<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Aleksandr Butakov</title>
    <description>The latest articles on DEV Community by Aleksandr Butakov (@saintbooth).</description>
    <link>https://dev.to/saintbooth</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3972571%2Fae68067d-c24c-48f3-946f-b167bd35c666.png</url>
      <title>DEV Community: Aleksandr Butakov</title>
      <link>https://dev.to/saintbooth</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/saintbooth"/>
    <language>en</language>
    <item>
      <title>PHP2Shell on real sites: a one-minute patch, a hidden PHP 5.6, and a host that broke the admin panel</title>
      <dc:creator>Aleksandr Butakov</dc:creator>
      <pubDate>Tue, 27 Oct 2026 11:07:18 +0000</pubDate>
      <link>https://dev.to/saintbooth/php2shell-on-real-sites-a-one-minute-patch-a-hidden-php-56-and-a-host-that-broke-the-admin-panel-3jka</link>
      <guid>https://dev.to/saintbooth/php2shell-on-real-sites-a-one-minute-patch-a-hidden-php-56-and-a-host-that-broke-the-admin-panel-3jka</guid>
      <description>&lt;p&gt;Most PHP2Shell coverage boils down to "update PHP." Fair enough, but I spent the past week doing exactly that on several production sites, and no two went the same way. One took about a minute. One turned up a PHP 5.6 binary hiding next to a patched web server. On one, the hosting provider's emergency fix broke the admin panel on the morning of the advisory, before the advisory was even published.&lt;/p&gt;

&lt;p&gt;If you run or maintain PHP sites, especially ones that haven't had their PHP version touched in years, here's what each of those looked like and what to do when "just update" isn't a one-liner.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;TL;DR.&lt;/strong&gt; PHP2Shell is two critical bugs in PHP itself. Updating WordPress or your CMS fixes nothing. Fixed releases are 8.2.35, 8.3.36, 8.4.27 and 8.5.12. On a supported branch it's a short maintenance task. On 7.4 or 8.1 it's a migration, and the interim hardening the PHP team lists does not make you safe.&lt;/p&gt;

&lt;h2&gt;
  
  
  What PHP2Shell is, in one screen
&lt;/h2&gt;

&lt;p&gt;PHP published &lt;a href="https://www.php.net/archive/2026.php#2026-10-20-1" rel="noopener noreferrer"&gt;the advisory on October 21, 2026&lt;/a&gt;. The press named the pair PHP2Shell:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-109618, "RCE via OOM corruption":&lt;/strong&gt; an out-of-memory error can interrupt PHP while internal request data is only half set up, and cleanup then treats that broken data as valid (&lt;a href="https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w" rel="noopener noreferrer"&gt;GHSA-w8hw-w9fx-j79w&lt;/a&gt;).&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;CVE-2026-109619, "RCE via stale global state":&lt;/strong&gt; state that belongs to one request can outlive its memory, and the next request handled by the same worker may use it (&lt;a href="https://github.com/php/php-src/security/advisories/GHSA-6wmf-4rv5-p3rp" rel="noopener noreferrer"&gt;GHSA-6wmf-4rv5-p3rp&lt;/a&gt;).&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Both are unauthenticated remote code execution, rated 9.2 on CVSS v4.0. According to the PHP team, every installation that serves web requests is at risk: PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed. The risk is rated highest for mod_php, where a demonstrated exploit fires before any application code runs. That's why a security plugin doesn't help: it never gets a say. On October 26 CISA added both CVEs to its &lt;a href="https://www.cisa.gov/known-exploited-vulnerabilities-catalog" rel="noopener noreferrer"&gt;Known Exploited Vulnerabilities catalog&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Don't mix it up with CVE-2026-111497. That one is a bug in the Brotli decompression library. PHP rebuilt only its Windows packages for it, and per the PHP team the vulnerable code is reachable through the cURL extension with automatic response decompression enabled. That's off by default in cURL, though applications and HTTP client libraries can turn it on.&lt;/p&gt;

&lt;h2&gt;
  
  
  First, find the PHP that actually serves your site
&lt;/h2&gt;

&lt;p&gt;The version that matters is the one handling web requests. It isn't always the one you see first.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;WordPress:&lt;/strong&gt; Tools → Site Health → Info → Server.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Response headers:&lt;/strong&gt; many servers report it in &lt;code&gt;X-Powered-By&lt;/code&gt;. On a cached site, add a throwaway query parameter so the response comes from PHP, not from a stored page.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;phpinfo():&lt;/strong&gt; a one-line &lt;code&gt;&amp;lt;?php phpinfo();&lt;/code&gt; file shows everything. Delete it right after. It's a gift to anyone scanning your site.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;As of late October 2026, anything below 8.2.35, 8.3.36, 8.4.27 or 8.5.12 within its branch is vulnerable, and so is every 7.x, 8.0 and 8.1. One caveat: Linux distributions and LTS vendors got the patches in advance, and their builds can keep an older version number after backporting the fix. If you see a distro 8.3.x, read the vendor's security notice before deciding either way. On shared hosting, ask the host.&lt;/p&gt;

&lt;h2&gt;
  
  
  Site 1: supported branch, own server, about a minute
&lt;/h2&gt;

&lt;p&gt;A store on Bitrix (the dominant commercial PHP CMS in Russia) with a CRM attached, running PHP 8.3.31.&lt;/p&gt;

&lt;p&gt;The update to 8.3.36 meant installing packages from the repository the server already used and restarting Apache. About a minute of actual work. The checks took longer:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;the module list was the same 59 before and after;&lt;/li&gt;
&lt;li&gt;
&lt;code&gt;php.ini&lt;/code&gt; was untouched;&lt;/li&gt;
&lt;li&gt;a custom opcache config survived, and the package's new default landed next to it as &lt;code&gt;.rpmnew&lt;/code&gt;;&lt;/li&gt;
&lt;li&gt;key pages and the admin panel returned 200.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Cron imports got a separate check the next morning, on their first run under the new version. Background jobs are where PHP problems surface last.&lt;/p&gt;

&lt;p&gt;On a supported branch, that's the whole cost of PHP2Shell: a short maintenance task.&lt;/p&gt;

&lt;h2&gt;
  
  
  Site 2: the host patched the web, the CLI was PHP 5.6
&lt;/h2&gt;

&lt;p&gt;A Bitrix site on shared hosting. The provider had already moved PHP-FPM to 8.2.35. I confirmed it from the &lt;code&gt;X-Powered-By&lt;/code&gt; header on several pages with the cache bypassed. Done, apparently.&lt;/p&gt;

&lt;p&gt;Then &lt;code&gt;php -v&lt;/code&gt; over SSH on the same server: PHP 5.6.&lt;/p&gt;

&lt;p&gt;The command-line binary and the one behind the web server are different PHPs, and nothing forces them to match. PHP 5 has had no security fixes for years. The advisory lists only web-facing setups (FPM, mod_php, CGI and the like), so the CLI isn't the PHP2Shell attack surface. But cron jobs and scripts run through it, so it belongs in your inventory. Checking only &lt;code&gt;php -v&lt;/code&gt; can mislead you in either direction.&lt;/p&gt;

&lt;h2&gt;
  
  
  Site 3: the host's emergency fix broke the admin panel
&lt;/h2&gt;

&lt;p&gt;On the morning of October 21, before the PHP team had published the advisory, a shared host restricted upload progress and blocked every POST request with the &lt;code&gt;multipart/form-data&lt;/code&gt; content type. Nobody warned the site owner. On that Bitrix store, product cards in the admin panel stopped saving, because the edit form submits as multipart.&lt;/p&gt;

&lt;p&gt;The first instinct with a sudden 403 is to blame the site: a module update, the CMS's own protection, a bad &lt;code&gt;.htaccess&lt;/code&gt;. Two curl requests settle it:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;&lt;span class="c"&gt;# A regular form POST goes through&lt;/span&gt;
curl -s -o /dev/null -w &lt;span class="s2"&gt;"%{http_code}\n"&lt;/span&gt; -X POST -d &lt;span class="s2"&gt;"a=1"&lt;/span&gt; https://example.com/
&lt;span class="c"&gt;# 200&lt;/span&gt;

&lt;span class="c"&gt;# The same POST as multipart/form-data is rejected&lt;/span&gt;
curl -s -o /dev/null -w &lt;span class="s2"&gt;"%{http_code}\n"&lt;/span&gt; -F &lt;span class="s2"&gt;"a=1"&lt;/span&gt; &lt;span class="s2"&gt;"https://example.com/?probe=test123"&lt;/span&gt;
&lt;span class="c"&gt;# 403&lt;/span&gt;

&lt;span class="c"&gt;# The probe never reached the site's access log,&lt;/span&gt;
&lt;span class="c"&gt;# so the request was cut off before Apache and PHP&lt;/span&gt;
grep &lt;span class="s2"&gt;"probe=test123"&lt;/span&gt; /path/to/site/access.log
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The 403 came from the host's nginx on every URL, static files included, and the marked request never showed up in the site's own log. That rules out &lt;code&gt;.htaccess&lt;/code&gt;, the CMS and the site's security settings.&lt;/p&gt;

&lt;p&gt;Support later explained the logic: the restrictions were temporary, until patches came out and PHP was updated. Patches usually exist only for supported versions, so on older ones they couldn't guarantee a fix, and recommended PHP 8.2 or newer after checking compatibility. On a VPS, the risk and the update are entirely the owner's job.&lt;/p&gt;

&lt;p&gt;In effect the host did what the PHP advisory suggests, only harder: instead of tuning upload limits, it shut multipart off entirely. Fair from their side, since they protect the platform, and your admin panel isn't their concern. Painful for a store that needs to edit products. On that site I updated PHP to 8.5.12, the latest fixed release, step by step, with a backup and a check at every stage.&lt;/p&gt;

&lt;p&gt;So "the host will handle it" is half true. For supported branches it often will. For old ones the host said it plainly: no guarantees. And its fix can cost you a working admin panel for a while.&lt;/p&gt;

&lt;h2&gt;
  
  
  If you're on 8.1 or older
&lt;/h2&gt;

&lt;p&gt;The PHP project no longer ships updates for these branches. It did publish minimal source backports for 7.4, 8.0 and 8.1, which help if you build PHP from source and have no LTS vendor. Distributions and LTS vendors ship their own patched packages, so ask yours. For a typical small site, the real fix is moving to a supported branch.&lt;/p&gt;

&lt;p&gt;Until then, the PHP team lists interim settings, with an explicit warning that only an update is a complete solution:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight ini"&gt;&lt;code&gt;&lt;span class="c1"&gt;; Based on the PHP advisory. Set at the server level (php.ini, FPM pool,
; Apache config), not via ini_set() or .user.ini, then restart every worker.
&lt;/span&gt;&lt;span class="na"&gt;memory_limit&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="o"&gt;-&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt;
&lt;span class="na"&gt;post_max_size&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;8M&lt;/span&gt;          &lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="mi"&gt;32M&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;the&lt;/span&gt; &lt;span class="n"&gt;site&lt;/span&gt; &lt;span class="n"&gt;needs&lt;/span&gt; &lt;span class="n"&gt;file&lt;/span&gt; &lt;span class="n"&gt;uploads&lt;/span&gt;
&lt;span class="na"&gt;upload_max_filesize&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;4M&lt;/span&gt;    &lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="n"&gt;must&lt;/span&gt; &lt;span class="n"&gt;stay&lt;/span&gt; &lt;span class="n"&gt;below&lt;/span&gt; &lt;span class="n"&gt;post_max_size&lt;/span&gt;
&lt;span class="na"&gt;session.upload_progress.enabled&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;
&lt;span class="na"&gt;fatal_error_backtraces&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;  &lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="no"&gt;PHP&lt;/span&gt; &lt;span class="mf"&gt;8.5&lt;/span&gt; &lt;span class="n"&gt;only&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;&lt;code&gt;memory_limit = -1&lt;/code&gt; closes one attack path and opens the door to memory exhaustion, so put a container or cgroup limit around the worker alongside it. This block buys days, not months.&lt;/p&gt;

&lt;h2&gt;
  
  
  What breaks on the jump to PHP 8
&lt;/h2&gt;

&lt;p&gt;WordPress or Bitrix core usually copes with a new PHP. What breaks is your own code and old plugins: custom modules, template overrides, event handlers written for PHP 7 and never touched since.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;code&gt;create_function()&lt;/code&gt; and &lt;code&gt;each()&lt;/code&gt; were removed in PHP 8.0. Code that calls them dies with a fatal error.&lt;/li&gt;
&lt;li&gt;PHP 8 stopped quietly coercing types. &lt;code&gt;array_key_exists()&lt;/code&gt; on &lt;code&gt;null&lt;/code&gt; or &lt;code&gt;count()&lt;/code&gt; on a non-array now throws a &lt;code&gt;TypeError&lt;/code&gt;, which in Bitrix often shows up as a 500 in the admin panel.&lt;/li&gt;
&lt;li&gt;Passing &lt;code&gt;null&lt;/code&gt; to built-in string functions is deprecated since 8.1, dynamic properties since 8.2. Deprecations are noise to clean up later. A fatal error is the outage.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The order that works for me: a copy of the site, the error log, fatal errors one at a time, then cron jobs and imports separately, because they fail in the background long after the homepage looks fine. Longer walkthroughs: &lt;a href="https://butakov.dev/journal/wordpress-php83-migration" rel="noopener noreferrer"&gt;WordPress after a PHP 8.3 switch&lt;/a&gt; and &lt;a href="https://butakov.dev/journal/bitrix-php8-legacy" rel="noopener noreferrer"&gt;legacy Bitrix on PHP 8&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why this one costs some sites so much
&lt;/h2&gt;

&lt;p&gt;Critical bugs happen in every branch of every language. If every site ran a supported PHP, PHP2Shell would be a routine maintenance day. It gets expensive only for sites that haven't been updated in five years. I think of it less as a bug than as a bill for those five years.&lt;/p&gt;

&lt;p&gt;Around ten of the client sites I look after run on Bitrix and WordPress and haven't changed PHP versions in years. The reasons repeat: "it works, don't touch it," not knowing it matters, nobody responsible for it, no budget to pay someone who would be.&lt;/p&gt;

&lt;p&gt;The strongest objection I hear: "We've been on PHP 7.4 for five years and never got hacked." True, and it made sense for a while. In my experience, most attacks on small sites came through plugins and site code, and a plugin update or a security plugin covered a lot of it. PHP2Shell sits below all of that. It fires before the application runs, and the PHP project has no regular release to ship for 7.4.&lt;/p&gt;

&lt;p&gt;The bill keeps growing on its own. A jump from 7.4 to 8.x breaks far more code than 8.2 to 8.3, because the big break came in 8.0, with more deprecations in 8.1 and 8.2. And PHP 8.2 itself reaches end of life on December 31, 2026, so today's "safe" version is two months from becoming the next old one. Small updates on a schedule cost less than one big update during a fire.&lt;/p&gt;

&lt;p&gt;An update closes the hole but doesn't remove anything already installed. If a site may have been compromised before you patched, that's a separate job. Here's &lt;a href="https://butakov.dev/journal/wordpress-keeps-getting-hacked" rel="noopener noreferrer"&gt;what a backdoor looks like after a "cleanup" and how to find it&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If your host did something similar on October 21, I'm curious how widespread the multipart block was. Tell me in the comments.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published on &lt;a href="https://butakov.dev/journal/php2shell-old-php-sites" rel="noopener noreferrer"&gt;butakov.dev&lt;/a&gt;. Written with AI assistance; the sites, the fixes and every claim here are mine and checked against the PHP advisory.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>php</category>
      <category>security</category>
      <category>wordpress</category>
      <category>webdev</category>
    </item>
    <item>
      <title>I replaced most of a dev team with Claude Code to build a production e-commerce store. Here's what broke.</title>
      <dc:creator>Aleksandr Butakov</dc:creator>
      <pubDate>Mon, 14 Sep 2026 05:23:10 +0000</pubDate>
      <link>https://dev.to/saintbooth/i-replaced-most-of-a-dev-team-with-claude-code-to-build-a-production-e-commerce-store-heres-what-gen</link>
      <guid>https://dev.to/saintbooth/i-replaced-most-of-a-dev-team-with-claude-code-to-build-a-production-e-commerce-store-heres-what-gen</guid>
      <description>&lt;p&gt;First post here. Figured a real project beats an introduction post, so: here's one, warts included.&lt;/p&gt;

&lt;p&gt;A women's clothing brand needed an online store synced to their inventory system, matched to an existing React prototype pixel for pixel, on a budget that ruled out the usual enterprise platforms. I built it solo: WordPress, WooCommerce, and Claude Code writing most of the code while I did the architecture and reviewed everything it touched. Not the demo-reel version of AI coding. The one with the actual bugs.&lt;/p&gt;

&lt;h2&gt;
  
  
  The default answer was too expensive, and it didn't fit the design
&lt;/h2&gt;

&lt;p&gt;The client's inventory lives in 1C, the ERP and point-of-sale system that runs most Russian retail, a corporate accounting platform that only talks to the outside world through XML. Ask around for how to sync a site with it and you get the same answer every time: build on Bitrix, the dominant commercial CMS there, which ships a stock exchange module and a pool of integrators who already know how to bill for it. Every market has its own version of this. Whatever the local business-software incumbent is, some platform built a plugin for it years ago, and agencies default to it because that's what they already know how to staff.&lt;/p&gt;

&lt;p&gt;For a one-location boutique, that default breaks on two fronts.&lt;/p&gt;

&lt;p&gt;Cost first. License, annual renewal, integrator fees: add it up and you're close to the entire budget of this open source build, before a single feature exists.&lt;/p&gt;

&lt;p&gt;Then design. The client had a React prototype already, a specific quiet-luxury look, four colors, two typefaces, zero border radius everywhere, and wanted it copied exactly, not approximated. Forcing that onto a stock theme takes longer than writing one from scratch.&lt;/p&gt;

&lt;p&gt;None of the actual requirements got smaller because the budget did: real-time stock from an ERP where every size-color pair carries its own count, card and instant-payment processing through a local processor, courier delivery, a cookie banner and consent checkboxes for local privacy law, ecommerce event tracking. All of it, on a budget that assumed almost none of it.&lt;/p&gt;

&lt;h2&gt;
  
  
  The stack
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;CMS: WordPress with a fully block-based theme (FSE) built from scratch, no page builder plugins&lt;/li&gt;
&lt;li&gt;E-commerce: WooCommerce and WooCommerce Blocks, mini cart through the Store API&lt;/li&gt;
&lt;li&gt;Backend: PHP 8, a modular &lt;code&gt;inc/&lt;/code&gt; structure, custom fields registered in code instead of through an admin UI&lt;/li&gt;
&lt;li&gt;Frontend: plain ES modules, no jQuery, &lt;code&gt;IntersectionObserver&lt;/code&gt; for scroll animations, &lt;code&gt;history.pushState&lt;/code&gt; for filter state&lt;/li&gt;
&lt;li&gt;Design tokens: &lt;code&gt;theme.json&lt;/code&gt; as the single source of truth, CSS variables generated from it&lt;/li&gt;
&lt;li&gt;ERP sync: CommerceML protocol, real time&lt;/li&gt;
&lt;li&gt;AI tooling: Claude Code, plus MCP servers for Playwright browser checks and SSH deploys&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  A block theme instead of a page builder
&lt;/h3&gt;

&lt;p&gt;No Elementor, no page builder plugin of any kind. On a budget project that's arithmetic, not taste: every builder plugin costs you performance, ties you to somebody else's release schedule, and inserts a layer between the design and the code you actually control.&lt;/p&gt;

&lt;p&gt;Design tokens moved straight into &lt;code&gt;theme.json&lt;/code&gt;, WordPress's native token file. Four colors, an 8px spacing scale, two typefaces. WordPress turns that into CSS variables on its own, and every stylesheet touches only those variables, never a raw value:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight css"&gt;&lt;code&gt;&lt;span class="nd"&gt;:root&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="py"&gt;--color-bone&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="c"&gt;/* background, roughly 70% of the canvas */&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="py"&gt;--color-ink&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="c"&gt;/* text and UI */&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="py"&gt;--color-bordeaux&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="c"&gt;/* sale badges and accents */&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="py"&gt;--color-pine&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="c"&gt;/* dark navigation */&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Border radius sits at zero everywhere except pill-shaped tags. That one rule caused more arguments with the agent than anything else on the project, more on that in a second.&lt;/p&gt;

&lt;p&gt;&lt;code&gt;functions.php&lt;/code&gt; does nothing but wire up modules:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight php"&gt;&lt;code&gt;&lt;span class="c1"&gt;// functions.php: wiring only, no logic&lt;/span&gt;
&lt;span class="k"&gt;require_once&lt;/span&gt; &lt;span class="nf"&gt;get_theme_file_path&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt; &lt;span class="s1"&gt;'inc/setup.php'&lt;/span&gt; &lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;require_once&lt;/span&gt; &lt;span class="nf"&gt;get_theme_file_path&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt; &lt;span class="s1"&gt;'inc/assets.php'&lt;/span&gt; &lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;require_once&lt;/span&gt; &lt;span class="nf"&gt;get_theme_file_path&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt; &lt;span class="s1"&gt;'inc/woocommerce.php'&lt;/span&gt; &lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;require_once&lt;/span&gt; &lt;span class="nf"&gt;get_theme_file_path&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt; &lt;span class="s1"&gt;'inc/ajax-handlers.php'&lt;/span&gt; &lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;One file, one job. Repeating sections became block patterns, header and footer became template parts, and WooCommerce's default templates got overridden only where the stock markup broke the design.&lt;/p&gt;

&lt;h2&gt;
  
  
  What vibe coding looks like on a real project
&lt;/h2&gt;

&lt;p&gt;The agent didn't build this alone. It wrote most of the code; I did the architecture, broke the work into tasks, and reviewed what came back. That's not less work than building it by hand. It's the same amount of thinking, compressed into review instead of typing.&lt;/p&gt;

&lt;p&gt;Three things kept it from falling apart, and the first one mattered more than I expected going in.&lt;/p&gt;

&lt;p&gt;Claude Code reads a project instructions file, &lt;code&gt;CLAUDE.md&lt;/code&gt;, at the start of every session. I put the entire design system in there as flat prohibitions: exactly four colors, no raw hex anywhere in the codebase, radius is zero, spacing only off the 8px scale. Leave an LLM alone and it improvises constantly, cheerfully, and exactly the way that wrecks a tight design system. With the rule sitting there in black and white, when the agent needs a new shade it doesn't reach for a hex code, it derives one with &lt;code&gt;oklch()&lt;/code&gt; off an existing token, because the file tells it that's the only move on the table.&lt;/p&gt;

&lt;p&gt;Memory came next, because the agent's context resets and this project didn't, it ran for months. Every real lesson, "WooCommerce Blocks doesn't load jQuery," "the ERP's delta exports quietly wipe out product variations," got written down in a markdown file the agent rereads on the next session. New session, same traps already mapped.&lt;/p&gt;

&lt;p&gt;And then verification, which I'd fight hardest to keep if someone told me to cut a corner. The agent has Playwright access through MCP: it opens the live page itself, clicks through the actual flow, takes a screenshot, and compares it to the prototype, instead of me squinting at a screenshot and taking its word for it. Rule: no "fixed" without a real check behind it. That rule paid for itself on a bug where product card links quietly stopped working, only on desktop, only in production. The agent reproduced it with an actual click through Playwright and traced it to &lt;code&gt;setPointerCapture&lt;/code&gt; inside the image carousel, which was grabbing pointer events and redirecting the click from the product link onto the carousel track underneath it. Nobody was finding that by staring at the carousel code. It looked fine. It was fine, mostly.&lt;/p&gt;

&lt;p&gt;Where it falls down, and I mean this as a flat statement, not a complaint: it will confidently fix the wrong root cause if you don't force a reproduction first. It drifts outside the design system the second a rule gets fuzzy instead of explicit. It loses the thread on anything long without something external reminding it what already happened. All three are process failures, not model failures.&lt;/p&gt;

&lt;h2&gt;
  
  
  The ERP integration: the part nobody warns you about
&lt;/h2&gt;

&lt;p&gt;This ate more hours than everything else combined, and it's the part that matters most if you're wiring any ERP or POS system into an ecommerce platform over an XML feed, not just this stack.&lt;/p&gt;

&lt;p&gt;The sync runs on CommerceML: the ERP exports packets on a schedule and on events, a plugin on the WordPress side parses them, catalog updates. Simple as a diagram. The traps are all in the details, and since CommerceML is a Russian national standard, the XML element names in the snippet below are genuinely Russian words. Not obfuscation, just the protocol.&lt;/p&gt;

&lt;p&gt;First bug: stock counts that weren't real. The storefront showed wrong numbers even though the sync ran clean, no errors, no warnings, everything green. Took diffing raw XML against what actually landed in the database to find it. In the protocol version we were on, stock arrives in separate files with a nested structure, split per warehouse:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight xml"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;Предложение&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;Ид&amp;gt;&lt;/span&gt;product-guid#variation-guid&lt;span class="nt"&gt;&amp;lt;/Ид&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;Остатки&amp;gt;&lt;/span&gt;
    &lt;span class="nt"&gt;&amp;lt;Остаток&amp;gt;&lt;/span&gt;
      &lt;span class="nt"&gt;&amp;lt;Склад&amp;gt;&lt;/span&gt;
        &lt;span class="nt"&gt;&amp;lt;Ид&amp;gt;&lt;/span&gt;warehouse-guid&lt;span class="nt"&gt;&amp;lt;/Ид&amp;gt;&lt;/span&gt;
        &lt;span class="nt"&gt;&amp;lt;Количество&amp;gt;&lt;/span&gt;3&lt;span class="nt"&gt;&amp;lt;/Количество&amp;gt;&lt;/span&gt;
      &lt;span class="nt"&gt;&amp;lt;/Склад&amp;gt;&lt;/span&gt;
    &lt;span class="nt"&gt;&amp;lt;/Остаток&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;/Остатки&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;/Предложение&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The plugin's parser expected a flat quantity field one level up and silently skipped the nested version. No error anywhere in the pipeline. The sync reports success. The data is just wrong, and nothing tells you that. Patched the parser, and I've diffed the plugin's code before every update since, because the next release could bring that bug back without so much as a changelog note.&lt;/p&gt;

&lt;p&gt;Two more surfaced close together. Product properties and variation characteristics turned out to be handled as two unrelated mechanisms in this protocol, separate code paths entirely, so a color set up as a property and a size set up as a characteristic land in WooCommerce through completely different logic. And the fast sync sends delta chunks instead of the full catalog on every run, which sounds efficient right up until you notice the plugin was rebuilding each product's entire variation set from whatever chunk it received, meaning any variation missing from that one chunk just vanished. The size range was draining out from under us for I don't know how long before anyone noticed, probably the client before me, honestly. Fixed with a mode that preserves variations across partial exports, plus a cleanup pass, since roughly a thousand orphaned variations had piled up in the database by the time we caught it.&lt;/p&gt;

&lt;p&gt;Last one, smaller: attribute terms were getting created with GUID slugs instead of readable ones, which broke filter URLs and made the admin panel borderline unreadable. Mapping table, repair script that runs after every sync, done.&lt;/p&gt;

&lt;p&gt;Honest takeaway: an ERP-to-WordPress sync can run reliably in production. Ours does, real time. But "install the plugin and walk away" was never on the table. Budget the time to read the plugin's source, patch its parser, and write your own verification scripts, because the vendor's tests, if they exist, aren't testing for your data.&lt;/p&gt;

&lt;h2&gt;
  
  
  WooCommerce Blocks quietly dropped jQuery, and a lot of tutorials haven't caught up
&lt;/h2&gt;

&lt;p&gt;Modern WooCommerce with block templates does not load jQuery on the storefront. At all. Every classic tutorial snippet built on &lt;code&gt;$('.variations_form').on('found_variation', …)&lt;/code&gt; has nothing left to hook into anymore, and there are still a lot of those tutorials ranking on page one. For a block theme I had to write a native variation resolver from scratch: collect the selected attributes, match them against the variation data, update price, photo, availability.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="c1"&gt;// No jQuery here: WooCommerce Blocks doesn't load it.&lt;/span&gt;
&lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="nx"&gt;match&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;variations&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;find&lt;/span&gt;&lt;span class="p"&gt;((&lt;/span&gt;&lt;span class="nx"&gt;v&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt;
  &lt;span class="nb"&gt;Object&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;entries&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;selected&lt;/span&gt;&lt;span class="p"&gt;).&lt;/span&gt;&lt;span class="nf"&gt;every&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="p"&gt;([&lt;/span&gt;&lt;span class="nx"&gt;attr&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nx"&gt;value&lt;/span&gt;&lt;span class="p"&gt;])&lt;/span&gt; &lt;span class="o"&gt;=&amp;gt;&lt;/span&gt; &lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="nx"&gt;v&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;attributes&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;attr&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;||&lt;/span&gt; &lt;span class="nx"&gt;v&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;attributes&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nx"&gt;attr&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt; &lt;span class="o"&gt;===&lt;/span&gt; &lt;span class="nx"&gt;value&lt;/span&gt;
  &lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Same story everywhere else in the frontend. AJAX catalog filters render inside the block product template. Filter state lives in the URL through &lt;code&gt;history.pushState&lt;/code&gt;, so a filtered link survives being forwarded to someone else. The mini cart runs on the Store API. Scroll animations use &lt;code&gt;IntersectionObserver&lt;/code&gt; and respect &lt;code&gt;prefers-reduced-motion&lt;/code&gt;. None of it is exotic, it's just modern frontend work that WordPress still makes you assemble by hand.&lt;/p&gt;

&lt;h2&gt;
  
  
  What shipped
&lt;/h2&gt;

&lt;p&gt;A working store on a custom block theme that matches the prototype, genuinely, not "close enough." Catalog with AJAX filtering, variable product cards with size and color options, a wishlist, product pages with a swipe gallery on mobile.&lt;/p&gt;

&lt;p&gt;Real-time sync with the ERP. Prices, stock, and variations arrive without anyone touching a spreadsheet, and after the stock parser fix, the storefront numbers actually match the warehouse.&lt;/p&gt;

&lt;p&gt;The client sent two full rounds of revisions, about thirty items between them, everything from typography tweaks to restructuring entire blocks. Each round cleared in a few days, not weeks. Total build time landed in weeks, not months.&lt;/p&gt;

&lt;p&gt;And the budget number holds up: zero dollars in platform licenses. WordPress, WooCommerce, the rest of the stack, all open source. The client's money went to actual work and hosting instead of a renewal invoice for a boxed platform.&lt;/p&gt;

&lt;h2&gt;
  
  
  When this fits, and when it doesn't
&lt;/h2&gt;

&lt;p&gt;Fits a small or mid-size store: custom design, an ERP that isn't a deep enterprise system, a budget that actually constrains the platform choice, and a developer willing to read exchange XML instead of just clicking through a plugin's settings screen.&lt;/p&gt;

&lt;p&gt;Doesn't fit a high-volume marketplace pushing thousands of orders a day. Doesn't fit real-time multi-warehouse reservation logic. Doesn't fit a client whose procurement process demands vendor support contracts on paper.&lt;/p&gt;

&lt;p&gt;On the AI part specifically: Claude Code didn't replace a developer here. It replaced part of a team. I wouldn't have shipped something this size alone, on this timeline, without it, or I'd have shipped it on a completely different budget, one of those two. But hand the same agent a project with no rules file, no memory between sessions, and no requirement to verify its own claims, and it will produce code that's confident and wrong in roughly equal measure. Building that process around it mattered more than any prompt I ever wrote.&lt;/p&gt;




&lt;p&gt;That's the first one. I build web stores and internal tools this way pretty regularly, usually pairing solo development with AI agents to hit budgets a full team can't match. More of this at &lt;a href="https://butakov.dev" rel="noopener noreferrer"&gt;butakov.dev&lt;/a&gt; if you want it, and happy to answer questions about the ERP side or the Claude Code setup in the comments.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>webdev</category>
      <category>programming</category>
      <category>tutorial</category>
    </item>
  </channel>
</rss>
