AI briefing - building sites

Best practice for AI agents creating or maintaining sites on the lazysite engine - the separation-of-concerns model and the failure modes to avoid.

Who this is for

A guide for AI agents (and the people directing them) creating or maintaining sites on the lazysite engine. It captures the separation-of-concerns model the engine is built around, and the failure modes that show up when an agent treats site-building like generating a single web page.

This is the "how to think about a lazysite site" briefing. For the mechanics see its siblings: content authoring, layouts and themes, publishing over WebDAV / the control API, and configuration.

The one principle: three layers that never touch

lazysite is built on a strict separation. Keep it.

The engine renders content through a layout, styled by a theme, and caches the result. Three sanity checks tell you whether you have kept them separate:

1. Could a non-technical person edit the page's words without seeing any HTML or CSS?
2. Could you restyle the entire site by swapping the theme, changing nothing else?
3. Could you reuse the layout on a completely different page?

If any answer is "no", you have fused layers that should be apart.

The anti-pattern: the monolithic all-in-one page

The most common mistake an AI makes is to generate one big HTML file with the markup, all the CSS, and the content fused together - a "web page", not a site.

Case study

united.explore.lazysite.io/lazysite-assets/summit-of-summits/ - a single page built by an assistant:

• 51 KB in one HTML file; 16.8 KB of CSS inlined in a single <style>.
• 108 <div>s, 10 <section>s of bespoke structure.
• 47 hardcoded hex colours (mixed in with 117 var(--theme-*) refs - so it half-knew about tokens, then hardcoded anyway).
• Fonts loaded directly from Google; nothing shared.
• Served from /lazysite-assets/… - the static-assets path. It bypasses the rendering engine entirely: no layout.tt, no site nav, no theme system, no cache pipeline, not registered in sitemap/llms/feeds.

Why that is wrong

• Not reusable. Nothing in those 16.8 KB helps the next page. Every new page restarts from zero.
• Cannot be restyled. A theme swap does nothing - the look is welded into the page, and the hardcoded colours ignore the tokens.
• Not maintainable or authorable. No one can edit the words without wading through markup. It is invisible to the manager UI and to AI publishing rules.
• Heavier and slower. 16.8 KB of CSS ships on this page alone instead of once, cached, in a shared main.css.
• Off the rails. As a raw asset it gets no nav, no canonical URL, no feed, no llms.txt entry - it is a dead end in the site.

What it should have been

That page is a look (a distinctive editorial theme) plus content. So:

• The distinctive visual style -> a theme (summit): its palette, fonts and spacing as tokens in theme.json; shared CSS in main.css.
• The page structure/hero/section rhythm -> a layout (or features added to an existing layout), reusable by any page that picks it.
• The words -> a Markdown page served by the engine, registered for discovery, restylable by swapping the theme.

Result: the same design, but reusable across a whole site, themeable, authorable, cached once, and on the rails.

The root cause: raw mode for ordinary pages

The monolith almost always traces back to one front-matter switch: api: true (equivalently raw: true). That is raw mode - the engine serves the file's body verbatim and skips the layout and the theme. A page built that way has to carry its own header, nav, footer and every line of CSS, because the shared chrome and the design tokens are never applied to it. Put ordinary pages in raw mode and you are hand-manufacturing monoliths, one per page.

Raw mode exists for a genuinely self-contained artifact - a single bespoke interactive widget or an embed fragment with its own content type. It is not for content. For a normal page, write plain Markdown (title/subtitle front matter, a prose body) and let the engine wrap it in the shared layout + theme.

Most of the "mechanical" busywork that raw-mode publishing seems to demand is a symptom that disappears the moment a page is Markdown:

• Rehoming embedded images into img/ paths - unnecessary. In Markdown you write !alt and upload the image once; images are only ever "embedded" because something authored a self-contained HTML blob.
• Reading the file back in ~200-line ranges so it doesn't truncate - unnecessary. A Markdown page is a screen or two. And do not read the source back to check your work at all: render it - preview_page (server-side render) or a plain GET of the page URL. Chunked source re-reads are a workaround for a problem the approach created.
• A page taking longer than a single upload - it gets faster, because you stop rebuilding the whole design on every page.

Build the design once; pages are then trivial

• Once per site: define the look as a theme (colour/font/spacing tokens) plus a layout, and activate them. That is where all the CSS and structure live - shipped once, cached once.
• Per page: a small .md - front matter, words, and ![](img/…). Upload the page (and any new image), clear the cache, verify with a rendered GET. Seconds, not surgery.
• Repeated blocks (cards, listings, galleries): a JSON data file + a loop, not copy-pasted HTML - so reordering can't break the markup and the data stays editable by anyone.

The scaling maths is the punchline. Monolith pattern: every page re-ships its own copy of the design (tens of KB), every edit is big-file surgery, and a restyle means touching every page - cost grows with the page count. Theme + layout pattern: the design is defined once (O(1)), each page is a few KB of content, and restyling the whole site is a single theme swap. As a site - or an estate of sites - scales, those two curves diverge hard: one gets slower and heavier per page, the other stays flat.

Rules of thumb

1. Content is Markdown - never raw mode. The body of a page is prose in .md, wrapped by the engine in the shared layout + theme. Do not set api: true / raw: true on a content page: that is raw mode, which serves the body verbatim with no layout and no theme, and is the usual source of monolith pages. Reserve raw mode for a genuinely self-contained artifact. If you are writing paragraphs of HTML in a page, stop - it belongs in the layout.
2. Structure is the layout. Header, nav, footer, hero, section scaffolding live in layout.tt. Build them once; every page inherits them.
3. Style is the theme, as tokens. Use var(--theme-*); never hardcode a colour or font in a page. A theme swap must be able to restyle everything.
4. Repeated blocks are data-driven, not copy-pasted HTML. Cards, feature grids, comparison tables, listings -> a JSON data file rendered with a FOREACH loop (or a components/*.tt component), so reordering can't break the markup and non-technical people can edit the data. Hand-written repeated <div>s are a smell.
5. A new capability is a reusable layout feature, not a one-off page hack. Needed a video hero? Add it to the layout, driven by a front-matter key (tt_page_var), so any page can opt in - don't bolt bespoke markup into a single page.
6. Serve through the engine. Never drop hand-authored HTML into /lazysite-assets/ (or any static path) as a "page". Assets are for images, video, fonts, main.css - not documents.
7. Minimal page-level inline HTML/CSS. A small one-off touch is fine; the bulk of styling and structure belongs in theme/layout. If a page's <style> is growing past a few lines, it wants to be in the theme.
8. Reuse before you build. Check the theme gallery and existing layouts first. A new look is a new theme, usually on an existing layout - rarely a new page type.
9. Register for discovery. Put pages in sitemap.xml / llms.txt, emit feeds where relevant. Server-rendered semantic HTML is what makes lazysite sites findable by search and AI.
10. Verify without polluting. When you screenshot/QA a live site, set your User-Agent to the opt-out marker lazysite-agent/<your-partner-id> (e.g. --user-agent "lazysite-agent/claude-dhcf"). The visitor-stats classifier treats that token - and the legacy claude-code-agent - as tooling, not a human visitor, so your QA hits stay out of the audience numbers.

Decision guide: page, component, layout, or theme?

• Just words / a one-off article -> a Markdown page.
• A block that repeats with the same shape (cards, rows, listings) -> a data file + loop/component.
• Chrome or a feature many pages share (nav, hero, footer, pager) -> the layout.
• A distinctive visual identity (palette, type, spacing) -> a theme (tokens), on top of a layout.
• "It's basically a whole custom design" -> still a theme + layout, never a monolithic page. If it feels too custom for that, the layout is the right home for the structure and the theme for the look.
• A single self-contained interactive widget / embed fragment -> the only case for raw mode (api: true / raw: true, with its own content_type) - never an ordinary content page.

Refactoring an inherited monolith

When you meet a page like summit-of-summits:

1. Separate the three strands. Pull the prose out into Markdown; note the repeated structural blocks (candidates for components/data); collect the visual decisions (colours, fonts, spacing).
2. Tokenise the look. Turn colours/fonts/spacing into theme.json tokens; move shared rules into the theme's main.css. Replace every hardcoded hex with a var(--theme-*).
3. Lift structure into the layout. Turn the hero and section scaffolding into layout markup / a component, driven by front-matter where it varies per page.
4. Re-home the content. The page becomes a short .md that selects the layout+theme and carries the words; register it for sitemap/llms/feeds.
5. Verify and cache-clear, then delete the raw asset page.

Done once, the design is now reusable, themeable and authorable - and the next page in that style costs almost nothing.

Publishing hygiene (hard-won)

• Keep json: data ASCII - the data loader drops non-ASCII (use - not em dashes, straight quotes, -> not arrows).
• Page <style> must be a single line - a leading # on its own line is parsed as a Markdown heading and mangles the CSS.
• Markdown is not processed inside raw block HTML - use <strong> inside a <div>, not **bold**.
• Clear the cache by re-activating the active layout after edits; a nav save alone does not drop already-cached HTML.
• Edit the active layout by copy-then-activate (stage a new layout dir, activate it, retire the old) - the live one is write-locked.
• One editor per file. If a page is open in the manager it is locked; don't fight the lock - edit the underlying data file instead.
• Screenshots use the lazysite-agent/<partner-id> UA (see rule 10) so QA traffic stays out of the human analytics.

Checklist before you call a site "done"

• No hand-authored HTML documents under /lazysite-assets/ or other static paths.
• No content page uses raw mode (api: true / raw: true) - only true self-contained artifacts do.
• Page bodies are Markdown; structural HTML lives in the layout.
• No hardcoded colours/fonts in pages - all var(--theme-*).
• Repeated blocks come from data, not copy-pasted markup.
• Shared features (hero, nav, footer) are layout features, not per-page hacks.
• The whole site restyles by swapping the theme.
• Pages are registered (sitemap/llms/feeds) and reachable from the nav.
• Verified live (with a non-human UA) and cache cleared.