02-web-platform.md             # + access log, journeys
├── 03-mobile-platform.md          # + store artifact tables, access log, journeys
├── 04-desktop-platform.md         # + per-OS/arch artifact table, version history
├── 05-user-journeys.md            # all platforms × 5 journeys, personas, handoff map
├── 06-hardware-integrations.md    # device families — or the explicit none-finding
├── 07-integration-architecture.md # + business-objective map
└── 08-pricing-revenue-sources.md  # + revenue estimates with method per figure

Every claim in the report carries one of three confidence tags — the reader always knows how much of it to believe:

🚀 Quick start

1 · Pick your tier

Nothing is required to start. Each tier adds capability; every tier degrades gracefully when missing.

flowchart LR
    T0["🟢 Tier 0 — zero-install<br/>built-in web search + fetch"] --> OUT
    T1["🔵 Tier 1 — live browser<br/>Playwright MCP — UI claims go Confirmed"] --> OUT
    T2["🟣 Tier 2 — parallel sweeps<br/>agy + jq — research fans out"] --> OUT
    OUT(["📦 The same full teardown either way —<br/>missing tools degrade the report honestly,<br/>they never block it"])

2 · Install

Tier 0 is one step:

cp -r saas-platform-teardown ~/.claude/skills/saas-platform-teardown

Then restart Claude Code (skills and MCP servers load at session start). For Tier 1/2 setup — Playwright MCP registration, agy permissions, and the verification snippets — see setup.md.

3 · Run it

> Do a full platform teardown of https://linear.app

or name the skill explicitly:

> Use the saas-platform-teardown skill on https://linear.app

Claude Code matches the request to SKILL.md's description, works phase by phase through the reference playbooks, and writes the report folder under ./<product-slug>-teardown/ in your working directory.

🔬 How it works

One URL enters; the kit runs recon first, fans the five research tracks out in parallel, then closes by synthesizing journeys and assembling the report (the dashed track is conditional):

flowchart TD
    URL(["🔗 One landing-page URL"]) --> P0["🔍 Phase 0 — intake &amp; recon<br/>tooling check · verified link inventory · sweep questions"]
    P0 --> WEB["🌐 Web audit"]
    P0 --> MOB["📱 Mobile audit"]
    P0 --> DESK["🖥 Desktop audit"]
    P0 --> SWEEP["📡 Research sweeps × 15<br/>stores · pricing · revenue · integrations"]
    P0 -.->|"if a touchpoint is found"| HW["🔌 Hardware deep-dive"]
    WEB --> JOUR["🚶 User-journey synthesis<br/>every platform × journey × persona"]
    MOB --> JOUR
    DESK --> JOUR
    SWEEP --> JOUR
    HW --> JOUR
    JOUR --> ASM["🧾 Report assembly<br/>every claim sourced, dated, confidence-tagged"]
    ASM --> OUT(["📁 product-teardown/<br/>9 linked .md files"])

The design principles:

• One skill, one job, progressive disclosure. The orchestrator is short. Depth lives in references/, loaded on demand — the documented Claude Code pattern, not an improvised one.
• Tools do what they're good at. Live DOM/UI truth comes from the configured browser (Playwright MCP when present), not from guessing what a page looks like. Broad, fast-moving facts (app store data, pricing, revenue estimators, integration directories) are swept by a dedicated research pass — fanned out in parallel through agy when installed, run sequentially on the built-in search otherwise. Claude Code does the synthesis, verification, and writing — the part that needs judgment.
• Decision-grade, not feature-list-grade. A teardown answers "is this product a real threat?": full user journeys per platform, business objectives behind every integration, explicit hardware answers, revenue/scale signals with methods, and release artifacts per platform — not just "an app exists."
• No credentials is a methodology, not a dead end. Self-serve trials are used when offered; genuinely gated products get the documented no-credential ladder (demo videos, help-center screenshots, reviews, Wayback, API docs) — and every platform section logs which methods produced its claims, so the reader can weigh verified truth against secondhand reconstruction.
• Every claim is sourced or flagged. The report template forces a Confirmed/Reported/Inferred distinction and a source link per section, because SaaS products change weekly and an undated, unsourced teardown is worthless in three months.

📦 Kit anatomy

This is not a single prompt. It's a skill package — a short orchestrator (SKILL.md) plus phase-specific reference playbooks, designed around how Claude Code's Skills system actually works: metadata loads at startup, the orchestrator loads when the task matches, and each playbook loads only when the orchestrator points at it. That progressive disclosure is what makes it reliable instead of a wall of instructions held in one context window.

flowchart LR
    A(["🗣 Teardown request"]) --> B["🏷 Skill description matches<br/>(metadata loaded at startup)"]
    B --> C["📄 SKILL.md<br/>short orchestrator"]
    C -->|"phase by phase, on demand"| D["📚 references/<br/>8 phase playbooks"]
    D --> E["🧱 Report template<br/>structure is forced, not remembered"]
    E --> F(["📁 product-teardown/"])

saas-platform-teardown/
├── SKILL.md                       # short orchestrator — loads when the task matches
├── references/                    # depth lives here, loaded on demand
│   ├── 01-intake-and-recon.md
│   ├── 02-web-platform-audit.md
│   ├── 03-mobile-platform-audit.md
│   ├── 04-desktop-platform-audit.md
│   ├── 05-integration-architecture.md
│   ├── 06-research-sweeps.md
│   ├── 07-report-assembly.md
│   └── 08-hardware-integrations.md
└── assets/report-template/        # the structure every report must follow

🧪 Example teardowns

Full teardowns generated by running this kit, committed as point-in-time research snapshots: SaaS products ship weekly — treat the specifics as dated; the methodology is the durable artifact. Provenance and run notes in examples/README.md.

❓ FAQ

📍 Roadmap

• Multi-harness adapters — same playbooks, other agent harnesses.
• More examples — examples are generated by running the finished kit and regenerate on minor-version milestones (next: hardware-centric).
• Community sweeps — new research sweeps follow the house shape (role / one question / sourcing / output shape); see CONTRIBUTING.md.

🤝 Contributing

Playbook conventions, the validation gate, and the example policy are in CONTRIBUTING.md. scripts/validate-skill.sh checks the kit's structural invariants; CI runs it plus markdownlint and a link check on every push. MIT licensed — see LICENSE.