Install in a monorepo

Monorepos work to the extent shadcn gives it for free, which is a real level of support and worth stating precisely rather than generously.

What works

shadcn already handles workspace layouts. It routes base components to your UI package and blocks to the app you ran the command in, and it rewrites imports accordingly. Two requirements: each workspace has its own components.json, and the workspaces agree on style, iconLibrary, and baseColor.

AgentBlog asks for card, badge, separator, avatar, and button by bare name, so they resolve through your configuration and inherit all of that with no work from you.

Where the files land

In a workspace, which root those last three land in is the genuine unknown, and getting it wrong scatters the agent layer somewhere nobody looks. It fails quietly, because the files do get written.

So agentblog doctor detects a workspace, reports where each of the three landed, and never blocks:

AgentBlog: pnpm workspace detected.
  agentblog.config.ts   apps/web/agentblog.config.ts
  AGENTS.md             AGENTS.md          (repo root)
  .claude/skills/       .claude/skills/    (repo root)
Verify these are where your tooling expects them.

If the split is wrong for your setup, move the files and update the single import in lib/config.ts. That is the only place @/agentblog.config is read, which is exactly why there is only one importer.

Turborepo

Nothing special is required. Add the blog to whichever app owns it and the usual build, lint, and typecheck tasks pick it up.

One setting is worth adding: give the content directory an input on the build task, so editing a post invalidates the cache.

{
  "tasks": {
    "build": {
      "inputs": ["$TURBO_DEFAULT$", "content/**", "agentblog.config.ts"]
    }
  }
}

Without it, Turborepo can serve a cached build that predates your new post. The symptom is a post that exists in the repository and not on the site.

What has not been tested

A monorepo fixture is not in CI yet. It joins when the first database-backed content source ships, because that is when workspace users become likely.

Until then the mechanism is sound and the disclosure above is accurate. If something lands in the wrong place, please open an issue with your workspace layout. That is more useful than a passing test written to fit the assumption.