Good docs do not talk about code examples in the abstract. They show real commands, realistic payloads, and enough context that a reader can copy, modify, and move on.
What Good Code Examples Look Like
Bulkhead OS supports plain code blocks for published content and Shiki-backed preview rendering for the editing workflow. The pattern is simple: keep a short explanation above the snippet, label the language clearly, and make the example copyable.
Shiki Preview Example
The editor preview endpoint accepts Markdown and a preferred theme. The response returns rendered HTML ready for a live preview pane.
curl -X POST https://docs.bulkheados.com/api/v1/editor-preview -H "Authorization: Bearer ENGINE_ADMIN_TOKEN" -H "Content-Type: application/json" -d '{
"markdown": "# Shipping docs faster\n\n~~~ts\nconst published = true\n~~~",
"theme": "dark"
}'
{
"html": "<h1 id=\"shipping-docs-faster\">Shipping docs faster</h1>\n<pre class=\"shiki github-dark\">...</pre>"
}
Shell Example
Shell examples should be safe to paste as-is. Show flags, URLs, and expected output in separate blocks when needed.
pnpm --filter @pobal-os/engine validate:routes
pnpm --filter @pobal-os/engine build
npx wrangler deploy
HTTP Example
For APIs, pair the request with the response. That makes the docs useful even before a reader opens their own tooling.
GET /api/v1/search?q=bulkhead&limit=3&type=inform&visibility=external
Accept: application/json
{
"results": [
{
"title": "Introduction",
"url": "/introduction",
"excerpt": "BulkheadOS is a company operating graph for complex software companies...",
"score": 0.82,
"node_type": "inform",
"domain": "Bulkhead Documentation"
}
],
"meta": {
"query": "bulkhead",
"count": 1,
"limit": 3
}
}
Markdown Example
When documenting authoring, include the exact Markdown a content editor would type.
## Ship the smallest helpful example
Use a real request, a realistic response, and one short note that explains why the example matters.
> If you can describe an endpoint, you can show it.
Rule of thumb: If a reader would need to invent sample data before they can try the feature, the example is still too abstract.