Docs that answer the question you opened the schema for
Responses now document their headers and body, endpoint descriptions render as markdown, and global search finds endpoints by URL fragment.
In this release
- New
Responses show their headers and body
Every response documents the headers it returns and the shape of its body, with examples resolved, so you can read an endpoint without chasing $refs.
- New
Endpoint descriptions render as markdown
Lists, links and code blocks in a description come out formatted instead of as one long line of raw text.
- Improved
Global search finds endpoints by path
Type a fragment of a URL — `pet/{id}` or just `checkout` — and fuzzy matching brings up the endpoints, not only exact node names.
- Fixed
Deleted endpoints stop reporting lint errors
Removing a path no longer leaves it behind in the linter, and linting runs against the document we publish rather than the working copy.
The Docs tab exists so you can read an API instead of parsing it. This release closes the three gaps that still sent people back to the raw document.
Responses, in full
A response is now documented the way a consumer needs it: the headers it sets, the shape of its body, and the examples the document carries — with $refs resolved in place. Reading what 201 actually returns no longer means opening two more nodes and holding them in your head.
Descriptions as the markdown they are
The OpenAPI spec says descriptions are CommonMark, and authors write them that way — a bullet list of caveats, a fenced snippet, a link to the guide. Printed as plain text they read as literal ** and - in the one view whose whole job is to be readable. Endpoint descriptions now render, using a strict subset of markdown that escapes the document's own content before emitting a single tag.
Finding an endpoint by its URL
Global search matched node names, which is only useful if you remember them. It now matches paths fuzzily, so pet/{id} or a bare checkout brings up the endpoints you meant. Names still match too — this is an addition, not a replacement.
And one that had been quietly wrong
Deleting a path left a tombstone the linter kept reading, so a removed endpoint went on reporting errors for a path that was no longer there. Linting now runs against the document we publish, which is the one you would ship.