pAI Teardown Plugin — Setup & Usage Guide

Version: 1.1.0 Built by: Palash Somani

---

What Is This?

pAI Teardown is a structured UX audit system you run inside Claude Desktop. It lets any PM, designer, or researcher run a rigorous, framework-backed review of any mobile app — in minutes, not days.

Every review produces:

• A Current State file — scored evidence, issues, screenshot-referenced findings (no recommendations — just facts)
• A Recommendations file — high-level AI-generated feature recommendations
• A Screenshot Manifest — every screen captured, named, and mapped to findings
• An optional Stakeholder PDF — consulting-grade report with embedded screenshots

It covers 10 UX modules (onboarding, homepage, discovery, personalization, content, engagement, monetization, AI, cognitive load, persona alignment) and is grounded in 10 expert frameworks (HEART, Hook Model, Nielsen, JTBD, Fogg, Reforge, Norman, Kano, AARRR, Cagan) — with built-in India consumer benchmarks.

---

Prerequisites

Before installing, make sure you have the following:

1. Claude Desktop App

• Download from: https://claude.ai/download
• Sign in with your work or personal Claude account
• Make sure you're on the latest version (Settings → check for updates)

2. Node.js (for Mobile MCP — required for automated screenshots)

• Download from: https://nodejs.org (v18+ recommended)
• Verify: node --version in your terminal — you should see a version like v20.10.0

3. Android Studio + Emulator (for automated app navigation)

• Download from: https://developer.android.com/studio

• After installing, create an Android Virtual Device (AVD):

  1. Open Android Studio → Device Manager (right sidebar or Tools menu)
  2. Click Create Device
  3. Choose a phone (e.g., Pixel 6) and a system image (API 33+ recommended)
  4. Click Finish → then click the ▶ Play button to start the emulator

• Install the app you want to review on the emulator:

  • Apps with a Play Store listing: open the Play Store inside the emulator, sign in, install the app like on a real phone
  • Apps shared as APKs: drag the .apk file onto the running emulator window — it installs automatically

• Verify the emulator is reachable via ADB (Android Debug Bridge): ADB is the command-line tool that talks to Android devices and emulators. Mobile MCP uses it under the hood. ADB ships with Android Studio's "Platform Tools" SDK component.

  With your emulator running, open a terminal and run:

  adb devices

  You should see output like:

  List of devices attached
  emulator-5554   device
  

  If you instead get command not found: adb, ADB isn't on your PATH yet. Add it:

  • macOS / Linux — add this line to ~/.zshrc (zsh) or ~/.bashrc (bash):

    export PATH="$PATH:$HOME/Library/Android/sdk/platform-tools"

    On Linux, the path is typically $HOME/Android/Sdk/platform-tools. Run source ~/.zshrc (or ~/.bashrc) to reload.
  • Windows — Settings → System → About → Advanced system settings → Environment Variables → edit Path → add %LOCALAPPDATA%\Android\Sdk\platform-tools. Restart your terminal.

  Re-run adb devices after fixing PATH. You must see your emulator listed before moving on — if ADB can't see it, Mobile MCP won't either.

Note: You do NOT need Android Studio if you're doing manual reviews (providing screenshots yourself). Mobile MCP is only required for automated navigation.

---

Installation: Step-by-Step

Step 1 — Download pAI-teardown from GitHub

The plugin is open and free.

1. Open https://github.com/Palash-AI/pAI-teardown in your browser
2. Click the green Code button → Download ZIP
3. Unzip the archive somewhere stable on your machine. Example: ~/pAI-teardown/ on macOS / Linux, or C:\Users\<you>\pAI-teardown\ on Windows
4. Don't leave it in Downloads/ — pick a folder you won't accidentally clean up

Alternative (for the technically-inclined): clone the repo with git so you can pull future updates later:

git clone https://github.com/Palash-AI/pAI-teardown.git ~/pAI-teardown

Step 2 — Open Claude Desktop

Open the Claude Desktop app and start a new conversation.

Step 3 — Load the Skill into Claude Desktop

Claude Desktop reads the skill straight from the folder you just downloaded — there's no plugin install step.

In your new conversation, send this message (replace the path with wherever you unzipped to):

Load the app-teardown skill from ~/pAI-teardown/skills/app-teardown/SKILL.md
end-to-end, plus ~/pAI-teardown/appteardown.md for the deep spec.
From now on, when I ask for a pAI teardown, follow the STEP 0–8 workflow
defined in that skill.

Claude will read the skill files and confirm it's ready.

Tip: If you use Claude Desktop's project / workspace feature, drop the pAI-teardown/ folder into a project once. Future conversations in that project will already have the skill loaded — you won't need to re-do Step 3 every time.

Step 4 — Set Up Mobile MCP (for Automated Screenshots)

Add Mobile MCP to your Claude Desktop configuration:

1. Open Claude Desktop → Settings → Developer → Edit Config
2. This opens the claude_desktop_config.json file in your text editor
3. Add the following inside the "mcpServers" block (or create the block if it doesn't exist):

{
  "mcpServers": {
    "mobile": {
      "command": "npx",
      "args": ["-y", "@mobilenext/mobile-mcp@latest"]
    }
  }
}

4. Save the file and fully quit and restart Claude Desktop (Cmd+Q on macOS / Alt+F4 on Windows — not just close the window)
5. Open your Android emulator (from Android Studio Device Manager) and install the app you want to review

Verify the install in two layers:

Layer 1 — ADB sees the emulator (OS level): in a terminal, run adb devices. You should see your emulator listed (e.g. emulator-5554 device). If you don't, the emulator isn't running or ADB isn't on your PATH — fix that first (see Prerequisites §3). Mobile MCP can't work if ADB can't see the device.

Layer 2 — Mobile MCP sees the emulator (Claude level): After restarting Claude Desktop, start a new conversation and type: list available mobile devices. Claude should return a list including your running emulator. If Layer 1 passes but Layer 2 fails, the issue is Mobile MCP / Claude Desktop config — not your emulator.

Step 5 — Verify Installation

In a new Claude Desktop conversation (with the skill loaded — see Step 3), type:

Run a pAI teardown of Duolingo onboarding as new-user, mode: speed

Claude should:

• Confirm the parameters with you
• List your emulator as the connected device
• Launch Duolingo automatically
• Navigate the onboarding flow, capturing ~3–5 screenshots
• Produce a current-state file + recommendations file + screenshot manifest in the pAI-teardown/ folder

If you see all three output files, the install is working end-to-end. You can now run any other app, module, or persona.

---

Using the Plugin

Basic Command

Type this to Claude Desktop in plain English. The skill recognises the pattern.

Run a pAI teardown of [app-name] [module] as [persona], mode: [speed | rigor]

Examples

Run a pAI teardown of Duolingo onboarding as new-user
Run a pAI teardown of Notion monetization as trial-user, mode: speed
Run a pAI teardown of Cred full as free-user, mode: rigor
Run a pAI teardown of Netflix engagement as paid-user

Parameters

Parameter	Required	Options
App name	Yes	Any app you can install on the emulator (Duolingo, Notion, Cred, your own product, a competitor)
Module	Yes	onboarding, homepage, discovery, personalization, content, engagement, monetization, ai, cognitive-load, persona-alignment, or full
Persona	Yes	new-user, free-user, trial-user, paid-user, lapsed-user
Mode	Optional	speed (≈15 min, top issues) or rigor (≈45-60 min, comprehensive). Default: rigor

---

The 10 Review Modules

#	Module	What It Covers	Key Frameworks
1	onboarding	First-run flow, time-to-value, friction	Fogg MAP, HEART Adoption
2	homepage	IA, navigation, visual hierarchy	Nielsen, Norman, Kano
3	discovery	Search, browse, content surfacing	JTBD, Nielsen Recognition
4	personalization	Recommendation quality, relevance	Hook Model, Reforge
5	content	Consumption UX, player, reader	JTBD, Norman Feedback
6	engagement	Retention loops, habit formation	Hook Model, Reforge Retention
7	monetization	Paywall UX, upgrade flows, pricing	Reforge, Fogg MAP, AARRR
8	ai	AI feature integration, discoverability	JTBD, HEART, Kano
9	cognitive-load	Visual clarity, decision fatigue	Nielsen, Norman, Fogg Ability
10	persona-alignment	Whether the app serves the user's job	JTBD, Cagan, Fogg MAP

---

Scoring Scale

Score	Label	What It Means
1–2	Critical	Fundamentally broken. Fix immediately.
3–4	Poor	Significant issues. High-priority.
5–6	Below Average	Functional but behind competitors.
7–8	Good	Solid, with minor optimization opportunities.
9–10	Excellent	Best-in-class. Benchmark-worthy.

Each score is backed by a specific framework citation and a competitive benchmark (e.g., "Duolingo reaches first lesson in 60 sec; this app takes 4 minutes").

---

Adding Context for Your App (Recommended)

For the most relevant reviews of an app you own, add a [your-app]-context.md file to the context/ folder inside pAI-teardown/. This context file teaches Claude about:

• Your product goals and north star metrics
• User personas (your actual ones, with names)
• Past review history
• Squad structure and ownership

If the file already exists, the skill will auto-load it at the start of every review of that app. After each review, Claude auto-appends findings to this file, so it learns over time.

Template structure for a context file:

# [App Name] Context

## Product Overview
...

## Key Metrics (North Star, L28, etc.)
...

## Personas
...

## Review History
(auto-appended by the skill after each review)

---

Adding a New Domain

The plugin ships with an education domain reference (domains/education.md) containing benchmarks from Duolingo, Coursera, Khan Academy, Unacademy, etc.

To add a new domain (e.g., OTT, Finance, E-commerce):

1. Create a file: domains/[domain].md
2. Include: domain-specific heuristics, best-in-class benchmarks with real metrics, common failure modes, key metrics
3. The orchestrator will auto-detect and load it based on the app type

---

Running Without an Emulator (Manual Mode)

If you don't have Android Studio set up, you can still run full reviews by providing screenshots manually:

1. Take screenshots of the app on your phone or any device
2. Drop them into pAI-teardown/screenshots/[run-id]/img/ using the naming convention from screenshot-protocol.md — e.g. 01-splash-new-user.png, 02-home-new-user.png
3. Tell Claude: Run a pAI teardown of [app] [module] as [persona] — screenshots are already in screenshots/[run-id]/img/

Claude will use the provided screenshots as evidence and still produce the full structured output.

---

Output Files

All review outputs are saved to your pAI-teardown/ folder:

File	Description
[app]-current-state-[persona].md	Evidence + scored issues (no recs)
[app]-recommendations-[persona].md	High-level AI-generated feature recommendations
screenshots/[run-id]/screenshot-manifest.md	All screenshots mapped to findings
[app]-teardown-[date].pdf	(On request) Full consulting-grade PDF with embedded screenshots
context/[app]-context.md	Auto-updated after each review

---

Troubleshooting

Problem	Solution
Claude doesn't seem to know about pAI teardown	Re-do Step 3: send the "Load the app-teardown skill from..." message at the start of the conversation. Claude Desktop loads context per-conversation.
Claude Desktop shows a "Python 3.13 required" page when I try to install Mobile MCP	You're on the wrong install path — Claude Desktop's Extensions / DXT marketplace offers a Python-based Mobile MCP that needs Python 3.13. Skip it. Use Step 4 above instead — the npx config-file route works with the Node.js you already installed in Prereq §2 and doesn't need Python 3.13.
Mobile MCP not connecting	First, isolate the layer: run adb devices in a terminal. If it doesn't list the emulator, the problem is OS-level (see the two adb rows below). If it does list the emulator but Claude still can't see it, ensure the emulator was running BEFORE you started the Claude conversation, and fully quit and restart Claude Desktop (Cmd+Q / Alt+F4) after editing the MCP config — closing the window isn't enough. Verify npx --version works in a terminal.
adb devices shows no devices	The emulator isn't actually running — go back to Android Studio → Device Manager → ▶. If it's already running but ADB still shows nothing, restart the ADB server: adb kill-server && adb start-server, then run adb devices again.
command not found: adb	ADB isn't on your PATH. Either call it by its full path (macOS default: ~/Library/Android/sdk/platform-tools/adb devices) or add platform-tools to PATH — see Prerequisites §3.
App not found on emulator	Install via Play Store (signed in) or drag the .apk onto the running emulator window
Screenshots failing to save	Claude will fall back to /tmp/ and copy to the project folder automatically. If issues persist, run a handshake test (see screenshot-protocol.md).
ModuleNotFoundError: reportlab when generating PDF	Run pip install reportlab Pillow (or pip3) — needed only for PDF output
PDF fonts look like Times / Courier	Install Poppins + Lato from https://fonts.google.com — double-click each .ttf to install
Unknown domain	Claude will use general frameworks only; add a domain file (see "Adding a New Domain") to improve results
Review feels generic	Add a [app]-context.md file to context/ with app-specific metrics and personas
Newer release on GitHub	cd into your folder and git pull (if you cloned), or re-download the ZIP and replace. Re-do Step 3 so Claude reloads the skill.

---

Privacy & Data

• All reviews run locally within your Claude session — no data is sent to third-party services
• Screenshots are saved to your local workspace folder only
• The [app]-context.md file is stored in your local folder and is only accessible to you

---

Feedback & Improvements

This plugin was built by Palash Somani and is actively evolving. If you:

• Find a bug or unexpected behaviour → DM Palash on LinkedIn
• Want to add a new module or domain → see "Adding a New Domain" above
• Want to add a new competitor benchmark → edit the relevant modules/[module].md file directly

---

Built with 10 expert frameworks, 10 review modules, India consumer context, and education domain benchmarks. v1.1.0