I’ve been sitting on a nagging suspicion for a while: my blog was fine for humans, but probably a mess for machines. Missing meta tags. No clean structured data. Nothing that told an AI crawler what this page actually is.
So I decided to fix it in one sitting—and I used Mythos running inside Claude Code to do the heavy lifting, pointed at the specification checklist over at https://specification.website. Here’s the exact workflow, and why I did it now instead of later.
The setup: point an agent at a spec
There’s a site at https://specification.website—”What a good website does.” It’s a checklist of what a well-built site should ship: Foundations, SEO, Accessibility, Security, Well-Known URIs, Agent Readiness, Performance, Privacy, Resilience, Internationalisation. Each topic is tagged Required or Recommended, from the boring-but-essential (<!doctype html>, <meta charset>, a real <title>) to the modern stuff (canonical URLs, theme-color, agent-readable metadata).
My instruction to the agent was deliberately simple:
Analyze my website against https://specification.website. Find everything I’m missing—required and recommended. Make a plan. Then fix it all.
That’s it. No hand-holding. I wanted to see how far an agent could get when the “spec” is an external source of truth instead of a vague “make my site better.”
It planned first, then let me decide
It didn’t just start editing files. It checked my pages against the spec and came back with a plan—a grouped list of what was missing across Foundations, SEO, Agent Readiness, Performance, and the rest. Standard agent behaviour; you’ve seen it.
The part that mattered was what came next.
I made a few decisions, then let it run
An agent shouldn’t get to decide everything. A handful of items needed a human call—things like which canonical form to standardize on, what to put in the metadata that’s genuinely mine, and how far to go on the recommended-but-optional items.
So I answered those few questions, and then it did the work: updated the files, wired in the missing pieces, and—this is the part people skip—tested it. It didn’t just claim the fixes were done. It verified the build and checked the changes held together before calling it finished.
Plan → decide → fix → verify. That loop is the whole game.
Why I did this now
Here’s the honest reason for the timing: this is a fantastic use of Mythos while it’s still cheap.
Right now, having an agent audit a whole site and ship the fixes fits inside a normal Claude subscription. That won’t last—these capabilities get more expensive as they get more valuable.
So do the one-off, high-leverage cleanup while it’s cheap. You only do it once per site, and the payoff lasts every time a crawler, search engine, or AI agent reads the site correctly from now on.
Takeaway
You don’t need to be an expert on any spec to hold your app or blog to one. Point a capable agent at a good external standard, make it plan before it edits, keep the few decisions that are actually yours, and demand it verify its own work. It applies to any codebase, not just an old one.
I did it in one sitting. And I did it now, on purpose—because getting an agent to bring something up to a standard is cheap today, and that’s exactly when you should spend it.
