AI Automation Engineer (Claude Code) for WordPress Site Migrations

We migrate real estate brokerage websites onto a custom WordPress platform. Each job is the same shape: take the broker's existing public website and rebuild it inside the platform so it looks the same to a visitor and all the data is set correctly in the back end. We have many of these to do, so the work is repeatable by design. How the platform works It is WordPress with a custom plugin and a GeneratePress child theme. Pages are not written by hand. They are assembled from a fixed set of section types, and each page is imported as a JSON structure through the plugin's importer. Content that a client should be able to edit lives in dashboard fields and user account meta, never hardcoded in PHP. Property listings and reviews are pulled in live from third party feeds. So a "good" migration means: the right section types, the right fields populated, and a visitor cannot tell the new site from the old one. What we have built so far We run the whole migration with Claude Code (Anthropic's agentic CLI) against a written 15 step process. We have custom Python and Node tools the agent calls: a Playwright based crawler and capture tool that records every page of the old site (text, headings, fonts, colors, spacing, images, links, screenshots at three widths), a breakpoint recorder, a motion recorder for sliders and carousels, and an importer that writes pages into WordPress. Checks are scripted so numbers decide pass or fail, not the AI's opinion. There is a local dashboard that shows what the agent is doing in real time. The problems we want solved Speed. A single site takes far too long. A recent overnight session ran nine hours of wall clock for four hours of real work. Verification is the hard part. Scripted checks have reported a clean pass on pages where half the content was never inspected. Vision models miss most real visual differences. We need a comparison layer we can trust to say "this page matches the old one" without a human squinting at screenshots. The agent drifts. Left alone, it invents new tooling mid build instead of using the process, then debugs its own invention for hours. Headless browser fragility. WebKit crashes on certain third party embeds, and one bad page can kill an entire run. We want resumable, isolated, retry-safe runs. Human in the loop, batched. Right now the run stops and waits. We want it to park blockers, keep working on everything independent, and ask all questions at the end. What you would do Work with us on the workflow itself, not on one site. Review how we orchestrate Claude Code, propose and build improvements, and prove each one with numbers on a real migration. You should have Real production experience with Claude, Claude Code, or agentic AI coding workflows WordPress development: custom plugins, themes, ACF style fields, the REST API Python and Node, Playwright or Puppeteer Automated visual and content comparison, QA automation, visual regression testing Opinions about where AI agents are reliable and where they are not To apply Do not send a generic proposal. Start your reply with the word BROKER so I know you read this.