REA + Claude Code (product-teardown skill)Partly tested🌏 World

Ganesh Iyer wanted to animate a story his daughter wrote, so he used Claude Code with the open-source tool REA to tear down two story-animation apps and see how they work

Ganesh Iyer (@ganeshiyer316) wanted a short animated video of a story his 7-year-old daughter wrote. Existing tools were pricey and didn't fit, so he tore down two of them first. He made one story in each app while the open-source tool REA (Reverse Engineer Anything) recorded what the app sent and received, keeping only field names, and had Claude turn that into a teardown report with his product-teardown skill: front-end steps, API calls, where media comes from and which AI providers are likely. Every finding is labeled Seen (in the traffic) or Best guess. He shares the skill free on GitHub. We ran the skill's bundled domain logger on our server; we did not install REA or tear down a real product.

Visit site REA + Claude Code (product-teardown skill) ↗
Pricing
Free
Region
Global
Where it's from
Overseas
Added
Nov 14, 2026
Tested on
Nov 14, 2026
Version we used
ganeshiyer316/claude-skills 81dd09b (product-teardown); REA via rea-agents (6.9.0 on npm at test time)
Categories
ResearchCoding
Original author
Ganesh Iyer @ganeshiyer316 ↗
Our test
Partly tested
Cost / time
REA and the skill are free and open source (MIT). The teardown uses your Claude Code or Codex quota; if the product is paid, you need your own legitimate account.
Source
X @ganeshiyer316 ↗GitHub ganeshiyer316/claude-skills ↗

The task

Before building something similar, learn what happens behind a web product when a user clicks: which APIs it calls, where media comes from and roughly which models it uses.

Steps

  1. Install REA's MCP server with npx rea-agents@latest setup and restart your agent (REA needs Node 24.22.1+ or 26.12.1+). Copy the product-teardown folder into ~/.claude/skills/ or ~/.codex/skills/. You also need Google Chrome, Node 22+ and jq.
  2. Launch Chrome with a remote-debugging port and a separate throwaway profile (Chrome refuses debugging on your everyday profile), and sign in to the product yourself in that window.
  3. Before you start, run the bundled log-origins.mjs. It records only each request's origin, resource type, MIME type and method, to reveal third-party CDN and AI hosts.
  4. Stop on the screen where the task begins, have the agent call REA's inspect_web_page with only the product's own origins allowed and JSON shape capture on, then do the task once within about 10 minutes (for example, make one video).
  5. Have the agent write the report per the skill: a diagram, the steps in order, APIs and field names, media sources and the inferred AI pipeline, each labeled Seen or Best guess, plus a list of unknowns.

Prompt

Use the product-teardown skill to tear down this web product. I'll sign in and make one story animation myself. Observe only, don't click or resend requests for me, and label every conclusion Seen or Best guess.

What the author says

Ganesh Iyer @ganeshiyer316 (X, 13 Nov 2026): did several product teardowns this week with Claude Code and REA. He tore down two story-animation apps, making one story in each while REA kept field names only. The report is detailed and shows how the product works, its tech stack and some design decisions, but it can't tell which AI model is used or what happens on the servers. He thinks the approach works for teardowns in any domain. The post includes summary images for the two apps.

Our test

On 14 Nov 2026 on Linux (Node 24.22.1) we cloned ganeshiyer316/claude-skills (81dd09b); product-teardown has a 132-line SKILL.md, 3 scripts and an HTML report template. We launched Chrome Headless Shell 159 with --remote-debugging-port=9222 and a throwaway profile, ran node log-origins.mjs 20 origins.json in the background, then opened https://ttqq.dev/zh/ (our own site) through the debugging endpoint. summarize-origins.sh listed 3 origins: ttqq.dev with 48 requests (document, scripts, styles, XHR and more), static.cloudflareinsights.com with 2 (our Cloudflare analytics) and raw.githubusercontent.com with 2 images; the file held only origin, resource type, MIME type and method, with no paths, headers or bodies. We did not install REA's MCP server (rea-agents is 6.9.0 on npm), have Claude write a report, or tear down any third-party product.

What we found when we tried it:The skill's domain logger records only origin-level data as described and surfaces the third-party services behind a page (on our site, Cloudflare analytics and GitHub-hosted images). A full teardown needs REA installed and you signed in to the product.

How to get in

The skill: https://github.com/ganeshiyer316/claude-skills/tree/main/skills/product-teardown ; REA: https://rea.tools/ .

Who it suits

Indie developers and product managers researching how comparable products are built, comfortable with Claude Code or Codex.

Pitfalls

  • Only tear down products you have a legitimate account for and follow their terms; the skill forbids bypassing paywalls, scraping at scale or probing for vulnerabilities.
  • You see only browser traffic, not server internals or the exact model, so the AI part is mostly inference.
  • REA keeps only origins you list exactly in advance; skip the domain logger and you'll miss third-party CDNs and AI hosts.
  • Apps often open the editor in a new tab; list browser targets again right before you start.
  • Strip your account IDs, project IDs and personal data from the report.

More in this category