This is the realistic first run: a single marketing site, Tier A (marketing only). Leave payments and auth for round two so you can shake out the workflow without wiring up Cloudflare, Stripe, and Resend at the same time.
The build itself is conversational — you describe what you want and your agent runs the commands. The steps below show the whole loop so you can follow along and steer.

Before you start

You’ve installed the binary and confirmed it runs:
sitekit --version

1. Install the companion skill

This makes any agent session you start know how to drive sitekit:
sitekit init-skill --target claude-code
Use --target opencode for OpenCode, or --target path:<absolute-path> for any other harness.

2. Create the project

Pick something close to a real, shippable product — the closer to real, the more you learn from the loop.
sitekit init <product-name> \
  --domain <product>.app \
  --description "<one-sentence pitch>" \
  --palette dark-stage
This writes sitekit.workspace.json and scaffolds content/, playbooks/, pages/, and public/images/. No HTML yet — pages are emitted on demand. cd into the new directory.
The --palette value must be one the chosen style pack declares. If it doesn’t apply, sitekit falls back to the pack’s first palette. You can also set the voice, style, and shape up front with --shape, --voice, and --style — that saves a default playbook so later sitekit run calls need no flags.

3. Hand the build to your agent

Open a fresh agent session in the project directory. The skill activates automatically. Say something like:
Build the marketing site. The product description is in sitekit.workspace.json. Read SITEKIT.md first, then show me the playbook composition before you start.
Your agent will:
1

Author the content

Populate content/messaging.yaml (the brand story) and content/pages/<page>.yaml (per-page copy) — sitekit keeps copy separate from markup.
2

Emit pages

Run sitekit run for each page. Every page lands at pages/<slug>.html with an inline playbook marker recording the voice + style + shape it was built against.
3

Author the HTML

Fill in each page’s <header>, <main>, and <footer> against the inlined style, voice, and shape guidance — not from a generic template.
4

Run verify and fix

Run sitekit verify, read each fix_hint, and iterate until it’s clean.
You watch and steer — when the agent drifts off-brand, say so, and it corrects.

4. Verify it yourself between rounds

You don’t have to take the agent’s word for it. Run verify directly:
sitekit verify            # full pipeline; exit 0 when no error-severity checks fail
sitekit verify --json     # machine-readable, to pipe into another tool
sitekit verify --strict   # warnings also block (the deploy gate)
The verify loop is not optional. Run it after every meaningful change — failures don’t get easier to fix later.

5. Deploy

sitekit doctor                   # what infrastructure setup is still missing
sitekit verify --strict          # final pre-flight
sitekit deploy --env staging     # smoke-test on a preview branch
sitekit deploy --env prod        # when staging looks right
sitekit deploy runs the verify gate by default and ships the Worker + Pages to Cloudflare. You’ll need a Cloudflare account, wrangler login once, and the Pages project name set with sitekit configure deploy.pages_project_name <name>.

Where to go next

Building is the first of four loops. Once the site exists, don’t re-run sitekit run on an already-authored page — it overwrites the body. Reach for the right loop instead:

The four workflows

Maintain, experiment, evolve, and adopt existing sites.

How it works

The mental model behind the verify loop and the playbook marker.

Playbooks

Voice + style + shape + valence — the composition that drives every build.

Command reference

Every command group, by task.