For the full list of changes, check out the Webiny 6.4.12external link release on GitHub.

Website Builder
anchor

Fixed MissingversionandmetadataFields in Single-Page Queries (#5727external link)
anchor

The GET_PAGE_BY_ID and GET_PAGE_BY_PATH queries in the Website Builder SDK were missing the version and metadata fields from their selection sets. If you needed these fields, you had to patch the SDK manually. Both fields are now included by default.

Fixed Stale Page Revision Data After Mutations (#5757external link)
anchor

Creating, publishing, unpublishing, or deleting page revisions did not update the revisions cache, so the editor’s revisions drawer and top bar dropdown displayed stale data until a page reload. Both now reflect these changes immediately, and publishing a revision marks the previously published one as unpublished.

Added MissingdeletePageRevisionGraphQL Mutation (#5757external link)
anchor

Deleting a page revision from the editor did not work, because the Admin app called a websiteBuilder.deletePageRevision mutation that the GraphQL API did not have. The mutation now exists. Deleting the latest revision promotes the previous one, and deleting the last remaining revision deletes the page.

Improved Revision List Actions and Confirmations (#5757external link)
anchor

Deleting, publishing and unpublishing a revision from the revisions drawer now ask for confirmation and show a toast with the outcome. Previously these actions ran on the first click and a failure was not reported. Delete revision is now also offered on locked and published revisions, matching the API.

Fixed Workflow Controls in the Page Editor (#5757external link)
anchor

With workflows enabled, the workflow controls were not applied to the page editor top bar. They now are, and once a review is approved the editor becomes writable again and offers Publish instead of New Revision. The revisions drawer no longer offers Publish revision when workflows are enabled, because it let a page under review be published past its workflow - publish from the top bar instead.

Fixed Duplicate Requests When Loading the Pages List
anchor

Opening the pages list sent the same listPages request several times. It is now sent once.

Permission Checks for Scheduled and Folder Publishing (#6173external link,#6240external link)
anchor

Scheduling a page to be published or unpublished now checks the permissions of the caller. When using folder level permissions, publishing and unpublishing a page requires write access to its folder.

Development
anchor

Debug Logger Tool (#5734external link)
anchor

A new opt-in, permission-gated debugger is now available for troubleshooting issues in environments where you cannot access AWS or the underlying code. When a user with the dev-tools.debug permission has capture on, debug payloads from instrumented call sites are returned in the GraphQL response under extensions.debug. The permission must be granted explicitly - full access does not include it.

Key details:

  • Capture is requested per-request via the x-webiny-debug header (set to 1, or a glob pattern like cms.os.* with ! to exclude).
  • Debug sessions are buffered in memory (capped at 4 MB), and payloads never reach CloudWatch—only the permitted user sees them.
  • The first instrumented call site is cms.os.list, which captures OpenSearch query and response data around entry listing.

In Admin, a Debug mode indicator in the header shows that capture is on and lets you download a JSON report or clear the collected data. The Debugger page, reachable at /debugger, has the capture toggle and the namespace filter. Collected sessions are held in memory only and do not persist beyond the browser tab.

After deploying this release, you must redeploy your API infrastructure (yarn webiny deploy api) and invalidate your CloudFront distribution to ensure the new x-webiny-debug header is forwarded correctly. Without this, the feature silently captures nothing.

Fixed Peer Dependency Ranges Being Locked During Package Publishing (#5777external link)
anchor

When publishing packages, peer dependencies were incorrectly locked to the exact version installed in the monorepo rather than preserving the original range. This caused installation failures with @webiny/website-builder-nextjs - npm would error with ERESOLVE if your project used a different Next.js version than the one pinned during publishing.

Peer dependencies are now published with their original version ranges. For example, @webiny/website-builder-nextjs now correctly declares next: >=15.0.0 <16.0.0, allowing installation alongside any compatible Next.js 15 version.

Updated Core Dependencies (#6168external link)
anchor

This release upgrades several core dependencies to their latest versions. The upgrade includes internal type fixes required by the new dependency versions — no changes to your project code are required.

Updated Dependencies (#6253external link)
anchor

Routine minor and patch dependency updates, including @aws-sdk/*, @pulumi/pulumi, @opensearch-project/opensearch, and various build/test tooling.

Admin
anchor

Fixed Keyboard Shortcut Leak on Component Unmount (#5757external link)
anchor

useHotkeys did not release its shortcuts when a component unmounted while they were registered. For example, a drawer that unmounted while open kept its esc binding, and the next component to register that shortcut threw an error.

FixedgetUserQuery When MissingwhereArgument (#6240external link)
anchor

The getUser query no longer fails with an unexpected error when the where argument is omitted.

Scheduler Validates Schedule Date Before Creating Actions (#6240external link)
anchor

Scheduling an action without a schedule date now returns the invalid date error instead of failing with an unexpected error.

Custom Scheduler Namespaces Require Permission Implementation (#6240external link)
anchor

If you have a custom scheduler namespace, you must now register a SchedulerPermissions implementation. Without one, getting, listing and cancelling scheduled actions returns a not authorized error. Additionally, the namespace_startsWith filter has been removed from ListScheduledActionsUseCase — use namespace instead.

API Keys and Roles Now Require Matching Permissions (#6268external link)
anchor

When creating or updating API keys and roles, the requested permissions are now validated against the permissions of the current identity. You can only grant permissions that you yourself hold (wildcards are taken into account). Attempting to assign permissions beyond your own scope returns a not authorized error. Existing permissions already stored on the API key or role can be kept unchanged.

Stricter Identity and Request Checks (#6173external link,#6203external link,#6240external link)
anchor

  • Cookie authentication: the authentication cookie is accepted only for the requests that need it, such as downloading private files.
  • CORS: stricter CORS headers on API responses.
  • Requests: GraphQL requests must be sent with the application/json content type, and JSON parsing is stricter.
  • Current user: getCurrentUser checks the identity of the caller.
  • Websockets: stricter permission checks.
  • Folders: creating a folder validates the folder type first.

Headless CMS
anchor

Fixed Full-Text Search Failing on Non-Text Fields in OpenSearch (#5819external link)
anchor

Full-text search queries in OpenSearch would fail when searchable JSON fields contained non-text values like dates, numbers, or booleans. The search now uses lenient mode, allowing these mixed-type fields to be searched without errors. Additionally, full-text search paths now correctly include parent object fields, and the contains filter in DynamoDB properly searches objects inside arrays.

If you have a custom CmsEntryOpenSearchFullTextSearch implementation, ensure you add lenient: true to your search configuration.

Added Model Access Guard for Content Models (#6240external link)
anchor

A new ModelAccessGuard provides an additional access rule layer for content models on top of the existing Headless CMS permissions. This allows for more granular control over which identities can access specific models.

Stricter Access Checks for Entries and Models (#6173external link)
anchor

  • Own entries: getting entries by ID and listing revisions respect the “own entries only” permission.
  • Private models: private content models are no longer reachable through getContentModel or the generic entry queries.
  • Scheduled publishing and bulk actions: both check the permissions of the caller.

Bulk Delete Now Validates Delete Permission per Entry (#6240external link)
anchor

Bulk delete now checks the delete permission for each entry. Entries the caller is not allowed to delete are skipped.

Entry Access Checks Use Stored Entry Data (#6268external link)
anchor

Updating an entry and creating a revision from an entry now check access against the stored entry rather than the submitted data. This closes a potential privilege escalation vector where manipulated request payloads could bypass access controls.

Entry Owner Cannot Be Changed via Updates (#6268external link)
anchor

The createdBy value sent when updating an entry or creating a revision from an entry is now ignored. This prevents unauthorized ownership transfers. Setting createdBy when initially creating an entry still works as expected.

Folder Level Permissions Apply to Publishing Operations (#6240external link)
anchor

When using folder level permissions, publishing, republishing and unpublishing an entry now requires write access to its folder. Previously, these operations only checked top-level permissions.

Tenant and Language Entries Require Matching Permissions (#6240external link)
anchor

Some resources available through the Headless CMS API now require their own permissions:

  • Tenant entries: A role with Headless CMS permissions but without the Tenant Manager permission can no longer read or change tenant entries through the CMS API.
  • Languages: A role with Headless CMS permissions but without the Languages permission (with write access) can still read languages, but can no longer create, update or delete them through the CMS API.

Record Locking Permission Checks and Expired Locks (#6203external link,#6268external link)
anchor

Record locking operations now require an admin user - anonymous callers and API keys are rejected - and listing all lock records requires full access or the record locking permission. An expired lock is now taken over by deleting the lock record and creating it again, instead of updating it.

File Manager
anchor

Tenant-Scoped Storage and Stricter Permission Checks (#6173external link,#6203external link)
anchor

  • Uploads: files, including multi-part uploads, are stored under tenant-scoped storage keys, and the upload input is validated.
  • Private files: downloading a private file requires an authenticated identity with the permission to read files.
  • Settings: updating the File Manager settings requires the settings permission, and the settings input is validated more strictly.
  • Performance: the S3 client is cached and reused.

Webhooks
anchor

Per-Webhook Signing Secrets (#6193external link)
anchor

Webhook signing secrets can now be configured individually for each webhook, rather than relying on a single global secret. You can set, change, or clear the signing secret via GraphQL, the SDK, or the Admin webhook form, and a webhook without its own secret keeps using the global one. The Admin interface also includes new actions to generate a random secret and copy it to your clipboard.

Webhook Settings Require the Webhook Permission (#6193external link)
anchor

The getSettings query returned the webhook settings, including the signing secret, without a permission check. It now requires the permission to manage webhooks, the same as the other webhook operations.

Stricter Webhook Endpoint Validation (#6203external link,#6240external link)
anchor

Webhook endpoint URLs must use HTTPS and can no longer point to a local or link-local address. Localhost is accepted only during local development. Deliveries to an existing webhook with such an endpoint now fail.

Webhook Redirects Are No Longer Followed (#6240external link)
anchor

When delivering a webhook, redirect responses (3xx) are no longer followed. The redirect response is recorded as the delivery result instead. This change hardens webhook security by preventing potential SSRF attacks through redirect chains.

Tenant Manager
anchor

Tenant Management Permission Checks (#6203external link)
anchor

Creating, installing, enabling, disabling and deleting tenants now checks the Tenant Manager permission of the caller. Enabling, disabling and deleting is limited to direct child tenants of the current tenant.

Audit Logs
anchor

Audit Logs Require the Audit Logs Permission (#6240external link)
anchor

The listAuditLogs query returned audit logs without a permission check. It now requires the Audit Logs permission. The getAuditLog query checks the permission before it loads the record, so it no longer reveals whether an audit log exists.

AI Power-Ups
anchor

Settings No Longer Expose Provider API Keys
anchor

The getSettings query rejects anonymous callers and never returns provider API keys. Only identities with full access see the masked keys. Updating the settings requires full access.

Workflows
anchor

Workflow Queries Require an Authenticated Identity (#6203external link)
anchor

The listWorkflows, listWorkflowStates, getWorkflowState and getTargetWorkflowState queries, and the createWorkflowState mutation, now reject anonymous callers.

Mailer
anchor

Mailer Settings Require the Mailer Permission (#6203external link)
anchor

The getSettings query now requires the mailer.settings permission, the same as saving the settings.

Infrastructure
anchor

Fixed GuardDuty Malware Protection Deployment and Destroy Failures (#6186external link)
anchor

The GuardDuty malware protection plan (used for File Manager threat detection) had missing dependency declarations. This caused two issues:

  • During deployment: the plan could be created before its IAM policy was attached, leaving it in ERROR status.
  • During destroy: the policy attachment could be removed before the plan, preventing GuardDuty from deleting its EventBridge rule and causing webiny destroy to fail with a BadRequestException.

The plan now correctly depends on its policy attachment, and the event-forwarding target depends on its role policy.

If your stack is already in a stuck state, a manual cleanup is still needed before webiny destroy will succeed.