<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Vadim Karnopelev on Medium]]></title>
        <description><![CDATA[Stories by Vadim Karnopelev on Medium]]></description>
        <link>https://medium.com/@vadimkarnopelev?source=rss-e4e992b50822------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*6vrImJUerRo_7Iva.</url>
            <title>Stories by Vadim Karnopelev on Medium</title>
            <link>https://medium.com/@vadimkarnopelev?source=rss-e4e992b50822------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 13:58:47 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@vadimkarnopelev/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[I was tired of every budgeting app wanting my bank login. So I built one that never asks.]]></title>
            <link>https://medium.com/@vadimkarnopelev/i-was-tired-of-every-budgeting-app-wanting-my-bank-login-so-i-built-one-that-never-asks-2f5b351ad80c?source=rss-e4e992b50822------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f5b351ad80c</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[personal-finance]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Vadim Karnopelev]]></dc:creator>
            <pubDate>Sat, 04 Jul 2026 08:11:43 GMT</pubDate>
            <atom:updated>2026-07-04T08:11:43.789Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*hUmpiCyV2QWzgEw5.png" /><figcaption><em>Waisly’s dashboard — your whole month at a glance, no bank login required.</em></figcaption></figure><p>I’ve downloaded and deleted more budgeting apps than I can count. Different logos, same first screen: “Connect your bank account to get started.”</p><p>That’s usually where I stop. I’d downloaded the thing four minutes ago and now it wanted my banking login. I know Plaid is a real company and the reassurances are probably true (“read-only, we never move money”), but it never felt worth it to me for the privilege of watching my own spending. So I’d close the app and go back to a notes file. Every time.</p><p>The other thing that got me was the fake “free.” You’d sign up, add three transactions, and hit a wall. The free version wasn’t a product, it was a trailer.</p><p>I kept waiting for someone to build the version I wanted and nobody did, so about a year ago I started building it myself. It’s called Waisly. It’s on the App Store and Google Play now, and I want to be straight about what it is and what it still isn’t.</p><h3>How it tracks money without your bank</h3><p>The whole app is built around one rule: it never asks for your bank login. So the money has to get in some other way. There are three, and you’ll probably settle on whichever one fits how you shop.</p><p><strong>Snap a receipt.</strong> Take a photo and it reads back the total, the store, the date, and the individual line items. This is the one I use for groceries, where the receipt is a mile long and I’d never type it out.</p><p><strong>Paste a payment text.</strong> Your bank already texts you every time you spend. Copy that message, paste it in, and it becomes a transaction. Same with a purchase confirmation or a booking email. People sleep on this one until they try it and realize half their spending arrives as a text already.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*ps_CN3oA3i1IG2bg.png" /><figcaption><em>Scanned, pasted, or typed, you end up on the same editable transaction. Nothing is a black box.</em></figcaption></figure><p><strong>Or just type it.</strong> Two taps. Nothing forces you into one way of doing it.</p><p>By default all of this stays on your phone. No bank connection, no Plaid, no “here’s your weekly report” email you didn’t ask for. If you want cloud backup and sync across devices you can switch it on, but the app works fine if you never do.</p><h3>The free version is the actual app</h3><p>This is the part I’m stubborn about. Free doesn’t mean crippled. Budgets, goals, analytics, exports, all of it, no cap and no countdown.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*ldj3H-BtFuaKkcnh.png" /><figcaption><em>Subscriptions, goals, and recurring income/expenses — one tap from the home screen.</em></figcaption></figure><ul><li><strong>Budgets</strong> that warn you <em>before</em> you overspend, monthly or per category.</li><li><strong>Subscription tracking</strong> so you can finally see the total monthly bleed, when each one renews, and how it’s crept up over a year. It nudges you before a trial flips to paid.</li><li><strong>Savings goals</strong> with an ETA that’s honest about your current pace instead of the pace you wish you had.</li><li><strong>Analytics and forecasting</strong> — spending trends, cash flow, top merchants, a health score, and a projection of where the month is heading if you keep going like this.</li><li><strong>A cost-of-living report</strong> that splits your spending by city and country. This one’s for the travelers and the people who moved somewhere new and want to know if it’s actually cheaper.</li><li><strong>Tags</strong> you can pile onto a transaction (a trip, reimbursables, the tax-deductible stuff), each with its own color and icon, then pull a report for any of them. They show up in exports too.</li><li><strong>A calendar view</strong> of money in and out, day by day.</li><li><strong>Exports</strong> to CSV, PDF, or JSON, because it’s your data and you shouldn’t have to ask.</li><li><strong>Multi-currency everywhere, 8 languages, PIN and biometric lock, dark mode.</strong></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*PQEjFMo8x4FqwZAY.png" /><figcaption><em>Item-level analytics — see exactly what you spend the most on, not just which category.</em></figcaption></figure><h3>So what do you actually pay for?</h3><p>The AI. Receipt scanning, the paste-to-parse thing, automatic subscription detection, and the spending insights. That’s the whole paid tier, $<strong>4.99 a month or $39.99 a year.</strong></p><p>I’ll tell you exactly why it’s paid: every scan and every parse runs on a model that charges me per call. If I gave the AI away for free I’d lose money on my most active users, which is a great way to kill an app. Everything that doesn’t cost me per-use stays free, and I’m not planning to move that line.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*Cwlwwm0g_3DzJVv0.png" /><figcaption><em>The entire paid tier, on one screen. Just the AI. Nothing else is locked.</em></figcaption></figure><h3>What it can’t do yet</h3><p>I’d rather you know now than feel misled after you install it:</p><ul><li><strong>It won’t read your texts or email in the background.</strong> You still paste or scan. Doing it automatically is the next big thing I’m working on.</li><li><strong>No shared or split wallets</strong> for couples and roommates. I want to build it. It’s not there.</li><li><strong>No web version.</strong> It’s a phone app for now.</li></ul><h3>If you give it a shot</h3><p>It’s basically me building this. Downloads are nice but what I actually want is to know where it falls apart. If the onboarding loses you, if a screen makes no sense, if something just annoys you, tell me. I read all of it.</p><p>And there’s one thing I genuinely can’t figure out from inside my own head: does skipping the bank connection feel like a relief, or does the pasting and scanning feel like a chore? I’ve been staring at this for a year and I’ve lost all objectivity. You haven’t.</p><p><strong>iOS:</strong> <a href="https://apps.apple.com/us/app/waisly/id6770818429">https://apps.apple.com/us/app/waisly/id6770818429</a> <strong>Android:</strong> <a href="https://play.google.com/store/apps/details?id=com.sharebird.vadim7j7.WaislyMobileApp">https://play.google.com/store/apps/details?id=com.sharebird.vadim7j7.WaislyMobileApp</a> <strong>Web:</strong> <a href="https://getwaisly.app/">https://getwaisly.app</a></p><p>Thanks for reading.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f5b351ad80c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Built a Website Cloner with AI Agents (And It’s Kinda Scary Good)]]></title>
            <link>https://medium.com/@vadimkarnopelev/i-built-a-website-cloner-with-ai-agents-and-its-kinda-scary-good-5d7dcb7ebe0c?source=rss-e4e992b50822------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d7dcb7ebe0c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[devtools]]></category>
            <dc:creator><![CDATA[Vadim Karnopelev]]></dc:creator>
            <pubDate>Thu, 12 Feb 2026 10:49:05 GMT</pubDate>
            <atom:updated>2026-02-12T18:33:06.198Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rHj3X-PCvYpDF-9LPPBzPw.png" /></figure><p>You know that feeling when you see a beautifull website and think “I wish I could build something like that”?</p><p>Yeah, me too. Except I got tired of manually inspecting elements, copying hex codes, and pretending I understand why designers use 47 shades of gray.</p><p>So I built something that does it for me. And honestly? It works better than it should.</p><h3>The Problem (AKA Why I Did This)</h3><p>Here’s my typical workflow when I see a nice landing page:</p><ol><li>Open DevTools</li><li>Start copying CSS values</li><li>Get distracted by how messy their code is</li><li>Forget what I was doing</li><li>Give up and use a template</li><li>Hate myself a little</li></ol><p>Sound familiar?</p><p>I wanted something that could just… look at a website and recreate it. Not pixel-perfect (that’s what screenshots are for), but structurally accurate with the right design tokens.</p><h3>Quick Setup</h3><p>Before we dive in, here’s how to get this running (takes like 2 minutes):</p><pre># 1. Clone the repo<br>git clone https://github.com/anthropics/buildmate<br>cd buildmate<br># 2. Install<br>python3 -m venv .venv<br>.venv/bin/pip install -e .<br># 3. Bootstrap your project (pick what you need)<br>buildmate nextjs+fastapi /path/to/project                # Full-stack cloning<br>buildmate nextjs+fastapi+scraping /path/to/project       # Full-stack + multi-page<br>buildmate nextjs /path/to/project                        # Frontend only</pre><p><strong>Stack breakdown:</strong></p><ul><li>nextjs - Frontend (Next.js + Tailwind)</li><li>fastapi - Backend (FastAPI + SQLAlchemy + JWT auth)</li><li>scraping - Multi-page site crawling (/clone-site command)</li></ul><h3>Enter the Clone Commands</h3><p>I added three skills (slash commands) that invoke specialized AI agents:</p><pre>/analyze-site &lt;url&gt;    # → site-analyzer agent<br>/clone-page &lt;url&gt;      # → ui-cloner + api-generator agents<br>/clone-site &lt;url&gt;      # → all three agents working together</pre><p>And here’s the kicker — it doesn’t just clone the frontend. It generates the <strong>backend too</strong>. Full-stack cloning with one command.</p><p>Let me show you what they actually do.</p><h3>Let’s Clone Cal.com (For Science)</h3><p>I picked <a href="https://cal.com/">cal.com</a> because:</p><ul><li>It’s open source (so no weird ethical stuff)</li><li>The design is clean and modern</li><li>Developers actually know what it is</li><li>Their landing page has good variety of components</li></ul><h3>Step 1: Analyze the Site</h3><pre>/analyze-site https://cal.com</pre><p>The site-analyzer agent wakes up and does its thing. It&#39;s not just taking a screenshot - it&#39;s actually understanding the page:</p><pre>Analyzing https://cal.com...</pre><pre>Extracting:<br>- Color palette (primary, secondary, accents)<br>- Typography (fonts, sizes, weights)<br>- Spacing system (margins, paddings, gaps)<br>- Component structure (hero, features, testimonials)<br>- Layout patterns (grid, flex, positioning)</pre><pre>Analysis saved to .agent-pipeline/site-analysis.md</pre><p>Here’s what the analysis looks like (abbreviated):</p><pre># Site Analysis: cal.com</pre><pre>## Design Tokens</pre><pre>### Colors<br>- Primary: #111827 (near black)<br>- Accent: #2563eb (blue)<br>- Background: #ffffff<br>- Text Secondary: #6b7280<br>- Border: #e5e7eb</pre><pre>### Typography<br>- Headings: Cal Sans, Inter (fallback)<br>- Body: Inter<br>- Sizes: 14px, 16px, 18px, 24px, 36px, 48px, 60px</pre><pre>### Spacing<br>- Base unit: 4px<br>- Common gaps: 16px, 24px, 32px, 48px, 64px</pre><pre>## Components Identified</pre><pre>1. **Navigation**<br>   - Logo left, links center, CTA right<br>   - Sticky on scroll<br>   - Mobile hamburger menu</pre><pre>2. **Hero Section**<br>   - Large heading with gradient text<br>   - Subheading with muted color<br>   - Two CTA buttons (primary + secondary)<br>   - Product screenshot below</pre><pre>3. **Features Grid**<br>   - 3-column layout<br>   - Icon + title + description pattern<br>   - Hover effects on cards</pre><pre>4. **Testimonials**<br>   - Carousel/slider<br>   - Avatar + quote + name + company</pre><pre>5. **Footer**<br>   - Multi-column links<br>   - Newsletter signup<br>   - Social icons</pre><p>The agent extracts the actual values, not just “it looks blue-ish”. These are real design tokens you can use.</p><h3>Step 2: Generate the Code</h3><p>Now the fun part:</p><pre>/clone-page https://cal.com --frontend nextjs --ui tailwind</pre><p>But here’s the thing — it doesn’t just yolo generate code. First, it shows you a plan:</p><pre>┌─────────────────────────────────────────────────────────────────┐<br>│                    CLONE GENERATION PLAN                        │<br>├─────────────────────────────────────────────────────────────────┤<br>│ Source: https://cal.com                                         │<br>│ Output: ./                                                      │<br>├─────────────────────────────────────────────────────────────────┤<br>│                                                                 │<br>│ FRONTEND (Next.js + Tailwind)                                   │<br>│ ├── Pages: 1 (landing)                                          │<br>│ ├── Components: 12 (5 sections, 7 UI)                           │<br>│ └── Output: ./frontend/                                         │<br>│                                                                 │<br>│ BACKEND (FastAPI)                                               │<br>│ ├── Models: 2 (User, Subscription)                              │<br>│ ├── Endpoints: 5 (auth, subscribe, etc.)                        │<br>│ └── Output: ./backend/                                          │<br>│                                                                 │<br>├─────────────────────────────────────────────────────────────────┤<br>│ Proceed? [y/n]                                                  │<br>└─────────────────────────────────────────────────────────────────┘</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*1dZOJaCcpLekaCrc.png" /></figure><p>You get to review before it writes anything. After confirmation:</p><pre>Generating...</pre><pre>./cloned/<br>├── frontend/<br>│   ├── src/<br>│   │   ├── app/<br>│   │   │   ├── page.tsx<br>│   │   │   └── layout.tsx<br>│   │   ├── components/<br>│   │   │   ├── ui/<br>│   │   │   │   └── Button.tsx<br>│   │   │   └── sections/<br>│   │   │       ├── Hero.tsx<br>│   │   │       ├── Features.tsx<br>│   │   │       └── Testimonials.tsx<br>│   │   └── hooks/<br>│   └── package.json<br>│<br>└── backend/<br>    ├── app/<br>    │   ├── api/routes/<br>    │   │   └── auth.py<br>    │   ├── models/<br>    │   │   └── user.py<br>    │   └── main.py<br>    └── pyproject.toml</pre><pre>Done! Frontend: npm run dev | Backend: uvicorn app.main:app</pre><h3>Running What You Cloned</h3><p>The clone generates the code, but you need to install dependencies:</p><pre># Frontend<br>cd cloned/frontend<br>npm install<br>npm run dev<br># → http://localhost:3000</pre><pre># Backend (in another terminal)<br>cd cloned/backend<br>pip install -e &quot;.[dev]&quot;<br>docker-compose up -d db   # Start PostgreSQL<br>alembic upgrade head      # Run migrations<br>uvicorn app.main:app --reload<br># → <a href="http://localhost:8000/">http://localhost:8000</a><br># → API docs at <a href="http://localhost:8000/api/docs">http://localhost:8000/api/docs</a></pre><p>Full-stack. One command to generate, a few more to run. I still can’t belive it works.</p><p>And here’s what it looks like running at localhost:3000:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*Vk5zdEFkSSRkSk0o.png" /></figure><p><strong>Live demo:</strong> <a href="http://calcom-clone-experiment.vercel.app/">calcom-clone-experiment.vercel.app</a> — yes, this is the actual output. I deployed it so you can see for yourself.</p><p><strong>Source code:</strong> <a href="https://github.com/vadim7j7/calcom-clone-experiment">github.com/vadim7j7/calcom-clone-experiment</a> — the full generated code, unedited.</p><h3>What You Actually Get</h3><p>Here’s the Hero component it generated:</p><pre>// components/Hero.tsx<br>export function Hero() {<br>  return (<br>    &lt;section className=&quot;pt-24 pb-16 px-4 md:px-8 lg:px-16&quot;&gt;<br>      &lt;div className=&quot;max-w-6xl mx-auto text-center&quot;&gt;<br>        &lt;h1 className=&quot;text-4xl md:text-5xl lg:text-6xl font-bold text-gray-900 mb-6&quot;&gt;<br>          Scheduling infrastructure<br>          &lt;span className=&quot;bg-gradient-to-r from-blue-600 to-purple-600 bg-clip-text text-transparent&quot;&gt;<br>            {&quot; &quot;}for everyone<br>          &lt;/span&gt;<br>        &lt;/h1&gt;</pre><pre>        &lt;p className=&quot;text-lg md:text-xl text-gray-600 max-w-2xl mx-auto mb-8&quot;&gt;<br>          The open source Calendly alternative. Self-host or let us handle it.<br>          White-label by design.<br>        &lt;/p&gt;</pre><pre>        &lt;div className=&quot;flex flex-col sm:flex-row gap-4 justify-center&quot;&gt;<br>          &lt;button className=&quot;px-6 py-3 bg-gray-900 text-white rounded-lg font-medium hover:bg-gray-800 transition-colors&quot;&gt;<br>            Get Started<br>          &lt;/button&gt;<br>          &lt;button className=&quot;px-6 py-3 border border-gray-300 rounded-lg font-medium hover:bg-gray-50 transition-colors&quot;&gt;<br>            View Demo<br>          &lt;/button&gt;<br>        &lt;/div&gt;<br>      &lt;/div&gt;<br>    &lt;/section&gt;<br>  );<br>}</pre><p>Is it pixel-perfect? No. Is it 90% there and actually usable? Yes.</p><h3>Why Agents Instead of Just… Code?</h3><p>Good question. Here’s why this is a multi-agent thing:</p><h3>The Site Analyzer Agent</h3><p>This one’s job is understanding, not generating. It:</p><ul><li>Fetches the page (handles JS rendering if needed)</li><li>Extracts visual patterns</li><li>Identifies component boundaries</li><li>Pulls exact design token values</li><li><strong>Discovers data models and API endpoints</strong></li><li>Creates structured documentation (both JSON and Markdown)</li></ul><p>It doesn’t write code. It writes a brief for someone who will. And now it looks for backend stuff too — what data the page displays, what API calls it makes, what auth methods exist.</p><h3>The UI Cloner Agent</h3><p>This one’s job is building the frontend. It:</p><ul><li>Reads the analysis (not the original site)</li><li>Knows framework-specific patterns</li><li>Generates idiomatic code for your stack</li><li>Applies the design tokens correctly</li><li>Creates proper file structure</li><li><strong>Asks for confirmation before generating</strong></li></ul><p><strong>And it doesn’t create giant monolithic components.</strong> The cloner follows atomic design:</p><pre>components/<br>├── ui/           # Atoms: Button, Card, Input<br>├── sections/     # Organisms: HeroSection, Features<br>└── layout/       # Templates: PageLayout</pre><p>So when it generates a HeroSection, it imports the Button from ui/ instead of copying button styles inline. You get components you can actually reuse across your project:</p><pre>// It generates THIS (reusable)<br>import { Button } from &#39;../ui/Button&#39;;</pre><pre>export function HeroSection() {<br>  return (<br>    &lt;div&gt;<br>      &lt;h1&gt;Title&lt;/h1&gt;<br>      &lt;Button size=&quot;lg&quot;&gt;Get Started&lt;/Button&gt;  {/* Reusable! */}<br>    &lt;/div&gt;<br>  );<br>}</pre><pre>// NOT this (copy-paste nightmare)<br>export function HeroSection() {<br>  return (<br>    &lt;div&gt;<br>      &lt;h1&gt;Title&lt;/h1&gt;<br>      &lt;button className=&quot;px-6 py-3 bg-indigo-600...&quot;&gt;  {/* Duplicated everywhere */}<br>        Get Started<br>      &lt;/button&gt;<br>    &lt;/div&gt;<br>  );<br>}</pre><p>Separation of concerns, even for AI agents.</p><h3>The API Generator Agent (New!)</h3><p>This is the new one. It handles the backend:</p><ul><li>Reads the site analysis (data models, endpoints, auth)</li><li>Generates FastAPI code (Rails and Express coming soon)</li><li>Creates SQLAlchemy models with proper relationships</li><li>Sets up JWT authentication (email + OAuth)</li><li>Generates database migrations with Alembic</li><li>Includes proper Pydantic schemas for validation</li></ul><p>The backend it generates isn’t a toy — it’s production-ready with:</p><ul><li>Async database operations</li><li>Password hashing (bcrypt)</li><li>Token refresh flows</li><li>CORS configuration</li><li>OpenAPI docs at /api/docs</li></ul><h3>Why This Matters</h3><p>If one agent did everything, it would:</p><ol><li>Mix up analysis with generation</li><li>Conflate frontend and backend concerns</li><li>Generate inconsistent code</li><li>Be harder to debug when things go wrong</li></ol><p>With three agents, you can:</p><ul><li>Review the analysis before generating anything</li><li>Generate frontend only (--no-backend)</li><li>Generate backend only (--no-frontend)</li><li>Swap output formats without re-analyzing</li><li>Debug issues at the right layer</li></ul><h3>All the Formats</h3><p>The cloner supports pretyt much everything:</p><p>FormatWhat You GethtmlSingle index.html (quick prototypes)reactCRA-style componentsnextjsApp router structurenativeReact Native + StyleSheetvueSingle File Componentssvelte.svelte files</p><p>And UI libraries:</p><pre>/clone-page --format react --ui tailwind   # Tailwind CSS<br>/clone-page --format react --ui shadcn     # shadcn/ui<br>/clone-page --format react --ui mantine    # Mantine<br>/clone-page --format react --ui chakra     # Chakra UI<br>/clone-page --format react --ui mui        # Material UI</pre><p>Same analysis, different output. The agent knows how each library works and generates appropriate code.</p><h3>But Wait, There’s More: /clone-site</h3><p>/clone-page is great for landing pages. But what if you want to clone an entire e-commerce site? Or a SaaS dashboard with 15 different page types?</p><p>That’s what /clone-site is for:</p><pre>/clone-site https://example-store.com --depth 2 --max-pages 20</pre><p>It crawls the site, discovers pages, and then does something smart — it <strong>deduplicates shared components</strong>.</p><pre>┌─────────────────────────────────────────────────────────────────┐<br>│                       SITE DISCOVERY                            │<br>├─────────────────────────────────────────────────────────────────┤<br>│ Pages Found:                                                    │<br>│ ├── / (landing)                                                │<br>│ ├── /products (listing)                                        │<br>│ ├── /products/:id (detail)                                     │<br>│ ├── /cart (form)                                               │<br>│ ├── /checkout (multi-step)                                     │<br>│ ├── /login (auth)                                              │<br>│ └── /account/orders (dashboard)                                │<br>│                                                                 │<br>│ Shared Components Detected:                                     │<br>│ ├── Header (all pages)                                         │<br>│ ├── Footer (all pages)                                         │<br>│ ├── ProductCard (3 pages)                                      │<br>│ └── Button (47 instances)                                      │<br>└─────────────────────────────────────────────────────────────────┘</pre><p>Instead of generating 47 different buttons, it creates ONE reusable Button component. The output is actually maintainable.</p><h3>Clone Site Options</h3><pre># Limit how deep to crawl<br>/clone-site https://example.com --depth 3</pre><pre># Only certain sections<br>/clone-site <a href="https://example.com/">https://example.com</a> --include &quot;/products/*,/categories/*&quot;</pre><pre># Skip admin pages<br>/clone-site <a href="https://example.com/">https://example.com</a> --exclude &quot;/admin/*&quot;</pre><pre># Frontend only (skip backend generation)<br>/clone-site <a href="https://example.com/">https://example.com</a> --no-backend</pre><pre># Custom output directory<br>/clone-site <a href="https://example.com/">https://example.com</a> --output ./my-clone/</pre><h3>Real Talk: What It’s Good For</h3><h3>Great For:</h3><ul><li><strong>Learning</strong> — See how designs translate to code</li><li><strong>Prototyping</strong> — Quick starting points</li><li><strong>Client work</strong> — “Make it look like X” becomes easier</li><li><strong>Component inspiration</strong> — Steal patterns, not pixels</li><li><strong>Design system extraction</strong> — Pull tokens from sites you admire</li></ul><h3>Not Great For:</h3><ul><li><strong>Pixel-perfect clones</strong> — Use Figma for that</li><li><strong>Complex interactions</strong> — Animations need manual work</li><li><strong>Dynamic content</strong> — It clones structure, not functionality</li><li><strong>Replacing designers</strong> — Please don’t</li></ul><h3>The Workflow I Actually Use</h3><p>Here’s my real process now:</p><pre># 1. Find inspiration<br># Browse Dribbble, land-book, or just cool sites I find</pre><pre># 2. Clone it (with confirmation)<br>/clone-page <a href="https://site-i-like.com/">https://site-i-like.com</a> --frontend nextjs --ui tailwind</pre><pre># 3. Review the plan, confirm<br># Agent shows me what it&#39;ll generate, I say yes</pre><pre># 4. Customize<br># This is where the real work begins</pre><p>For multi-page sites:</p><pre># Clone an entire e-commerce site<br>/clone-site https://cool-store.com --depth 2 --max-pages 15</pre><pre># Review the plan (pages, components, models, endpoints)<br># Confirm</pre><pre># Get a full-stack app with:<br># - Frontend: Next.js with all pages and shared components<br># - Backend: FastAPI with models, routes, auth</pre><p>The clone isn’t the final product. It’s the starting point that would have taken me a day to set up manually.</p><h3>But Wait, Does It Actually Work?</h3><p>Here’s the thing — generated code is just code. It might have bugs. It might not even compile. That’s why Buildmate doesn’t stop at generation.</p><p>After cloning, I run the code through the same validation pipeline as any other feature:</p><pre># Verify the clone actually works<br>/verify</pre><p>This kicks off the quality gates:</p><pre>Running quality gates...</pre><pre>TypeScript: Checking types...<br>  ✓ No type errors</pre><pre>Lint: Running ESLint...<br>  ⚠ 3 warnings in Hero.tsx<br>  → Auto-fixing...<br>  ✓ Fixed</pre><pre>Build: Running next build...<br>  ✓ Build successful</pre><pre>Tests: Running test suite...<br>  ⚠ No tests found for new components</pre><h3>The Frontend Agents Jump In</h3><p>If something’s broken, you don’t have to fix it manually. The grind agent handles it:</p><pre>/grind</pre><pre>Grind agent activated...</pre><pre>Issues found:<br>1. Missing &#39;use client&#39; directive in Testimonials.tsx<br>2. Image component not using next/image<br>3. Unused import in Footer.tsx</pre><pre>Fixing...<br>  ✓ Added &#39;use client&#39; to Testimonials.tsx<br>  ✓ Replaced &lt;img&gt; with &lt;Image&gt; component<br>  ✓ Removed unused import</pre><pre>Re-running quality gates...<br>  ✓ All checks pass</pre><p>The grind agent knows Next.js patterns. It doesn’t just delete the error — it fixes it properly.</p><h3>Want a Code Review Too?</h3><pre>/review</pre><p>The frontend-reviewer agent looks at the generated code with fresh eyes:</p><pre>## Code Review: Cloned Components</pre><pre>### Overall<br>Clean component structure. Design tokens applied consistently.</pre><pre>### Suggestions</pre><pre>**Hero.tsx**<br>- Consider extracting gradient colors to tailwind config<br>- Button styles could be a reusable component</pre><pre>**Navbar.tsx**<br>- Mobile menu needs aria-labels for accessibility<br>- Consider usePathname for active link highlighting</pre><pre>**Performance**<br>- Add loading=&quot;lazy&quot; to below-fold images<br>- Testimonials carousel could use dynamic import</pre><p>Now you have AI-reviewed, validated code instead of “hope this works” code.</p><h3>The Full Clone-to-Production Flow</h3><pre># 1. Clone (with confirmation)<br>/clone-page https://cal.com --frontend nextjs --ui tailwind</pre><pre># 2. Verify both frontend and backend compile<br>/verify</pre><pre># 3. Fix any issues automatically<br>/grind</pre><pre># 4. Get code review<br>/review</pre><pre># 5. Ship it<br>/ship</pre><p>Five commands. From “I like that website” to “full-stack PR ready for review”.</p><p><strong>Want the full orchestrated experience?</strong> Run /pm Clone cal.com as a Next.js app and let the orchestrator handle everything - planning, cloning, verification, review. One command, zero babysitting.</p><h3>Tips From Actually Using This</h3><p><strong>Tip 1: Simple pages work best</strong></p><p>Landing pages, marketing sites, dashboards — these clone well. Complex web apps with lots of state? Not so much.</p><p><strong>Tip 2: Check the analysis first</strong></p><p>Sometimes the analyzer misses things or gets confused. Quick review saves headaches later.</p><p><strong>Tip 3: Mix and match</strong></p><p>Analyze three sites, take colors from one, typography from another, layout from the third. The analysis files are just documentation — combine them however you want.</p><p><strong>Tip 4: Use it for learning</strong></p><p>Seriously, analyzing well-designed sites and seeing how they structure things taught me more than any CSS course.</p><p><strong>Tip 5: Don’t be creepy</strong></p><p>Clone patterns and approaches. Don’t clone someone’s actual business and pretend it’s yours. That’s not cool.</p><h3>Try It Yourself</h3><p>After setup (see Quick Setup at the top), start Claude Code and run:</p><pre># Single page (full-stack)<br>/clone-page https://cal.com --frontend nextjs --ui tailwind</pre><pre># Frontend only<br>/clone-page <a href="https://cal.com/">https://cal.com</a> --frontend nextjs --no-backend</pre><pre># Multi-page site (needs +scraping stack)<br>/clone-site <a href="https://example-store.com/">https://example-store.com</a> --depth 2</pre><p>That’s it. Go clone something cool.</p><h3>What’s Next?</h3><p>Things I’m still thinking about:</p><ul><li><strong>Screenshot comparison</strong> — Visual diff between original and clone</li><li><strong>More backend frameworks</strong> — Rails and Express are on the roadmap</li><li><strong>Design token export</strong> — Figma/Tokens Studio format</li><li><strong>Animation extraction</strong> — Detect and document transitions</li></ul><p>But honestly, the current version does 90% of what I need. Full-stack cloning with three agents is already kind of insane.</p><p>Got questions? Found a site that breaks it? Let me know.</p><p>And if this saved you time, maybe grab me a coffee? ☕</p><p><a href="https://buymeacoffee.com/vadim7j7"><strong>☕ Buy Me A Coffee</strong></a></p><p><em>Built with </em><a href="https://github.com/vadim7j7/buildmate"><em>Buildmate</em></a><em> — because manually setting up AI agents is so 2024.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d7dcb7ebe0c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Actually Use Buildmate (The Practical Guide)]]></title>
            <link>https://medium.com/@vadimkarnopelev/how-to-actually-use-buildmate-the-practical-guide-9180d2580e14?source=rss-e4e992b50822------2</link>
            <guid isPermaLink="false">https://medium.com/p/9180d2580e14</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Vadim Karnopelev]]></dc:creator>
            <pubDate>Mon, 09 Feb 2026 00:10:55 GMT</pubDate>
            <atom:updated>2026-02-09T00:10:55.753Z</atom:updated>
            <content:encoded><![CDATA[<p><em>because “just read the docs” isn’t helpful and we both know it</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7ywd23sDapSHySoP6QLCMw.png" /></figure><blockquote><strong><em>Hit the paywall?</em></strong><em> </em><a href="https://medium.com/@vadimkarnopelev/i-got-tired-of-being-a-one-man-dev-team-so-i-built-this-thing-3940da901888"><em>Read this article for free here</em></a></blockquote><h3>So You Installed It, Now What?</h3><p>Ok so maybe you read my first post about Buildmate. Or maybe you just found the GitHub and thought “cool I’ll try it.” Either way — you got it installed but now you’re staring at your terminal like “…what do I actually do with this thing?”</p><p>Yeah I get it. Docs are great and all but sometimes you just want someone to show you how they actually use the thing. So thats what this is. Real workflows. Real commands. Stuff I actually do every day.</p><p>Lets go.</p><h3>Quick Setup (In Case You Skipped It)</h3><p>If you already did this skip ahead. If not:</p><pre>git clone https://github.com/vadim7j7/buildmate.git<br>cd buildmate<br>python3 -m venv .venv &amp;&amp; .venv/bin/pip install -e .</pre><p>Then bootstrap your project:</p><pre># Pick your stack<br>buildmate rails /path/to/my-app<br>buildmate nextjs /path/to/my-app --ui=tailwind<br>buildmate rails+nextjs /path/to/my-app<br># Or just use a preset<br>buildmate --profile saas /path/to/my-app</pre><p>Not sure what stacks are available? Check with:</p><pre>buildmate --list           # all stacks<br>buildmate --profiles       # preset combos<br>buildmate --options nextjs # options for a stack</pre><p>After this you got a .claude/ folder with all the agent stuff. Thats it. Your ready.</p><h3>Your First /pm Command</h3><p>Ok so /pm is the main thing. PM stands for Project Manager. Its the command that kicks off the whole workflow - planning, building, testing, reviewing, everything.</p><p>Lets try it. Open your project and run claude, then type:</p><pre>/pm Build a user profile page with avatar, bio, and edit functionality</pre><p>Heres what happens:</p><p><strong>Phase 1 — Planning (this part is interactive)</strong></p><pre>PM: Ok so for user profile, I&#39;m thinking:<br>Requirements:<br>    - Profile page showing avatar, name, bio<br>    - Edit mode to update info<br>    - Avatar upload (or URL?)<br>    - Save/cancel buttons<br>    Questions:<br>    - Should avatar be file upload or just URL input?<br>    - Where does this page live? /profile? /settings?<br>    - Need to create User model or exists already?<br>You: URL is fine for now. /profile route. User model exists.<br>PM: Cool, heres the plan:<br>    [shows feature file with all the details]<br>    Ready to build?<br>You: yep go</pre><p><strong>Phases 2–5 (automatic from here)</strong></p><p>Now it just runs. You dont do anything.</p><pre>[Phase 2: Implementation]<br>→ Spawning backend-developer...<br>→ Creating ProfileController<br>→ Adding routes<br>→ Spawning frontend-developer...<br>→ Creating ProfilePage component<br>→ Creating EditProfileForm component<br>[Phase 3: Testing]<br>→ Spawning testers...<br>→ Writing controller specs<br>→ Writing component tests<br>→ Running quality gates...<br>→ Grind agent fixing 2 lint errors...<br>[Phase 4: Review]<br>→ Spawning reviewers...<br>→ Backend review: APPROVED<br>→ Frontend review: APPROVED<br>[Phase 5: Verification]<br>→ Starting dev server...<br>→ Testing /profile route...<br>→ Component renders ✓<br>→ Form submits ✓<br>Done! Score: 0.89 (B+)</pre><p>And now you have a working profile page. With tests. Reviewed. Verified.</p><p>I know it sounds like magic but its not really. Its just agents doing their specific jobs in order. Backend dev writes backend. Frontend dev writes frontend. Tester writes tests. Each one is focused on one thing.</p><h3>The Agents (Who Does What)</h3><p>Quick reference cause I always forget:</p><p>AgentWhat It DoesWhen It Runs<strong>PM/Orchestrator</strong>Plans the work, coordinates everyoneFirst, stays in control<strong>Backend Developer</strong>Writes Rails/FastAPI codePhase 2<strong>Frontend Developer</strong>Writes React/Next.js codePhase 2<strong>Backend Tester</strong>Writes backend testsPhase 3<strong>Frontend Tester</strong>Writes frontend testsPhase 3<strong>Grind</strong>Runs linters in loop, fixes errorsPhase 3<strong>Verifier</strong>Actually runs the code to test itPhase 3–4<strong>Reviewers</strong>Code reviewPhase 4<strong>Eval</strong>Scores the codePhase 5<strong>Security Auditor</strong>Checks for vulnerabilitiesWhen you ask</p><p>They live in .claude/agents/ as markdown files. You can read them if your curious. Or edit them - more on that later.</p><h3>Commands You’ll Actually Use</h3><h3>/pm “task”</h3><p>The big one. Full workflow.</p><pre>/pm Add pagination to the products list<br>/pm Fix the login bug where sessions expire too fast<br>/pm Build checkout flow with Stripe integration</pre><p>Use this for anything substantial. Features, big fixes, new pages.</p><h3>/verify</h3><p>Tests your code by actually running it. This ones newer and honestly I use it all the time now.</p><pre>/verify                              # test last changes<br>/verify --component Navbar           # test specific component<br>/verify --endpoint POST /api/users   # test specific endpoint<br>/verify --page /dashboard            # test whole page</pre><p>If something fails it tries to fix it automatically. Up to 3 times.</p><h3>/new-model and /new-component</h3><p>Quick scaffolding. Creates the thing plus tests plus all the boilerplate.</p><pre>/new-model Product name:string price:decimal active:boolean<br>/new-component ProductCard<br>/new-component forms/CheckoutForm</pre><p>Way faster than doing it manually. Follows your project conventions too.</p><h3>/parallel</h3><p>Got independent tasks? Run them at the same time.</p><pre>/parallel &quot;Add user avatars&quot; &quot;Add product images&quot; &quot;Add company logos&quot;</pre><p>Creates separate git worktrees, runs agents in each, merges when done. I use this when I got a bunch of similar things to do.</p><h3>/security</h3><p>Security audit. Run it before you deploy pls.</p><pre>/security</pre><p>Checks for injection, auth issues, XSS, CSRF, all that OWASP stuff.</p><h3>/eval</h3><p>Just want a score without the whole workflow?</p><pre>/eval</pre><p>Gives you the grade. Good for checking your own work.</p><h3>/test and /review</h3><p>Sometimes you just want one thing:</p><pre>/test          # just run tests<br>/review        # just review current changes</pre><h3>Git Workflow (The /branch → /ship Flow)</h3><p>Ok so this one is newer and I use it all the time now.</p><p>Instead of manually creating branches, writing commit messages, pushing, writing PR descriptions at 5pm when your brain is fried… just:</p><pre>/branch user-auth                    # creates feature/user-auth<br># ... do your work or use /pm ...<br>/ship                                # commits, pushes, creates PR</pre><p>Thats it. It auto-generates commit messages from your changes. Writes the PR description for you. Even runs quality gates before shipping.</p><p>More examples:</p><pre>/branch login-fix --type fix         # creates fix/login-fix<br>/branch user-auth --issue 42         # links to github issue #42<br>/sync                                # rebase on main if youre behind<br>/ship --draft                        # create PR as draft</pre><p>The /sync one is useful when main has moved ahead and you need to catch up. It handles the rebase (or merge if you prefer) and force-pushes for you.</p><pre>/sync                    # rebase on main<br>/sync --merge            # merge instead of rebase<br>/sync --base develop     # sync with develop branch</pre><p>And /ship does everything at once:</p><pre>/ship                    # commit + push + create PR<br>/ship --draft            # same but draft PR<br>/ship --no-pr            # just commit and push, no PR</pre><p>No more “oh wait I forgot to push.” No more staring at the PR template trying to remember what you changed. It just does it.</p><h3>Customizing Your Setup</h3><h3>Editing Agents</h3><p>The agents are just markdown files. You can edit them.</p><pre>.claude/agents/<br>├── orchestrator.md      # the PM brain<br>├── backend-developer.md # Rails/FastAPI dev<br>├── frontend-developer.md<br>├── grind.md<br>└── ...</pre><p>Open one up. Youll see instructions, patterns, do’s and donts. Change whatever you want.</p><p>For example maybe you want the backend dev to always use a specific gem:</p><pre>## Rules<br>- Always use Pagy for pagination (not Kaminari)<br>- Always use Pundit for authorization<br>- ...</pre><p>Just add it to the file. Next time it runs itll follow your rules.</p><h3>Adding Patterns</h3><p>Got coding patterns you always follow? Add them to .claude/patterns/.</p><p>These are reference docs the agents read. Stuff like “how we structure services” or “our API response format.”</p><h3>Changing Quality Gates</h3><p>In the stack config you can change what commands run:</p><pre>quality_gates:<br>  lint:<br>    command: bundle exec rubocop<br>    fix_command: bundle exec rubocop -A<br>  tests:<br>    command: bundle exec rspec</pre><p>Add more, remove some, change commands. Whatever works for your project.</p><h3>Real Workflows I Actually Use</h3><h3>Building a New Feature</h3><p>This is like 80% of what I do:</p><pre>1. /pm Build [feature description]<br>2. Answer a few questions about requirements<br>3. Approve the plan<br>4. Go get coffee<br>5. Come back, its done<br>6. Maybe tweak something small manually<br>7. Commit and push</pre><p>Thats it. Most features take 10–15 minutes now instead of half a day.</p><h3>Fixing a Bug</h3><pre>1. /pm Fix [describe the bug and where it is]<br>2. Agent reads the code, finds the issue<br>3. Fixes it<br>4. Writes a test so it doesnt happen again<br>5. Done</pre><p>For small bugs I sometimes just do it myself cause its faster. But for anything tricky I let the agent figure it out.</p><h3>Adding to Existing Code</h3><pre>1. /pm Add [thing] to [existing feature]<br>2. Agent reads the existing code first<br>3. Follows the patterns already there<br>4. Adds the new thing in the same style</pre><p>This is where it really shines tbh. It actually reads your code and matches the style. No more “this new code looks totally different from everything else.”</p><h3>Quick Scaffolding Day</h3><p>When I’m setting up a new project I’ll just rapid fire:</p><pre>/new-model User email:string name:string<br>/new-model Product title:string price:decimal<br>/new-model Order user:references total:decimal<br>/new-component Navbar<br>/new-component Footer<br>/new-component ProductCard<br>...</pre><p>Boom. All the basics in like 5 minutes. With tests.</p><h3>Common Issues (And How to Fix Them)</h3><p><strong>“Its stuck in a loop”</strong></p><p>The grind agent will try 10 times then give up. If its not converging theres probably something weird going on. Read the error, might need manual fix.</p><p><strong>“Agent didnt follow my style”</strong></p><p>Edit the agent file in .claude/agents/. Add specific rules about your style. Be explicit - &quot;always use X, never use Y.&quot;</p><p><strong>“Tests keep failing”</strong></p><p>Check if your test setup is correct. Sometimes the agent writes good tests but the test environment isnt configured right. Run the tests manually to see whats up.</p><p><strong>“It made too many files”</strong></p><p>Yeah it can get enthusiastic sometimes. You can tell it in the initial prompt: “keep it simple, minimal files.”</p><p><strong>“The code works but I dont like how its structured”</strong></p><p>Use /review to get feedback, or just refactor yourself. The agents are good but theyre not perfect. Your judgment still matters.</p><h3>Tips From Actually Using This Thing</h3><ol><li><strong>Be specific in your prompts.</strong> “Build a login page” is ok. “Build a login page with email/password, remember me checkbox, forgot password link, and redirect to dashboard after” is better.</li><li><strong>Answer the planning questions.</strong> Dont just say “whatever.” The more context you give, the better the output.</li><li><strong>Use /verify often.</strong> Catches dumb mistakes before you even see them.</li><li><strong>Read the feature files.</strong> They live in .claude/context/features/. Good for remembering what was built and why.</li><li><strong>Edit the agents.</strong> Seriously. Make them yours. Add your patterns, your rules, your preferences.</li><li><strong>Trust but verify.</strong> The code is usually good but always do a quick look before shipping.</li><li><strong>Use the git flow.</strong> /branch → do work → /ship. No more manual PR descriptions at 5pm.</li></ol><h3>Whats Next</h3><p>Working on website cloning — point it at any site and itll generate your own version using your UI library. Thats gonna be a whole other post tho.</p><p>For now, go try this stuff. Break things. Fix things. Let me know what works and what doesnt.</p><p>GitHub: <a href="https://github.com/vadim7j7/buildmate">github.com/vadim7j7/buildmate</a></p><h3>Found This Useful?</h3><p>If this saved you some time maybe buy me a coffee?</p><p><a href="https://buymeacoffee.com/vadim7j7"><strong>☕ Buy Me A Coffee</strong></a></p><p><em>Now go build something cool.</em></p><p><em>- Vadim</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9180d2580e14" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Got Tired of Being a One-Man Dev Team (So I Built This Thing)]]></title>
            <link>https://medium.com/@vadimkarnopelev/i-got-tired-of-being-a-one-man-dev-team-so-i-built-this-thing-3940da901888?source=rss-e4e992b50822------2</link>
            <guid isPermaLink="false">https://medium.com/p/3940da901888</guid>
            <category><![CDATA[developer-tools]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Vadim Karnopelev]]></dc:creator>
            <pubDate>Sun, 08 Feb 2026 08:21:49 GMT</pubDate>
            <atom:updated>2026-02-08T08:21:49.005Z</atom:updated>
            <content:encoded><![CDATA[<p><em>aka how I finally stopped mass-complaining about linter errors at 2am… mostly</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mqpLBy8E-eLctk2c.png" /></figure><h3>Ok So Here’s the Thing</h3><p>Its 3am. I’m so tired. Probably on my fifth coffee at this point, maybe sixth idk I lost count. Eyes are burning from staring at the screen. And I’ve been fighting with the SAME rubocop error for like an hour now.</p><p>You know the cycle right? Run rubocop. 3 errors. Fix them. Run again. Now its 2 different errors. Fix those. Run again. 5 ERRORS. HOW. I literally just fixed things and now theres MORE???</p><p>(I once spent 4 hours debugging something and it turned out I had a typo in a variable name. Four. Hours. I don’t wanna talk about it.)</p><p>Anyway so I’m sitting there at 3am, questioning my life choices, and I’m thinking — wait. Didn’t I become a developer to BUILD cool stuff? Not to fight with semicolons and copy paste Stack Overflow answers about “unexpected end of input”?</p><p>Thats basically why I made Buildmate. I got tired of doing everything myself lol.</p><h3>What Is This Exactly</h3><p>Ok so like… you know how in a real company you got different people doing different things:</p><ul><li>Theres a developer who writes the code (and googles stuff constantly)</li><li>A tester who breaks everything (on purpose… hopefully)</li><li>Code reviewer who points out you forgot null checks again</li><li>Some PM who keeps everyone from losing their minds</li><li>Security person who tells you everything is wrong</li></ul><p>Buildmate gives you all of these. But as AI agents. That actually talk to eachother and do stuff while you go get lunch or whatever.</p><pre>Me: &quot;/pm Build user authentication with OAuth&quot;<br>Buildmate: &quot;got it&quot;<br>         → spawns a backend dev (writes the Rails code)<br>         → spawns frontend dev (does the React stuff)<br>         → spawns testers (writes tests, runs them)<br>         → runs linting in a loop until everything passes<br>         → actually RUNS the code to verify it works<br>         → fixes its own mistakes automatically<br>         → spawns reviewers (finds my mistakes)<br>Me: *drinking coffee* &quot;nice&quot;</pre><p>I know it sounds wild. When I first got this working I didnt believe it either.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/702/0*j6BUE3J9OUz3XzX3.png" /></figure><h3>The Linter Loop From Hell (And How I Fixed It)</h3><p>We’ve ALL been here:</p><pre>$ rubocop<br>3 offenses detected<br>*fixes them carefully*<br>$ rubocop<br>2 offenses detected (different ones??)<br>*fixes those too*<br>$ rubocop<br>5 offenses detected<br>*stares at screen*<br>*questions existence*</pre><p>I swear I’ve lost years of my life to this.</p><p><strong>The fix:</strong> Theres this thing called the Grind Agent. It runs your linters in a loop and fixes stuff automatically. Until everythings green. Or until it gives up after 10 tries.</p><pre>Grind Agent:<br>  Iteration 1: rubocop → 3 errors → fixing...<br>  Iteration 2: rubocop → 0 errors, nice<br>  Iteration 3: rspec → 2 failures → fixing...<br>  Iteration 4: rspec → all green<br>Status: CONVERGED<br>Fixed:<br>  - app/models/user.rb:15 - frozen_string_literal<br>  - spec/models/user_spec.rb:28 - nil vs empty string</pre><p>You didn’t do anything. It handled it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/0*6HSOyeboq_TGvWSq.png" /></figure><h3>“LGTM” Isn’t Really a Code Review</h3><p>Be real. When your tired and the PR has 47 files, “Looks Good To Me” really means “I looked at maybe 3 files and gave up.”</p><p>We all do it. Its human nature.</p><p><strong>The fix:</strong> Theres an Eval Agent that scores your code with actual grades.</p><pre> Thing	       Weight	  Score<br>Correctness	30%	  0.85<br>Code Quality	25%	  0.90<br>Security	20%	  0.95<br>Performance	15%       0.80<br>Test Coverage	10%       0.75<br>Final		0.87 	  (B)</pre><p>Actual numbers. No more arguing about tabs vs spaces. It tells you stuff like “line 47 might have an N+1 query” or “this method is doing 5 things, split it up.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/928/0*FRqIkQYoeMIGdH_i.png" /></figure><h3>It Actually Tests Itself (This Is the Cool Part)</h3><p>Most AI coding tools just write code and hope it works. You run it, it crashes, you spend 30 minutes debugging.</p><p>I got tired of that too.</p><p>So theres /verify. It actually RUNS your code:</p><p><strong>Backend:</strong> Starts dev server, makes real HTTP requests, validates responses</p><p><strong>Frontend:</strong> Opens a browser, navigates to your page, takes screenshots, checks if components render, looks for console errors</p><p>But heres the best part. If something fails? It fixes itself.</p><pre>[Verification] Creating HeroSection...<br>[Testing]<br>- Starting dev server... ✓<br>- Looking for .hero-section...<br>- Component NOT FOUND ✗<br>[Analyzing]<br>- Component exists but not exported<br>- Adding export...<br>[Retry 1/3]<br>- Component found ✓<br>- No console errors ✓<br>Verification passed after 1 fix.</pre><p>It built the component. Tested it. Found it wasnt exported. FIXED it. Tested again. All by itself.</p><p>Backend too:</p><pre>[Verification] POST /api/users<br>- Making request...<br>- Status: 500 ✗<br>[Analyzing]<br>- Missing user_params method<br>- Adding to controller...<br>[Retry 1/3]<br>- Status: 201 Created ✓<br>Verification passed after 1 fix.</pre><p>The stupid mistakes that take 20 minutes to debug? Gone.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/756/0*XdbxW099gcb9AJ4f.png" /></figure><h3>Security (Actually Important)</h3><p>Theres a Security Auditor agent. Checks OWASP stuff — injection attacks, auth problems, XSS, CSRF, etc.</p><pre>## Security Report<br><br>Found: 2 issues<br><br>1. [MEDIUM] Possible SQL injection<br>   File: app/models/search.rb:45<br>   Fix: use parameterized query<br>2. [LOW] No rate limiting on login<br>   Suggestion: add rack-attack<br><br>Verdict: PASS WITH WARNINGS</pre><p>I used to forget to check for this. Now it just happens.</p><h3>How to Get Started</h3><pre>git clone https://github.com/vadim7j7/buildmate.git<br>cd buildmate<br>python3 -m venv .venv &amp;&amp; .venv/bin/pip install -e .<br><br># Bootstrap your project<br>buildmate rails /path/to/my-app<br>buildmate nextjs /path/to/app --ui=tailwind<br>buildmate rails+nextjs /path/to/app  # fullstack<br><br># Or use a preset<br>buildmate --profile saas /path/to/app</pre><p>Then:</p><pre>/pm &quot;Build user authentication&quot;</pre><p>And watch it work.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*EsYivGf2z2SmlOPF.png" /></figure><h3>Commands Cheat Sheet</h3><pre>/pm &quot;thing&quot;			Full workflow - plan, build, test, review<br>/verify				Actually runs your code to test it<br>/verify --component Hero	Test specific component<br>/verify --endpoint /api/users	Test specific endpoint<br>/parallel &quot;a&quot; &quot;b&quot;		Do multiple things at same time<br>/new-model Name			Create model + migration + spec + factory<br>/new-component Name		Create component + test<br>/security			Security audit<br>/eval				Score the code</pre><p>The /verify one is new and honestly its my favorite now.</p><h3>FAQ</h3><p><strong>“What if the AI makes mistakes?”</strong></p><p>It will. But:</p><ul><li>Verify agent RUNS your code and catches runtime errors</li><li>If something breaks, it fixes automatically (up to 3 times)</li><li>Grind agent catches lint/type errors</li><li>Eval agent scores it so you know if its good</li></ul><p>Most AI tools just write code and pray. This one tests it.</p><p><strong>“Will it work with my project?”</strong></p><p>Yeah. It just creates a .claude/ folder. Doesn’t touch your code unless you ask.</p><p><strong>“What frameworks?”</strong></p><p>Rails, Next.js, FastAPI, React Native. More coming.</p><h3>Coming Soon: Website Cloning</h3><p>Working on something kinda crazy:</p><pre>/analyze-site https://some-cool-website.com<br>/clone-page https://some-cool-website.com/pricing</pre><p>It will look at any website, extract the design, and generate YOUR code using YOUR UI library. See a landing page you like? Clone it.</p><p>Will write a whole post about it when its ready. <strong>Follow me so you don’t miss it.</strong></p><h3>Try It</h3><pre>git clone https://github.com/vadim7j7/buildmate.git<br>cd buildmate<br>python3 -m venv .venv &amp;&amp; .venv/bin/pip install -e .<br><br>buildmate rails /path/to/your-app<br>/pm &quot;Build something cool&quot;</pre><p>Takes like 2 minutes.</p><h3>Support This Thing</h3><p>If Buildmate saved you from a 3am debugging session — maybe consider buying me a coffee?</p><p><a href="https://buymeacoffee.com/vadim7j7"><strong>☕ Buy Me A Coffee</strong></a></p><h3>Links</h3><ul><li><strong>GitHub:</strong> <a href="https://github.com/vadim7j7/buildmate">github.com/vadim7j7/buildmate</a></li><li><strong>Buy Me A Coffee:</strong> <a href="https://buymeacoffee.com/vadim7j7">buymeacoffee.com/vadim7j7</a></li></ul><p>Star the repo if you like it. And open issues if something breaks, I actually read those.</p><p><em>Built late at night with mass too much coffee</em></p><p><em>- Vadim</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3940da901888" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>