Start with the Operator Manual
How authorization, search, Markdown access, and repository sources work in the Operator Manual.
Task-first navigation, scoped access
The manual is organized around operator tasks. Audience metadata controls which of those tasks a reader may discover; it does not divide the source of truth into separate conceptual shelves.
Four audiences, one identity
The manual uses the same Supabase identity and authorization records as Local Media Desk. It does not maintain a parallel list of documentation users. Every page declares one audience, and the server decides on every request.
| Audience | Who can read it |
|---|---|
| Public | Anyone, signed in or not |
| Client | A user with at least one active workspace membership |
| Developer | A person holding the platform.docs.read grant, a platform operator who can view all tenants, or an active workspace administrator or owner |
| Admin | A platform operator who can manage tenants or platform permissions |
Workspace administrators and owners automatically have documentation-reader access. Sign in with your usual workspace identity; no separate reader grant is needed. This follows your current admin access and ends when that access is removed.
platform.docs.read is the grant for an outside documentation collaborator. It
opens Developer pages only: it gives no view of any tenant or client data, no
Admin pages, and no Client pages, which still need an active workspace
membership. The grant is issued and revoked like any other platform grant; see
the Docs collaborator access runbook on the Engineering shelf.
Some client pages narrow further to a specific tenant, account, or tenant permission. A page with those constraints is included only when your current membership satisfies all of them.
Search follows access
Navigation and search are assembled from the same access decision as the page route. If a page is outside your scope, its title, path, breadcrumbs, and text are absent from the sidebar and from search results — it is not merely unclickable.
Reading a page outside the browser
Copy as Markdown
Every page carries a copy button that fetches that page's Markdown using your own session.
/llms.txt
An index of the pages you can read, for an assistant that follows links.
/llms-full.txt
The full text of those same pages in one response.
All three are scoped exactly like the site: they contain the pages your request is allowed to read and nothing else. The "open in an external assistant" action is offered only on public pages, because an outside service fetching a URL carries no session of yours and would receive a sign-in page instead.
Pages that come from a committed file also show a last-updated date taken from the file's git history.
Engineering documentation is mirrored, not retyped
The Engineering shelf renders the repository's own Markdown — architecture
notes, operational runbooks, and architecture decision records — directly from
docs/ in the source tree. Links between those files resolve inside the
manual; links to anything else in the repository open on GitHub. The shelf is
developer scope, so it is absent for everyone else.
When access changes
Reload the page after an administrator changes your membership or platform grant. The manual reads current database grants and does not keep a persistent authorization cache.
Last updated on