Floorplan: A Plain-English Map of Your AI-Built App, Drawn by Claude Code or Codex

Hayden Smith released Floorplan, an MIT-licensed plugin that makes Claude Code or Codex draw an interactive, plain-English map of an AI-built codebase, with a Health check for leaked keys, open Supabase tables, open Firebase rules and fakeable Stripe webhooks. The GitHub repository gained more than 2,000 stars between October 27 and its November 2, 2026 trending appearance, and the changelog shows near-daily releases from 0.3 to 0.6.1 on November 1, 2026. Floorplan requires Node.js 18+ and two install commands, uses an existing Claude or ChatGPT plan with no API keys, and its README cautions that the AI-written names and summaries can be wrong and the Health check is not an audit.

If you ship apps with Claude Code or Codex and could not explain how they fit together, Floorplan https://github.com/haydenschmitty/floorplan?ref=explainx is built for you. It is a new MIT-licensed plugin by Hayden Smith that asks your agent to read your project and draw an interactive, plain-English map of the whole app , then adds a Health check for the mistakes AI-built apps often make. The repo was trending on GitHub on November 2, 2026, with more than 2,000 stars gained since October 27, and its changelog shows releases almost daily from 0.3 to 0.6.1 November 1 . This guide covers what Floorplan does, how to install it, what the Health check really looks for, how it compares with similar tools, and where you should stay careful. TL;DR | Question | Answer | |---|---| | What is it? | A plugin that makes Claude Code or Codex draw a zoomable map of your codebase | | Who is it for? | People who build with AI and do not read the code | | License | MIT, per the repo's LICENSE file | | Cost | Uses your existing Claude or ChatGPT plan, no API keys, per the README | | Install | Two commands in Claude Code or Codex; Node.js 18+ required | | Standout feature | Health check for leaked keys, open Supabase tables, open Firebase rules, fakeable Stripe webhooks | | Latest release | 0.6.1 on November 1, 2026 | | Main caveat | The AI describes your parts, so names and summaries can be wrong; the check is not an audit | What problem is Floorplan solving? The README opens with a blunt scenario: you built an app, probably with a lot of AI help, and it works. But if someone asks "what happens when someone checks out?" or "what breaks if I change this?", the honest answer is no idea. That gap is real for anyone working with coding agents. We have covered the same pain from other angles, such as the Graphify knowledge graph skill https://explainx.ai/blog/graphify-codebase-knowledge-graph-skill-claude-code-cursor-codex-trending-2026 that gives agents a queryable map, and the Claude Code Game Studios template https://explainx.ai/blog/claude-code-game-studios-49-agents-74-skills-template-github-trending-guide-2026 that scaffolds a whole agent team. Floorplan targets the human on the other side of the keyboard. Its output is a picture you read, not a graph the agent reads. How does Floorplan work? Per the README and the project's how-it-works docs, the pipeline has two halves. 1. Scripts do the mechanical work. Plain Node scripts with no dependencies find repos, parse code, match web requests to server endpoints, and lay out the map. Supported languages include TypeScript, JavaScript, Python, Go, Rust, Java, Kotlin and Scala, plus Vue, Svelte and Astro files. Endpoint detection covers Express, Koa, Hono, Next.js, Firebase Cloud Functions, Flask, FastAPI, Django, Go's net/http, Gin, Echo, chi, Actix, Axum and Spring. 2. The agent does the reading. Claude or Codex decides what each part of your app is and describes it in plain words, for example "takes card payments at checkout" instead of "Stripe API wrapper". The result is a floorplan/index.html that opens in your browser. The README says the folder ignores itself in git and nothing is committed. In Claude Code, after your first map, the agent asks whether future sessions should know the map exists, and only then adds a note to CLAUDE.local.md changelog 0.4.4 . Coding repository map illustration, a folder opening into a branching network, used for the Floorplan codebase map Features worth knowing - A map at every zoom. Start with the whole app, then double-click into a part, a group of files, a file and its endpoints. - Real connections across repos. It follows imports and matches each frontend request to the server endpoint that answers it, even when frontend and backend live in separate repos. - Start from a screen. Pick a page and see its parts, requests, server code, data and outside services. - Ask the map. A question like "How does a refund reach the customer's card?" returns an answer plus a mini-map of the pieces involved. The Ask bar answers through the Claude Code CLI or Codex on your own account. - Recent work. After the AI changes code, the cards it touched light up green, each with a note on why. This is the feature that makes the tool useful during daily agent work, because it turns a diff into a picture. - Export. One HTML file you can send to anyone, without the Ask bar and without your folder path or git remotes changelog 0.4.0 . - Blueprints. Give it a PRD or spec and it draws a dashed-card map of what you would need to build, plus a review of open decisions. - Phones and tablets. Version 0.5.0 added a touch layout; the changelog credits contributor @menachem-kurkus for it. You can see the repo's own demo, a coffee-subscription app called Brewline, through the live demo link in the README. How do I install and try it? The README lists both routes. Claude Code desktop Code tab or CLI : claude plugin marketplace add haydenschmitty/floorplan claude plugin install floorplan@floorplan Codex : codex plugin marketplace add haydenschmitty/floorplan codex plugin add floorplan@floorplan In the Codex app you can instead open Plugins, choose Add a marketplace, enter haydenschmitty/floorplan and install Floorplan. Start a new session, open your project for split repos, open the folder that holds both and ask: "Make me a floorplan of this app." In Codex you can type $floorplan . Later, say "Update the floorplan" and only changed files are re-read. Good first prompts from the README's table: "Where do we talk to Stripe?", "If I change the users table, what breaks?", "Which files handle login?" and "Fix the important issues from the floorplan's Health check". If you want a different way to keep agents and files in sync, our guides to Beacon's cross-harness agent memory https://explainx.ai/blog/beacon-asymptote-labs-open-source-cross-harness-agent-memory-claude-code-codex-cursor-2026 and AgentPlane's side-by-side control plane https://explainx.ai/blog/agentplane-local-control-plane-claude-code-codex-cursor-side-by-side-2026 cover neighbouring needs. What does the Health check actually look for? This is the part that made us pay attention, because it targets the mistakes that keep showing up in AI-assisted projects. The changelog for version 0.6.0 October 31 lists five checks: 1. Secret keys sent to the browser. An OpenAI, Stripe or Supabase service-role key in a NEXT PUBLIC , VITE , REACT APP or similar variable, which every visitor can read. Keys public by design, such as Supabase's anon key or Stripe's publishable key, are left alone. 2. Supabase tables anyone can read and change. Tables whose row-level security is never switched on, worked out by replaying the migrations in order, so a table locked later or dropped does not count. Permissive policies are flagged as worth a look. 3. Firebase rules that let anyone in. Firestore, Storage and Realtime Database rules open to everyone, including "test mode" rules, with the date they stop working. 4. Stripe webhooks anyone could fake. Code that acts on events such as checkout.session.completed without verifying Stripe's signature. 5. Permissive CORS. A smaller note, worth fixing when login cookies are allowed along with it. Version 0.6.1 November 1 then cut false alarms: bcrypt and argon2 hashes are no longer reported as passwords, encrypted values in .env files from tools such as dotenvx and sops are left alone, and each issue now has a "Not a problem" button with an optional reason. A box can also open a GitHub issue for a false alarm, with a preview of exactly what it says; the changelog says it never includes your code or values. Coding secrets protection illustration, a lock guarding keys, representing the Floorplan Health check for leaked secrets The README states the limit plainly: it is "a check, not an audit". Treat a clean Health panel as a good sign, not a clearance. If you ship anything with payments or user data, a human review and a proper scanner still matter. For the wider picture of how agents and credentials go wrong, see our reporting on Claude Code agents with personal credentials on a defence network https://explainx.ai/blog/accenture-contractor-personal-claude-code-agent-dod-niprnet-two-months-wired-what-is-reported-2026 , and consider AgentBeam https://agentbeam.com/?ref=explainx , the agent security platform from the explainx.ai team, to stop agents before they take dangerous actions. How does it compare with Archify and Understand Anything? The README includes its own comparison table, so keep in mind this is the author's framing. | | Floorplan | Archify | Understand Anything | |---|---|---|---| | Question it answers | How does the app I built work, and what should I worry about? | How do I show this clearly? | How does this big codebase work, down to functions? | | Made for | People who build with AI and do not read code | Anyone explaining a system or idea | Developers learning a large codebase | | Output | Zoomable map, screens, requests across repos | Polished diagrams in 10 types | Knowledge graph of files, functions, classes | | Only here | Health check, Recent work, blueprints from a PRD | Image and video exports | Diff impact, role-adaptive dashboard | | Runs in | Claude Code, Codex | Claude Code, Codex, Cursor, OpenCode and more | Claude Code, Codex, Cursor, Copilot, Gemini CLI and more | The author's advice: pick Floorplan to understand and check an app you own, Archify for a diagram in a doc or talk, and Understand Anything to study a large codebase function by function. Note that Floorplan runs only in Claude Code and Codex, so Cursor or Gemini CLI users should look at the other two Archify https://github.com/tt-a1i/archify?ref=explainx and Understand Anything https://github.com/Egonex-AI/Understand-Anything?ref=explainx . What about privacy and safety? Any plugin that reads your whole repo deserves scrutiny. What the project says about itself: - The scan, the map and the small server run locally; the server only accepts connections from your own machine. A fix in 0.3.2 made it answer only to the computer's own addresses, to block DNS-rebinding attacks from websites. - It has no hooks, no MCP servers, no dependencies and no install scripts, and downloads nothing. Its only outside request is the opt-in Floorplan Cloud waitlist email. - Each release ships a CHECKSUMS.txt so you can verify your installed copy, and CodeQL scans every change; its first scan found six spots to tighten, all fixed 0.4.2 . - The Health check never stores or shows secret values, only where they are. These are the author's claims in the README and docs; we read them but did not audit the code line by line. The repo is also days old, with a single maintainer driving most changes. Pin a version, read the plugin files before installing, and do not run it on a repo with secrets you cannot rotate until you are comfortable. One more detail matters: the AI reads your files through your own account, like any Claude Code or Codex task. If you open the map inside the Claude app, the README says Claude uploads the map names, descriptions and structure, not source code to your private claude.ai account, and asks first. Where does it fall short? - Descriptions are model-written. A plain-English label is only as good as the model's reading. The README invites you to rename parts "rename 'Main server' to 'Shop API'" , and your edits are kept on update. - Coverage is framework-specific. Matching requests to endpoints works for the frameworks listed above; other stacks get "lighter support". - Cost scales with size. The README says big projects are split across parallel helpers and use your plan like any other task. - It is young. Created recently, version 0.6.1, and fast-moving. Expect changes. Why this matters for AI-built software Coding agents have made it easy to produce working software faster than anyone can understand it. Tools like Floorplan treat comprehension as a first-class feature: not just generating code, but showing the owner what exists, what a change touched and what is exposed. The Health check checks the exact set of mistakes that tend to appear when a model scaffolds Supabase, Firebase or Stripe code on a first pass. The "Recent work" board is the most interesting idea. A diff is hard to review for non-programmers; a map where the changed cards are green and each says why is something a founder or designer can actually read. If it holds up on large real-world repos, expect copycats inside the agents themselves. For a wider look at skills and plugins, see our guide to Claude Code mods and TypeScript plugins https://explainx.ai/blog/claude-code-mods-typescript-plugins-guide-2026 and the AWS AI-DLC workflows https://explainx.ai/blog/awslabs-aidlc-workflows-ai-dlc-steering-rules-claude-code-codex-cursor-kiro-guide-2026 for another take on structuring agent-led development. Bottom line Floorplan is worth ten minutes if you build with Claude Code or Codex and want to see what you have made. Try it on a small project first, read the Health findings with a critical eye, and keep your normal security process. Primary sources: the Floorplan repo https://github.com/haydenschmitty/floorplan?ref=explainx and its changelog https://github.com/haydenschmitty/floorplan/blob/main/CHANGELOG.md?ref=explainx .