How chrismbs.link was built

At a glance

Prompts are quoted exactly as typed, typos included. That's the real process.

Why "chrismbs.link"?

"The domain is something I just came up with, in thinking it would be something that's brandable and short. I like the dot-link because it's an easy way for me to remember that it's something to be shared. It's not really like an inbound page for sharing who I am; it's really just for the goal of it being shareable resources."

Chris

• Short and brandable. Easy to say out loud in a video or type from a slide.
• .link signals the job. This domain is for sharing resources, not for introducing Chris. chrislatham.ca and THE LATHAM (thelatham.net) do that.
• MBS is a middle name Chris has used online for a long time. It wasn't given at birth; he chose it, and it stands for Mind, Body, and Soul.

Starting point: a guide worth sharing

The site started with one file: a Custom GPT to Plugin Migration Guide (version 3). It was a useful, well-built HTML guide, but it had generic corporate styling and no author.

What happened

• Claude found the brand colours in brand/creative-kit.md (Warm Paper, Vibrant Magenta, Deep Teal, Warm Crimson) and the fonts from THE LATHAM newsletter site (Space Grotesk, Inter, JetBrains Mono).
• It saved a new file and left the original untouched.
• The gradient hero became a near-black block with warm paper text and a magenta underline.
• It added a THE LATHAM top bar, a byline, a newsletter sign-up block, and a footer, all linking to thelatham.net.
• It took screenshots at desktop and phone widths to check the result.

The idea

With the rebranded page done, the real question came up: where do these pages live?

"This was done in Claude using Terminal. That helped create the guide and the plan. It gave me two options, and I picked the best option to move forward with."

Chris

What Claude did

• Loaded its Cloudflare skill and checked Cloudflare's current docs before recommending anything.
• Recommended: Cloudflare Workers with static assets, deployed from a GitHub repo, plus a small build script that reads every page and generates the homepage automatically.
• Explained the workflow: drop an HTML file in pages/, publish, and it's live. Each page's title and description become its card on the homepage.
• Covered privacy: "unlisted" pages (work by link, hidden from the homepage and Google) and truly private pages behind Cloudflare Access.
• Compared alternatives: Claude Artifacts, Cloudflare Pages drag-and-drop, Netlify/GitHub Pages, and R2 storage, each with a "good for" and a "downside".
• Asked two questions before building: which domain, and where the project should live.

The build: empty folder to live site

What Claude did, step by step

1. Checked before building. Confirmed git, the GitHub CLI, and Wrangler (Cloudflare's command-line tool) were installed and logged in. Confirmed chrismbs.link was already on Cloudflare and that nothing was live on it yet, so there was nothing to overwrite.
2. Created the project: ~/Documents/GitHub/chrismbs-link-html-library
3. Made the favicon set. Copied the avatar as-is, plus PNG versions made with the Mac's built-in image tool: 32px for browser tabs and 180px for iPhone home screens. WebP alone doesn't show up in every browser.
4. Wrote the build script (build.mjs, no outside dependencies). On every deploy it:
   • copies every page into the published folder
   • adds the favicon to any page that doesn't have one
   • reads each page's title, description, category, and date
   • generates the homepage: cards sorted newest-first, a search box, and category filters
   • generates a branded 404 page, a sitemap.xml, and a robots.txt
5. Designed the homepage and 404 in the brand colours, with a THE LATHAM sign-up block.
6. Wrote the page title and meta description: "Chris Latham's Library: Free Guides, Checklists & Tools", plus social-share tags and structured data for search engines.
7. Wrote the Cloudflare config: serve the published folder as a static site, use the 404 page for missing URLs, and attach the custom domain.
8. Added basic security headers.
9. Tested locally before going live. Ran the site on the Mac, checked every route (homepage, guide, .html → clean URL, favicon, 404, sitemap), and took screenshots at desktop and phone widths.
10. Wrote a README explaining how to add a page.
11. Saved and published the code to a private GitHub repo.
12. Deployed, and hit the first real snag.
13. Went live. Under an hour after the idea prompt.

The answer was chrismbs.link/custom-gpt-to-plugin-migration-guide. The file name minus .html becomes the URL.

Polish: credits and naming

Small prompts, each deployed and checked live within a minute or two.

Claude updated the byline and footer and deployed. The change took about 20 seconds to show up, because of Cloudflare's cache.

Claude updated it in 4 places (byline, footer, author tag, meta description), then flagged that the homepage still said "Chris Latham" and asked before changing it.

Claude updated the homepage footer (shared with the 404 page) and the author info. It deliberately left the homepage title and Google description as plain "Chris Latham", because search results often drop emoji and that's the name people search for. It flagged that choice for Chris to decide.

Saving the rules

Claude Code has a memory that carries between conversations. These prompts turned one-off decisions into standing rules.

• The note was rewritten: projects going straight to production go in ~/Documents/GitHub/; prototypes and experiments go in the existing dev-projects folder.
• The incorrect note was deleted so the two couldn't conflict.
• Claude also saved, unprompted, which Cloudflare account is Chris's, so future deploys never land in a client account.

The saved rule: every knowledge-sharing page goes on chrismbs.link with:

• a short link
• a self-referencing canonical tag
• a meta description without a byline
• category and date tags
• brand colours and fonts
• the "Created by Chris MBS Latham 🍔 with support of the AI Robots" credit
• links to THE LATHAM

Then it gets deployed, pushed, and verified live.

Short links, and the SEO lesson behind them

What happened

• Short links became a one-line tag. Each page lists its own short links, and the build writes Cloudflare's redirects file automatically:

<!-- in the page's <head> -->
<meta name="aliases" content="/gpt-to-plugin" />

• Safety checks: the build stops instead of deploying if a short link is badly formatted, uses an invalid redirect code, is already taken by another page, or would hide a real page. All four failure cases were tested on purpose.
• Tracking tags carry through: /gpt-to-plugin?utm_source=slack keeps the ?utm_source=slack on the final page.
• Removed the byline from the meta description and added the canonical tag.
• Deployed, verified live, and updated the README.

Saved as redirects-and-short-links-explained.txt.

Sharing it

A page only matters if people open it.

Claude rewrote it to lead with the December 11 deadline and the things that catch people out (Custom Actions and sharing settings don't transfer), list what's inside the guide, and use the short link.

Four new versions:

1. "I did the homework"
2. "Here's a toolkit"
3. Short and casual
4. Start a conversation

Claude recommended #4 for building relationships, if the advice in it matched what Chris was actually doing. All four were saved to slack-messages-gpt-to-plugin-guide.txt.

What's next

Claude looked through the workspace and suggested 10 pages built from material that already exists, including:

• an AI-writing-signs checklist
• "How I host all my one-page guides for free" (this story)
• a 301 vs. 302 explainer
• AI-for-marketers lessons (flagged: check ownership if it's workplace material)
• a camera-confidence checklist
• Season 1 card-to-CRM tool comparisons

Analytics

Why two tools: Chris set up Plausible and Piqo accounts himself, outside the conversation. They're the two he's using to collect data for the long term.

"…just to get data for long term. And if I ever change anything, I'll be able to see that."

Chris

• Added automatically: the build inserts both scripts into every page, including the homepage and 404, so new pages never need them pasted in.
• No double-counting: if a page already has either script, the build skips adding it.
• Verified live: both scripts appear exactly once on every page, and both load from their providers without errors.
• A catch, explained: visits through a short link count as visits to the full page, because the redirect happens before analytics loads. Use ?utm_source= tags to see which share link people used.

This is the home for everything

The site was originally built for one page, the GPT-to-Plugin guide. Then Chris realized it was the right home for other work he'd been building.

"The AI course marketing page—I've been working on that for a few months—so I realized it's a good place for this to live… I'd originally built this to do the custom GPTs to the plugin page. Then I realized there's a bunch of other stuff I've been working on that needs to be added here as well."

Chris

FREE AI Course for Marketers went live at chrismbs.link/ai-course-for-marketers (short link /ai-marketers) on Oct 15, giving months of work a public home. It was published in a different conversation, so its prompts aren't captured here.

How publishing works now

Once a page is finished and Chris confirms it, he just tells the AI to publish.

1. Copy the page into the site's pages/ folder in the GitHub project
2. Tag it with the standard tags: short link, canonical, category, date
3. Build and deploy to Cloudflare Workers
4. Commit and push to GitHub
5. Check the live page

It's live within a few minutes, with no dashboard, no file uploads, and no manual steps.

"Anytime anything gets completed on a page and I confirm, I just tell the AI to publish it, and it will copy the page to the GitHub folder. Then it will write, commit, and push it to GitHub, which will get it onto the Cloudflare workers and live within a few minutes."

Chris

The final setup

chrismbs-link-html-library/        (private GitHub repo)
├── pages/                        ← the only folder Chris normally touches
│   ├── custom-gpt-to-plugin-migration-guide.html
│   └── ai-course-for-marketers.html
├── public/                       ← favicons, homepage CSS, security headers
├── build.mjs                     ← homepage, 404, sitemap, short links, favicon + analytics
├── wrangler.jsonc                ← Cloudflare config, custom domain, pinned account
└── README.md                     ← how to add a page

Commit history: the build, as Git recorded it

Lessons from the build

Still open

• Optional: GitHub-triggered deploys (Cloudflare Workers Builds). Publishing already works through the AI. This would only matter for pushing from somewhere else, like uploading a file on GitHub from your phone.
• Origin of the GPT guide itself (v1–v3): how it was first made isn't captured yet.
• Homepage title and description still say plain "Chris Latham", on purpose, for search.
• Automatic canonical tags for new pages: offered, not built yet.