Changelog

Authors on non-legacy content management sites now see a grouped, clearly labelled draft actions menu with updated copy throughout — no more “branch” language, and actions are organized into Draft Details, Actions, and Draft Management sections.

Desktop and mobile layouts differ: mobile includes Request review and Publish changes inline; desktop keeps those in the edit bar. An Exit button for proposed drafts and a responsive editing bar with continuous label truncation are also included.

The API can return a grouped Added / Modified / Removed summary of a draft’s changes — separating documentation changes from settings changes — enabling the review modal to show reviewers exactly what a draft touches.

Display names resolve from post-change then pre-change data, so renamed and deleted entities still show their correct labels. An “other changes” bucket surfaces any changes that cannot be attributed, rather than silently omitting them.

Before requesting review, the API can now identify other proposed drafts on the same site that touch the same pages, patterns, or settings areas, giving authors a heads-up about potential conflicts.

The check is advisory and never blocks a review request; a mayHaveUndetectedOverlap flag is returned when either draft has changes that cannot be classified.

AI agents querying design tokens now receive accurate, namespace-qualified paths and the exact CSS custom-property names emitted by the site’s token build, reducing hallucinated variable names.

Calling list-tokens with no arguments returns a digest of token counts by namespace; passing an include prefix filter returns flat token entries with cssVar, path, and value. The get-component tool gains section selection via include.

The “Contact Knapsack” modal and CSAT feedback form, which submitted to a retired Zendesk integration, have been removed from the app.

Settings, MCP, Design, Develop, Prototype, and related nav links no longer appear — and their routes are now denied — when viewing a deployed App Client, preventing unintended access to administrative surfaces from read-only environments.

The authoring workspace (latest, branch instances, local dev) is completely unaffected; only standalone deployed clients are gated.

Help and documentation links throughout the app now point to the correct pages on the new docs platform, resolving broken links that previously led to 404 errors.

Heading elements inside markdown blocks no longer include id attributes generated by the markdown renderer, eliminating per-render deprecation warnings from the server.

App UI renders markdown client-side with its own heading ID assignment, so this change does not affect the table-of-contents or any user-visible heading anchors in the authoring workspace.

Component prop defaults, descriptions, slot titles, allowed component constraints, and authored ordering are now fully emitted into the spec tables that AI agents read over MCP, eliminating incorrect inferences caused by missing data.

Props tables now include Default and Description columns; slots tables include Title, Required, Only One, Constraints, and Allowed Components columns, all in the order you authored them.

Asset-set CSS and JavaScript files no longer 404 or trigger MIME errors when multiple asset sets share source files from the same directory.

A race condition caused the server to register static routes at one URL prefix while serving clients a different prefix; pages could load unstyled as a result. URLs are now derived deterministically from the asset set configuration.

Fixes a race condition that caused asset-set CSS to be served from a mismatched URL, producing a MIME-type error that blocked styles from applying in the browser.

When multiple asset sets shared a source directory, non-deterministic Promise resolution could cause the server’s static routes and the client’s reported URLs to disagree, resulting in a 404 and an HTML response where CSS was expected.

Large-pattern get-pattern calls that previously exceeded timeout thresholds on a cold cache now complete in seconds by sharing a single renderer-discovery request across concurrent callers.

Fixes a bug where newly uploaded component builds were silently ignored by live AI conversations — tool handlers now correctly serve the latest content after a revalidation.

Users who see the concurrent editing banner can now click “Learn more” to open the published documentation on concurrent editing in a new tab.

Text-editor blocks with malformed content now degrade gracefully to empty HTML instead of triggering error-level log floods on every ingest run.

A code guard was added so any typeless TipTap content (from any site or source) silently renders empty rather than throwing, preventing recurrence beyond the specific blocks repaired.

Admins can now find every page or pattern where unsupported formatting would have been silently lost, navigate between affected locations, and recover content without data loss.

A new ‘Unsupported formatting’ section in Doctor lists affected pages with direct links; a global banner lets admins step through them. Admins see salvaged plain text with a warning badge; other visitors see the content preserved invisibly until an admin saves the block.

Authors can now see who last published and when, and browse full publish history with search and pagination, directly from the editing bar.

Save, publish, branch creation, and review requests are now refused server-side when a site is frozen for migration, preventing content loss during the cutover window.

Asset images that failed in every browser due to a TLS certificate mismatch on the legacy host are now served from the correct CDN and display as expected.

When a teammate saves changes to the page or pattern you’re viewing, a footer banner appears so you know concurrent edits are happening — no more silent overwrites.