- User API key create/list/revoke on StackServerApp and AsyncStackServerApp
- Team API key create/list/revoke on both facades
- check_api_key validates any key and returns user_id/team_id
- Tests for all sync and async API key methods
- list_contact_channels, create_contact_channel for user channels
- send_verification_code with optional callback URL
- verify_contact_channel to validate codes
- Both sync and async facades
- Tests for list, create, send verification, verify on sync facade
- Tests for all contact channel methods on async facade
- Covers optional fields and callback URL parameter
- grant_permission, revoke_permission for team and project scope
- list_permissions with optional direct filter
- has_permission boolean check, get_permission single lookup
- Both sync and async facades
- Tests for grant, revoke, list, has, get permission on sync facade
- Tests for all permission methods on async facade
- Covers team-scoped and project-level permissions
- Add add_team_member and remove_team_member via /team-memberships
- Add send_team_invitation, list_team_invitations, revoke_team_invitation
- Add list_team_member_profiles and get_team_member_profile
- Add user_id field to TeamMemberProfile for profile filtering
- Both sync and async facades have identical method surfaces
- Tests for add_team_member, remove_team_member
- Tests for send/list/revoke team invitations
- Tests for list_team_member_profiles and get_team_member_profile
- Async equivalents for all membership/invitation/profile operations
- Tests for get_team, list_teams, create_team, update_team, delete_team
- Tests for get_team_by_api_key two-step lookup
- Async equivalents for all team CRUD operations
- Uses respx mocking pattern consistent with existing user tests
- list_sessions(user_id) returns list of ActiveSession for a user
- get_session(session_id, user_id) filters from list_sessions, returns ActiveSession or None
- revoke_session(session_id, user_id) deletes session via DELETE endpoint
- Both sync and async facades implement all three methods
- StackServerApp (sync) and AsyncStackServerApp (async) facade classes
- get_user returns ServerUser or None on 404
- list_users with cursor/limit/order_by/desc/query/include_restricted/include_anonymous
- create_user sends only non-None fields
- update_user uses _UNSET sentinel to distinguish not-provided from explicit None
- delete_user returns None
- get_user_by_api_key performs two-step lookup via /api-keys/check then /users/{id}
- Both classes support context managers and compose API client + optional token store
- Tests for get_user, list_users, create_user, update_user, delete_user
- Tests for get_user_by_api_key two-step lookup
- Tests for sentinel pattern in update_user (None vs not-provided)
- Both sync and async facades covered with respx mocks
- TokenStore ABC with get_stored_access_token, get_stored_refresh_token, compare_and_set
- MemoryTokenStore, ExplicitTokenStore, RequestTokenStore implementations
- Module-level registry shares MemoryTokenStore instances per project_id
- resolve_token_store handles memory/dict/request/None init variants
- CAS refresh algorithm (sync + async) with 20s expiry buffer and 75s iat threshold
- Token refresh via form-encoded POST to /api/v1/auth/oauth/token
- Per-instance threading.Lock and asyncio.Lock for concurrency safety
- 34 passing tests covering all store types, registry, helpers, and CAS branches
- Add AuthState, TokenPartialUser, decode_access_token_claims to public API
- Add sync_authenticate_request and async_authenticate_request exports
- Existing exports unchanged
- TokenPartialUser frozen dataclass for unverified JWT payload extraction
- AuthState frozen dataclass for authentication results
- decode_access_token_claims extracts user info without signature verification
- _extract_token_from_headers handles Authorization and x-stack-auth headers
- sync_authenticate_request and async_authenticate_request compose with JWKS verifier
- PaginatedResult[T] generic wrapping nested {items, pagination: {next_cursor}} shape
- Convenience properties .next_cursor and .has_next_page
- Package __init__.py exports all models, errors, and PaginatedResult
- SyncAPIClient/AsyncAPIClient intentionally excluded from public API
- Tests verify pagination with ServerUser and Team types
- BaseAPIClient generic base with header construction, URL building,
response parsing via x-stack-actual-status and x-stack-known-error
- SyncAPIClient wrapping httpx.Client with context manager support
- AsyncAPIClient wrapping httpx.AsyncClient with async context manager
- Exponential backoff retry for idempotent methods (GET/PUT/DELETE)
- 429 rate limit handling with Retry-After header support
- POST/PUT/PATCH with no body sends {} as JSON
- All 26 tests passing
- BaseUser and ServerUser with camelCase aliases and millis-to-datetime properties
- Team, ServerTeam, TeamMemberProfile, TeamInvitation with timestamp conversions
- ActiveSession with GeoInfo nested model
- ContactChannel and TeamPermission/ProjectPermission
- Tests for user and team model parsing, inheritance, and field conversion
- StackAuthError base with code/message/details and from_response() factory
- 13 category subclasses: Authentication, NotFound, Validation, PermissionDenied,
Conflict, OAuth, Passkey, ApiKey, Payment, Email, RateLimit, Cli, Analytics
- Complete _ERROR_CODE_MAP with 143 error codes from known-errors.tsx
- StackAuthModel base class with Pydantic v2 ConfigDict and millis conversion
- Package exports via __init__.py with __all__
- Test StackAuthError base class storage and string representation
- Test all 13 category subclasses inherit from StackAuthError
- Test from_response() dispatch for known and unknown error codes
- Test StackAuthModel config and millis conversion
- Package skeleton with pyproject.toml (stubs only, tests expected to fail)
## Summary
- replace the dashboard feedback form's Web3Forms submission with an
authenticated internal backend endpoint
- send support and feature-request notifications through Stack Auth's
native internal email pipeline
- share internal project auth headers in the dashboard and add backend
E2E coverage for support feedback
## Testing
- pnpm typecheck
- pnpm lint -- "src/components/feedback-form.tsx"
"src/components/stack-companion/feature-request-board.tsx"
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Internal feedback submission endpoint with automated internal email
notifications
* New internal email builder and sending utility; recipient list
configurable via env
* **Enhancements**
* Feedback form requires sign-in, disables submit when unauthenticated,
and tightens validation
* Centralized header helper for authenticated internal requests
* Feature request board gates actions for signed-out users and improves
upvote/submit reliability
* Runtime retrieval/validation of the feature-tracking API key and
streamlined user handling
* **Tests**
* End-to-end tests covering internal feedback flows, validation, and
email delivery
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
## Documentation
* Updated setup instructions across all documentation to clarify that
the publishable client key is only required when your project
configuration enforces it, removing confusion about unconditional
requirements.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
- Introduced a fallback mechanism for the private sign-up risk engine,
allowing for zero-score assessments when the primary engine is
unavailable.
- Updated Next.js configuration to support dynamic resolution of the
private risk engine, including aliasing for both Turbopack and Webpack.
- Added a new fallback implementation in
`private-sign-up-risk-engine-fallback.ts` to ensure consistent behavior
during builds.
- Adjusted `risk-scores.tsx` to utilize the new compiled engine,
improving error handling and logging for risk assessment failures.
This update improves the robustness of the sign-up risk scoring system
and enhances the development experience by streamlining engine
resolution.
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Improvements**
* Sign-up risk engine is initialized and validated at startup for more
predictable performance.
* If the risk engine is unavailable or invalid, the system immediately
returns safe zero-risk scores to avoid runtime failures.
* **Tests**
* End-to-end tests updated to match the new engine initialization and
detection behavior.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Konstantin Wohlwend <[email protected]>
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Refined analytics query error classification so certain database
errors are treated as known/handled, reducing exposure of internal
diagnostics in responses.
* **Tests**
* Added end-to-end tests verifying safe (masked) error responses,
preventing leakage of restricted column/identifier details and
constraining suggestion text in error messages.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Enhanced analytics event processing to properly handle edge cases when
data contains certain truncated special characters or emoji sequences,
ensuring data integrity.
* **Tests**
* Added coverage for analytics data edge case handling.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
### Context
We encountered an out of memory error when running verify-data-integrity
against the prod database. This was the error:
`FATAL ERROR: Ineffective mark-compacts near heap limit Allocation
failed - JavaScript heap out of memory`. This was one of the things
preventing verify-data-integrity from running successfully in prod.
### Summary of Changes
Local stress testing with constrained heap and memory telemetry revealed
that the rise in used heap memory was directly proportional to the
number of api calls. Investigation revealed that the `currentOutputData`
array was growing with each api call and was kept in memory. Since it
was still being appended to, it was actively kept in the heap. We
refactor the script to no longer use it, and for the two flags
`--save-output` and `--verify-output` that used it before, we refactor
them to not need to. `--save-output` now streams responses to disk as
JSONL and `--verify-output` now compares each response immediately and
discards it.
We also note a potential source of a future memory leak in the
`allUsers` array that is populated in memory for each project. We
refactor to paginate instead. Note that this didn't cause a memory leak
on local, this is a preventive measure.
### Out of Scope
fetching all transactions in the payments section of the script is
another potential cause for concern, but since the payments section of
the script will be refactored soon, we defer that discussion.
- Updated pnpm-lock.yaml to include 'rolldown' as a new optional
dependency and upgraded 'minimatch' to version 10.2.4.
- Added a debug log statement in risk-scores.tsx to indicate when the
sign-up risk engine is disabled in the public build.
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Signup risk scoring now returns a neutral (zero) score when the
scoring engine is unavailable; scoring errors are logged and surfaced
consistently instead of being silently swallowed.
* Invalid engine shapes now fail loudly rather than falling back
silently.
* **Chores**
* Updated private engine reference and adjusted tests to reflect the new
loading/resolution behavior.
* Expanded Next.js output tracing to include private package files used
by API routes.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Enhances sign-up process with Turnstile integration for fraud
protection. Builds on top of fraud-protection-temp-emails.
Made with [Cursor](https://cursor.com)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Cloudflare Turnstile bot-protection across signup/sign-in flows
(including SDK JSON mode).
* Email deliverability checks via Emailable.
* Sign-up risk scoring with persisted risk metrics and country code
tracking.
* UI: country-code selector, risk-score editing in user details, users
list refresh button, and Turnstile signup demo pages.
* **Bug Fixes**
* Use actual sign-up timestamp for reporting/metrics.
* **Documentation**
* Expanded knowledge base on Turnstile, risk scoring, and env
configuration.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Konstantin Wohlwend <[email protected]>
Co-authored-by: BilalG1 <[email protected]>
Co-authored-by: Armaan Jain <[email protected]>
Co-authored-by: nams1570 <[email protected]>
### Context
We get a lot of noise on sentry about the result admin validation
failing from the onList handler. However, we no longer care about
validating reads for trustedDomains. Via config pushes, people can set
them to anything anyway. There's no value in being stricter on reads
than on writes.
### Summary of Changes
We scope our schema changes to just the onList and read handlers for
trusted domains for the most part, though relaxing the params validation
also affects delete. In practice, delete needs to do an exact match to
find what needs to be deleted so this is fine. Also, without relaxing it
for delete, you wouldn't be able to delete a domain you had previously
put in.
DB migration compat / Back-compat — Current branch migrations with ${{ needs.check-migrations-changed.outputs.base_branch }} branch code (push) Has been cancelled
DB migration compat / Forward-compat — Current branch code with ${{ needs.check-migrations-changed.outputs.base_branch }} branch migrations (push) Has been cancelled
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
adds 2027-track npm package and updates middleware.
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Added asynchronous visit tracking to improve analytics collection.
* **Chores**
* Added a new tracking dependency.
* Refined middleware: switched to a default export with improved typing
and async handling.
* Expanded redirect path mappings and improved header handling for more
reliable navigation.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
### Context
In a [previous PR](https://github.com/stack-auth/stack-auth/pull/1258),
we hardcoded the mode to link because local emulator wasn't in a ready
state yet. Soon after, we started encountering three failing tests on
dev
The three cli tests that were failing on dev were failing because they
required the create mode flag to be set. The hardcode to link made the
create paths unreachable. Since we don't have local emulator working,
allowing users to pass in opts.mode would be bad practice- they'd be
triggering local emulator actions without the local emulator being set
up.
Also, there was a failing AI endpoint test. The unified AI endpoint
tests are set up so that if certain env variables are not present,
certain tests aren't run. In practice, if the openrouter key isn't set,
the tests that require actually processing a full AI endpoint result
without forwarding to prod will be skipped. The failing test was meant
to just check schema validation but it performed a full request instead.
### Summary of Changes
We just skip the tests for now. They'll only become relevant when
"create" is a legitimate workflow, which necessitates the function of
local emulator. There is no regression risk because the flow they're
testing isn't active yet, and so the only thing we could possibly test
is that passing the create mode will invoke a certain function which
isn't helpful at this state.
The unified AI endpoint failing test was reworked, another test
accomplishes the same schema validation effect. We don't lose coverage
by axing the failing test because other AI tests already test valid
request bodies (if they weren't valid, they wouldn't get a response).
DB migration compat / Back-compat — Current branch migrations with ${{ needs.check-migrations-changed.outputs.base_branch }} branch code (push) Has been cancelled
DB migration compat / Forward-compat — Current branch code with ${{ needs.check-migrations-changed.outputs.base_branch }} branch migrations (push) Has been cancelled
This PR implements unified AI endpoint and custom dashboards.
**Unified AI Endpoint**
We now use a single endpoint throughout the codebase that makes the call
to openrouter. Specifically, email drafts, email templates, email
themes, wysiwyg, cmd centre ai search and docs ai, all use this unified
ai endpoint. All the tools are defined in the backend, all the prompts
exist in the backend.
How to review this PR for unified ai endpoint:
This PR will be easier to review if we look at the different folders
that were affected.
under packages - We added streaming functionality, and made renaming
changes
under docs - there are three files that have changed
package.json - we updated the package (we were previously using a very
old version of the package)
route.ts - we changed the call from a direct call to openrouter to the
unified ai endpoint
ai-chat.tsx - because of updating the package, we had to make changes to
adapt to the latest versions of the package
under backend
route.ts - the main unified ai endpoint. this endpoint uses various
support files
forward.ts - this is the forward to production functionality
models.ts - consists of the models, and the rules for selecting those
models
prompts.ts - consists of the base prompt + specific system prompts
depending upon the usage
schema.ts
every single file under ai/tools folder - which as the name suggests,
consists of the implementations of the different tools that can be
provided to the llm
route-handlers - added support for streaming to SmartRoute and response
under dashboard
ai-search/route.ts - refactored the file to use unified ai endpoint
chat-adapters.ts - refactored the file to use unified ai endpoint and
created extra checks for the ai generated code
**Custom Dashboards**
We let the user write their query in english. We then use AI to create
dashboards that are interactive, live and savable. This PR includes a
new package called dashboard-ui-components. This package has components
that are used in the dashboard and more importantly, these components
are being imported from esm in the ai generated code for custom
dashboards. We also change the bar at the top for the products pages.
How to review this PR:
Review the new package (package/dashboard-ui-components), the setup and
the files inside it.
Review the schema changes in stack-shared/src
Review the changes in dashboard. The following changes have been made
Updated the design-components folder since we moved the dashboard
components to the new package
Updated imports for these components accordingly
Updated the title bar of the product pages
Created the files for custom dashboards under the dashboards folder and
components under commands/create-dashboard
Created a script under dashboard/scripts that generates the file with
type definitions that would go to the llm
Review the backend
Started using unified ai endpoint
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Added custom dashboards feature allowing users to create and manage
personalized dashboards with AI assistance.
* Integrated AI-assisted dashboard code generation with visual preview
and editing capabilities.
* Introduced new AI query endpoints supporting stream and generate modes
with configurable model quality/speed settings.
* **Improvements**
* Reorganized UI components into a dedicated component library package
for better code reuse.
* Enhanced chat architecture with improved message handling and tool
integration.
* Updated AI provider integration with improved configuration
management.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
---------
Co-authored-by: Konsti Wohlwend <[email protected]>
Co-authored-by: Bilal Godil <[email protected]>
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: Copilot <[email protected]>
<!--
Make sure you've read the CONTRIBUTING.md guidelines:
https://github.com/stack-auth/stack-auth/blob/dev/CONTRIBUTING.md
-->
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **Bug Fixes**
* Fixed authentication errors in the Claude agent.
* **Refactor**
* Simplified the CLI initialization process to default to cloud project
linking, removing interactive prompts.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
Ejiro Asiuwhu
MantraandGitHub
Konstantin Wohlwend
BilalG1andGitHub
Aman GanapathyandGitHub
MadisonandGitHub
