How the agent works
An agent-native engine that shows its work.
Before every job — building a form, setting up a blog, wiring SEO — your agent reads a small, tested skill. Plain markdown. No proprietary magic.
The problem with most AI builders
The AI is a black box.
You ask. You get a result. You can't see why. When the output looks generic, when something breaks subtly, when the same prompt does different things on different days — there's nothing to audit. The AI just "knows best." Trust the magic.
That's how Lovable, Bolt, v0, and the AI features in Wix and Squarespace all work. The model has its own opinions and you live with them.
SiteShip's approach
Tested playbooks, read on every job.
SiteShip is built around a small library of skills — markdown documents that describe exactly how the engine handles forms, blogs, SEO, payments, embeds, and the rest. The agent reads the matching skill before it touches your site, every time.
That's why the first try usually works. The agent isn't guessing — it's following a tested pattern that the engineering team curates and updates. New gotcha discovered? Skill updated. Every site benefits, immediately.
And it doesn't stop at the skill. Before any change ships, the engine runs diagnostics across the whole site — broken component references, missing alt text, dropped form actions, placeholder copy you forgot to swap. Anything off-contract gets fed back to the agent, which fixes it and tries again.
Write. Check. Correct. Repeat — until the change is actually right. Most AI tools generate and pray. SiteShip generates and verifies.
What your agent reads
The skills, in full.
Site Architecture Reference
File-based website using LiquidJS templating and Tailwind CSS (compiled by the engine on your machine and served at `/tailwind.css`).
Blog
The engine auto-populates all blog data — you just create the files with the right structure. No configuration needed beyond the file layout.
Data Pages
Store structured content in `data/*.json` files. Every file is available in every template as `{{ data.filename }}` — no configuration needed.
Third-Party Embeds
Adding an external service requires two things: declaring the domain in your site config so the security policy allows it, and writing the embed HTML. The engine handles the rest.
Forms
Forms post to `/api/forms/{id}`. The hosted runtime handles validation, honeypot spam protection, and stores submissions in a per-site Durable Object with built-in SQLite — queryable from your dashboa…
Local SEO
The engine auto-generates all schema.org JSON-LD from structured data in `data/` and `site.yaml`. You set the data — the engine handles the markup.
SEO
The engine auto-generates JSON-LD schema, OG tags, sitemap.xml, RSS, and canonical URLs from your frontmatter and `data/site.yaml`. Your job is to set the right data — the engine handles the markup.
Shopify E-commerce
Add Shopify products and checkout to any page using Storefront Web Components. Shopify handles inventory, payments, and checkout — the site just displays products.
Stripe Payments
Add payment buttons to any page using Stripe's embeddable web components. Stripe handles the entire checkout — the site just embeds a button.
You don't need to read these. They're for the agent. We document them anyway — we'd rather show our work than ask you to trust a black box.
Why this matters
The opposite of a black box.
Auditability.
Read the skill. See what the agent is supposed to do. If your form, blog, or SEO ever ships wrong, you have somewhere to point.
No surprises.
Same skill, same site, same result. The agent isn't free-styling. It's following a documented contract — predictable, repeatable, debuggable.
No lock-in.
The engine contract is documented. Export your site as HTML anytime. Take the workspace with you. Plain files, no proprietary glue.