## Summary
- Inject non-secret application variables (`isSecret: false`) into front
component `process.env` via the existing Web Worker `setWorkerEnv`
mechanism
- Filter secret variables server-side in the resolver so they never
reach the browser
- Set application variables before system variables (`TWENTY_API_URL`,
`TWENTY_APP_ACCESS_TOKEN`) to prevent override
- Wire up environment variable keys in the logic function code editor
for TypeScript autocomplete
## Test plan
- [x] Unit tests for `buildNonSecretEnvVar` (6 passing)
- [x] Typecheck passes for `twenty-front` and `twenty-server`
- [x] Install an app with both `isSecret: false` and `isSecret: true`
variables, open a front component, verify only non-secret vars appear in
`process.env`
- [x] Open a logic function editor, verify autocomplete suggests
declared variable keys
## Summary
- Adds `ENCRYPTION_KEY` (primary) and `FALLBACK_ENCRYPTION_KEY`
(decrypt-only fallback for rotation) env vars to twenty-server, with
backward-compatible fallback to `APP_SECRET` when `ENCRYPTION_KEY` is
unset.
- Introduces a versioned ciphertext envelope `enc:v2:<keyId>:<base64>`
using AES-256-GCM with HKDF-SHA256 derived per-context keys. The 8-hex
`keyId` fingerprint lets every row identify which physical key encrypted
it, so rotation routes directly to primary or fallback without trial
decryption; GCM's auth tag gives true integrity (legacy CTR has none).
- Migrates `ConnectedAccountTokenEncryptionService` to the new envelope
and plumbs `workspaceId` through every caller, so per-workspace HKDF
context binds each row to its tenant.
The remaining encryption sites (`jwt-key-manager`, `config-storage`,
`postgres-credentials`, `application-variable`, TOTP) stay on the legacy
unprefixed CTR path and will be migrated in follow-up PRs. The
operator-facing rotation runbook is out of scope here.
### Format details
`enc:v{N}:{keyId}:{base64}` — `N=2` is the only version produced by new
writes (`v1` exists for backward-compatible decryption of existing
connected-account rows). `keyId =
sha256(rawKey).slice(0,4).toString('hex')`. The CHECK constraint on
`core.connectedAccount.{accessToken,refreshToken}` is relaxed from `LIKE
'enc:v1:%'` to `LIKE 'enc:v_:%'` so both versions pass.
### Key resolution
| `ENCRYPTION_KEY` | `FALLBACK_ENCRYPTION_KEY` | `APP_SECRET` | Encrypt
with | Decrypt try order |
|---|---|---|---|---|
| set | set | (any) | `ENCRYPTION_KEY` | match `keyId` → primary →
fallback |
| set | unset | (any) | `ENCRYPTION_KEY` | match `keyId` → primary |
| unset | set | set | `APP_SECRET` | match `keyId` → `APP_SECRET` →
fallback |
| unset | unset | set | `APP_SECRET` | match `keyId` → `APP_SECRET` |
| unset | unset | unset | startup error | n/a |
## Test plan
- [x] `npx nx typecheck twenty-server` — clean
- [x] `npx jest
'secret-encryption|connected-account-token-encryption|connected-account-refresh-tokens|encrypt-connected-account-tokens|connection-provider-oauth-flow'`
— 87 tests pass
- [x] New `secret-encryption.service.versioned.spec.ts` covers: key
resolution table (no-key error, APP_SECRET fallback, ENCRYPTION_KEY
precedence), v2 round-trip with/without workspaceId, GCM tamper
rejection, workspaceId-mismatch rejection, keyId-based primary→fallback
routing, missing-key error names the fingerprint, v1 legacy decryption,
no-prefix legacy decryption, malformed envelope rejection.
- [x] Updated `connected-account-token-encryption.service.spec.ts`
covers workspaceId binding and HKDF context isolation.
- [x] Updated slow instance command spec verifies workspaceId is
threaded through encryption and the relaxed `enc:v_:%` LIKE pattern
matches both v1 and v2.
- [ ] Manual E2E: connect a Gmail account on a freshly deployed instance
with `APP_SECRET` only → confirm `core.connectedAccount.accessToken` is
`enc:v2:<keyId>:<base64>`.
- [ ] Manual E2E: rotate — set `ENCRYPTION_KEY=<new>` and
`FALLBACK_ENCRYPTION_KEY=<old APP_SECRET>`, restart, confirm
pre-rotation rows still decrypt and new rows carry the new `keyId`.
- [ ] Manual E2E: missing key — set `ENCRYPTION_KEY=<new>` without the
fallback, confirm decrypt error names the old `keyId` so the operator
can identify the missing key.
## Summary
`ImapFlow` is an `EventEmitter`; per Node.js semantics, an emitted
`'error'` event with no listener becomes an uncaught exception that
exits the process. Both ImapFlow construction sites in `twenty-server`
(`ImapClientProvider` used by all messaging flows, and
`testImapConnection` in the connection-wizard validator) currently build
the client without attaching a permanent `'error'` listener, so a
transient socket condition (idle timeout, network blip, server-side
disconnect) crashes `twenty-server` and triggers a container restart
with a ~1 min HTTP 502 window for end users.
This patch attaches an `'error'` listener at each call site that logs
the error and lets `imapflow`'s internal reconnect handle recovery. Same
shape / same precedent as #20143 (Redis session-store client) which
fixed#20144.
Closes#20509.
## What changed
-
`packages/twenty-server/src/modules/messaging/message-import-manager/drivers/imap/providers/imap-client.provider.ts`:
`ImapClientProvider.createConnection` now attaches `client.on('error',
...)` between construction and `connect()`.
-
`packages/twenty-server/src/engine/core-modules/imap-smtp-caldav-connection/services/imap-smtp-caldav-connection.service.ts`:
`testImapConnection` does the same on its short-lived test client.
Both listeners log via the existing `Logger` instance (matching the
resolver-level logging already in `ImapClientProvider.getClient`) and
surface `error.stack` so transient socket conditions are observable but
no longer fatal.
## Crash this fixes (real production stack)
```
node:events:487
throw er; // Unhandled 'error' event
^
Error: Socket timeout
at TLSSocket.<anonymous> (/app/node_modules/imapflow/lib/imap-flow.js:795:29)
at TLSSocket.emit (node:events:509:28)
at Socket._onTimeout (node:net:610:8)
...
Emitted 'error' event on ImapFlow instance at:
at ImapFlow.emitError (/app/node_modules/imapflow/lib/imap-flow.js:397:14)
code: 'ETIMEOUT',
```
End-user impact: server process exits cleanly (code 0), Docker / k8s
restarts it; the DB, worker, redis, and caddy containers are unaffected
— only the API server dies, taking the GraphQL/REST surface offline for
a ~1 min health-check warmup.
## Test plan
- [x] `npx nx typecheck twenty-server` (planned — relying on CI for
verification)
- [x] `npx nx lint:diff-with-main twenty-server` (planned — relying on
CI for verification)
- [x] Manually reproduced the crash on `v2.2` by hitting an
IMAP/SMTP/CalDAV outbound flow with Gmail; with the patch applied
locally to the running container (verified the listener fires and logs
without process exit), the server stays up across the same trigger
sequence.
- [ ] Unit-level coverage: behavior is "listener exists, doesn't throw"
— not easily covered without a contrived socket-mock test. Existing call
sites have no unit tests today; happy to add one if a reviewer prefers,
otherwise mirroring the convention from #20143 which merged without a
new test.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: neo773 <neo773@protonmail.com>
Billing is now decremented per-step, not per-turn. The onStepFinish
callback in chat-execution.service.ts calls a new
decrementAndCheckAvailableCredits method on each model step, so Redis is
debited incrementally as the agent runs rather than all at once at the
end.
Credit exhaustion stops the stream mid-run. When a step depletes the
remaining credits, a hasNoMoreAvailableCredits flag is set and passed
into the stopWhen predicate of streamText, causing the agent to halt
before starting the next step.
A new credits-exhausted event is introduced. After the stream drains and
the response is persisted, if credits ran out the job publishes a
dedicated credits-exhausted event to the frontend instead of the normal
message-persisted event.
The frontend handles this new event. useAgentChatSubscription has a new
credits-exhausted case that sets a BILLING_CREDITS_EXHAUSTED-coded error
on the atom, closes the writer, and stops the streaming state —
triggering the existing AiChatCreditsExhaustedMessage UI.
## Summary
Fixes#20502
/claim #20502
The navigation drawer's collapsed state is persisted to \`localStorage\`
via \`isNavigationDrawerExpandedState\`. When a user collapses the
drawer in the main app and then opens settings via a **direct URL,
refresh, or new tab**, the settings layout renders in collapsed mode —
no \`useOpenSettingsMenu\` call is made in those paths to force
expansion.
\`StyledAnimatedContainer\` (which controls the outer drawer width) used
raw \`isNavigationDrawerExpanded\` with no settings-route override.
Inner components (\`NavigationDrawerItemsCollapsableContainer\`) already
guard with \`isExpanded = isNavigationDrawerExpanded || isSettingsPage\`
— the outer container simply needed the same treatment.
**Fix:** derive \`isExpanded = isSettingsDrawer ||
isNavigationDrawerExpanded\` using the already-in-scope
\`useIsSettingsDrawer()\` result and pass it to both
\`StyledAnimatedContainer\` and \`StyledContainer\`. 1 derived variable,
2 prop changes, no new hooks or state.
**Changed files:**
-
\`packages/twenty-front/src/modules/ui/navigation/navigation-drawer/components/NavigationDrawer.tsx\`
## Test plan
Verified via code review and CI. The fix is structurally identical to
the existing \`isSettingsPage\` guard already used in
\`NavigationDrawerItemsCollapsableContainer\` — the outer container
simply lacked the same treatment. No new hooks, no side effects, no
state mutations.
Manual UI verification (collapse → navigate to settings → confirm
expanded) was not performed against a running instance. If the
maintainers want to verify, the logic path is the same as the inner
component guard that already ships in production.
> **Note (2026-05-12):** PR #20508 was submitted after this PR with a
\`useEffect\`-based approach. That approach has already received a
review comment from a team member noting that \`useEffect\` should be a
last resort per the project's own React guidelines. This fix uses no
effects or imperative state — only a derived variable.
---
> [!NOTE]
> **AI-Assisted Contribution**
> This patch was generated by
[Mesopredator](https://github.com/GusFromSpace), an autonomous code
intelligence system.
> Static analysis located the bug, the fix was written and verified with
\`tsc --noEmit\` + \`node\` test suite, and reviewed by a human before
submission.
> Please treat this as a community contribution and request changes if
needed.
## Issue
As per the developer docs, the local setup uses the `npx nx
database:reset twenty-server` command, which seeds 4 workspaces. This
works correctly for multi-workspace mode
(`IS_MULTIWORKSPACE_ENABLED=true`) and integration tests but causes
issues in single-workspace mode (`IS_MULTIWORKSPACE_ENABLED=false`) or
when switching from multi-workspace mode back to single-workspace mode.
Also, the default mode is single-workspace but 4 workspaces are already
seeded in the database. As a result,
`WorkspaceDomainsService.getDefaultWorkspace()` selects the newest
workspace (Empty4), which is intended only for integration testing and
contains no user data.
There is also an existing warning log mentioning fallback to the Apple
seed workspace when multiple workspaces are found in single-workspace
mode i.e `IS_MULTIWORKSPACE_ENABLED=true`, but it was never implemented.
```
if (workspaces.length > 1) {
Logger.warn(
` ${workspaces.length} workspaces found in database. In single-workspace mode, there should be only one workspace. Apple seed workspace will be used as fallback if it found.`,
);
}
```
Although we could replace with `"nx command-no-deps --
workspace:seed:dev --light"` in `project.json` for `database:reset`,
which will only seed one workspace but it wont resolve issue when
switching from multi-workspace mode back to single-workspace mode or for
integration test `with-db-reset`.
## Changes
- Implement fallback behavior already hinted by existing warning logs.
- Ensure the Apple seed workspace is used as the fallback in
single-workspace mode when multiple workspaces exist.
This improves:
- Local developer onboarding experience.
- Switching between multi-workspace and single-workspace development
modes.
- Consistency during local development and integration testing.
## Related PR
- #19822
- #20464
These PRs only resolves the issue for docker environments but not for
local development setup.
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
**Root cause:** getWorkflowRunContext(stepInfos) builds a Record<string,
unknown> from the previous steps' results. There is no workspaceId key
in it, so context.workspaceId as string silently evaluated to undefined.
That undefined was then passed all the way down to
WorkspaceCacheService.getOrRecompute, **which correctly throws** when
workspaceId is not a valid UUID.
Before :
<img width="525" height="130" alt="Screenshot 2026-05-13 at 14 58 54"
src="https://github.com/user-attachments/assets/0549b4dc-7063-44e5-95a1-00a460a6d7f1"
/>
Introduced with billing v2 yesterday, since then, workspaceId is needed
to bill credit usage
# Introduction
Adding `no-miused-promise` lint rule to the twenty-server
In order to flag such pattern
```ts
// ❌ Flagged — forEach doesn't await the callback
items.forEach(async (item) => {
await process(item);
});
```
## What happened
- Refactored the code-interpreter driver to have a async onResult (
which is also expected by e2b )
- Workspace manager still dirty solution including force cast
- Basic fixes
## Summary
- Replace the dynamic `RestApiMetadataController` (which parsed
`/rest/metadata/*path` and proxied to internal GraphQL) with two
dedicated controllers: `ObjectMetadataController` and
`FieldMetadataController`.
- Drop the GraphQL hop: reads hit Postgres directly via TypeORM
repositories; writes call the existing
`{create,update,delete}One{Object,Field}` service methods.
- Introduce a new clean response shape behind a workspace feature flag
(`IS_REST_METADATA_API_NEW_FORMAT_DIRECT`) — see grace period below.
- Update the OpenAPI spec so the REST playground reflects the (default)
legacy shape during the grace period.
## Why
The legacy metadata controller was over-complex: it routed every method
through a path parser, a set of GraphQL query-builder factories, an
internal GraphQL call, and a
`cleanGraphQLResponse` post-processor. Operation names from GraphQL
(`createOneObject`, `updateOneField`, …) leaked straight into REST
responses. The internal-GraphQL hop also gave us
nothing on metadata reads — pagination, filtering, and serialization all
happen against the same Postgres tables either way.
## Feature flag & grace period
`IS_REST_METADATA_API_NEW_FORMAT_DIRECT` (workspace-scoped):
- **Existing workspaces:** flag absent → resolves to `false` → **legacy
response shape** (no behavior change).
- **Newly created workspaces:** flag seeded to `true` via
`DEFAULT_FEATURE_FLAGS` → **new response shape** from day one.
- **Toggle:** support-assisted (no frontend); customers contact us to
opt into the new shape early.
- **Removal:** the flag, the legacy adapter utils
(`to-legacy-{object,field}-metadata-response.util.ts`), and the
parametrized test wrapper get deleted after the grace window. New shape
becomes the only shape; OpenAPI flips to new shape; POST loses the
conditional and reverts to a declarative response.
## Response shapes
| Operation | Legacy (flag OFF, default for existing) | New (flag ON) |
|-----------|-----------------------------------------|---------------|
| `GET /rest/metadata/objects` | `{ data: { objects: [...] }, pageInfo,
totalCount }` | `{ data: [...], pageInfo, totalCount }` |
| `GET /rest/metadata/objects/:id` | `{ data: { object: {...} } }` | `{
... }` |
| `POST /rest/metadata/objects` | `201 { data: { createOneObject: {...}
} }` | `201 { ... }` |
| `PATCH/PUT /rest/metadata/objects/:id` | `{ data: { updateOneObject:
{...} } }` | `{ ... }` |
| `DELETE /rest/metadata/objects/:id` | `{ data: { deleteOneObject: {
... } } }` | `{ ... }` |
Same matrix for `/rest/metadata/fields`. Cursor params
(`starting_after`, `ending_before`, `limit`) and `totalCount` are
preserved across both shapes. POST returns `201` in both (old
controller already did — the doc on main saying `200` was wrong).
## Implementation notes
- Reads go straight to Postgres with TypeORM cursor pagination
(`paginateByIdCursor` util, mutually-exclusive `starting_after` /
`ending_before`). No cache on this path — caching +
filterable pagination didn't combine cleanly.
- Object endpoints inline `fields[]` via a single follow-up `WHERE
objectMetadataId IN (...)` query.
- Controllers read the flag via `FeatureFlagService.isFeatureEnabled`
and conditionally pass the result through a legacy-shape adapter util
before returning.
- Per-domain REST exception filters
(`{Object,Field}MetadataRestApiExceptionFilter`); the `exceptionCode →
httpStatus` switch is extracted to a util so it can be merged with the
existing GraphQL handler later.
- New controllers live inside the metadata domain modules
(`metadata-modules/{object,field}-metadata/controllers/`) to match
existing precedent (view-field, view, page-layout, …).
- Removes: `RestApiMetadataController`, `RestApiMetadataService`,
`metadata/query-builder/`, `clean-graphql-response.utils.ts`.
- Integration tests are parametrized over both flag values via
`describe.each` — both shapes are asserted in CI.
- OpenAPI fixes inherited from the migration (kept as-is): documents
flat `fields: [...]` rather than the obsolete `{edges:{node:[...]}}`
wrapping; always emits `totalCount`; POST
status `201`. These match what customers actually receive on both
shapes.
Note: Next goal is to implement something similar for graphql and remove
nestjs-query dependency for those 2 entities, then generalise it.
Note2: We have the same issue with Core Rest API such as
```json
{
"data": {
"createCompany": {
"id": "123e4567-e89b-12d3-a456-426614174000",
"createdAt": "2026-05-07T12:14:52.769Z",
"updatedAt": "2026-05-07T12:14:52.769Z",
"deletedAt": "2026-05-07T12:14:52.769Z",
...
```
with "createCompany" here which is odd compared to REST standards (FYI
@etiennejouan @charlesBochet)
## Before (Without feature flag)
<img width="1346" height="712" alt="Screenshot 2026-05-12 at 20 50 38"
src="https://github.com/user-attachments/assets/316ce225-1045-4aac-97a9-60fd537eb1ec"
/>
<img width="1378" height="729" alt="Screenshot 2026-05-12 at 20 52 24"
src="https://github.com/user-attachments/assets/a621ab6f-e4f8-44d5-817c-1efd25d33c30"
/>
## After (With feature flag)
<img width="1376" height="728" alt="Screenshot 2026-05-12 at 20 50 46"
src="https://github.com/user-attachments/assets/2424d9c5-e4ed-497c-8e5c-6b54d78675e4"
/>
<img width="1375" height="727" alt="Screenshot 2026-05-12 at 20 51 47"
src="https://github.com/user-attachments/assets/101d957f-38ed-45d9-ab7b-f4f4eb983397"
/>
---------
Co-authored-by: prastoin <paul@twenty.com>
## Summary
Closes#20195
Fix phone field unique constraints so phone numbers are considered
unique by both `primaryPhoneNumber` and `primaryPhoneCallingCode`.
- Include `primaryPhoneCallingCode` in the shared phone composite unique
constraint metadata
- Align the frontend settings composite field config with the backend
metadata
- Return all included unique composite subfields when building
create-many conflict fields
- Match composite unique conflict fields as a group during create-many
upserts
## Root Cause
Phone composite metadata only marked `primaryPhoneNumber` as part of the
unique constraint. That made different international phone numbers with
the same national number conflict, for example `+1 123456789` and `+32
123456789`.
## Test Plan
- `yarn workspace twenty-shared build`
- `jest --runTestsByPath <index action handler and create-many utility
specs>`
- `prettier --check <touched files>`
- `oxlint --type-aware <touched files>`
- `nx run twenty-shared:typecheck`
- `nx run twenty-server:typecheck`
- `nx run twenty-front:typecheck`
---------
Co-authored-by: mkdev11 <MkDev11@users.noreply.github.com>
Co-authored-by: Etienne <45695613+etiennejouan@users.noreply.github.com>
Co-authored-by: prastoin <paul@twenty.com>
## Summary
Extends the asymmetric signing work from #20467 to cover **every
remaining `JwtTokenTypeEnum` value**: `LOGIN`, `WORKSPACE_AGNOSTIC`,
`FILE`, `API_KEY`, `APPLICATION_ACCESS`, `APPLICATION_REFRESH`,
`APP_OAUTH_STATE`, plus the ACCESS-shaped session token issued by the
code interpreter tool.
After this PR, every JWT the server signs is ES256 with a `kid` pointing
at the current `core."signingKey"` row, while legacy HS256 tokens (no
`kid` header) remain verifiable indefinitely through the existing
fallback in `JwtWrapperService.resolveVerificationKey`. No new entity /
migration / config: this is a pure routing change on top of the
infrastructure that already shipped.
## Why
`#20467` only flipped `ACCESS` and `REFRESH` to ES256. Every other JWT
type was still HS256-signed against the global `APP_SECRET`, which kept
the original blast radius (a leaked `APP_SECRET` invalidates *every* JWT
type forever). Migrating the rest unifies the sign path on rotatable
per-server private keys without forcing any token reissue.
## Mechanical changes
### Sign side (8 services)
- `LoginTokenService.generateLoginToken`
- `TransientTokenService.generateTransientToken`
- `WorkspaceAgnosticTokenService.generateWorkspaceAgnosticToken`
- `ApplicationTokenService.signApplicationToken` (`APPLICATION_ACCESS` +
`APPLICATION_REFRESH`)
- `ApiKeyService.generateApiKeyToken`
- `FileUrlService.signFileByIdUrl` / `signWorkspaceLogoUrl`
- `ConnectionProviderOAuthFlowService.signState` (`APP_OAUTH_STATE`)
- `CodeInterpreterTool.generateSessionToken`
Each call site swaps `jwtWrapperService.sign(payload, { secret:
generateAppSecret(...), ... })` for `await
jwtWrapperService.signAsync(payload, { expiresIn, [jwtid] })`. The
`generateAppSecret` calls on the sign side are dropped (verifier-side
`generateAppSecret` stays in `resolveVerificationKey` for the HS256
fallback).
### Verifier side
- `WorkspaceAgnosticTokenService.validateToken` now goes through
`verifyJwtToken` instead of the bespoke `verify({ secret })` path, so
new ES256 tokens are accepted while the legacy HS256 fallback inside
`resolveVerificationKey` still serves the old shape.
- `JwtWrapperService.sign()` is kept (legacy compat / tests) but is now
strictly deprecated — there are no remaining production callers.
### Async ripple (`signFileByIdUrl` was synchronous)
- `FileUrlService.signFileByIdUrl` and `signWorkspaceLogoUrl` are now
`async`; the `signUrl` callback used by `getRecordImageIdentifier` is
widened to accept `Promise<string | null>`.
- Every direct/indirect caller is updated: admin panel (user lookup +
statistics + top workspaces), search service
(`computeSearchObjectResults`, `getImageIdentifierValue`), workspace
resolver (`logo` resolver, public workspace by domain/id),
`WorkspaceMemberTranspiler` (now `async toWorkspaceMemberDto[s]` /
`toDeletedWorkspaceMemberDto[s]` / `generateSignedAvatarUrl`),
`UserService.loadSignedAvatarUrlsByUserId`,
`UserWorkspaceService.castWorkspaceToAvailableWorkspace`,
workspace-invitation, approved-access-domain, agent-chat-streaming,
agent-message-part resolver, navigation-menu-item record identifier,
file-ai-chat / file-core-picture / file-email-attachment / file-workflow
/ files-field services, rich-text & files-field query result getters,
and the code-interpreter tool.
## Backward compatibility
- **Legacy HS256 tokens (no `kid`)** keep verifying via
`resolveVerificationKey` → `extractAppSecretBody` → `generateAppSecret`
for both `workspaceId`-bearing and `userId`-bearing payloads.
- The `API_KEY` HS256-via-ACCESS-secret fallback (#16504) still kicks in
inside `verifyJwtToken` for pre-2025-12-12 API keys.
- No payload shape changes, no DB writes, no env var changes — old
tokens issued by `main` continue to authenticate.
## Tests
### Unit (all green locally — 63/63)
Updated specs for every migrated service to mock `signAsync` instead of
`sign` and assert the new option shape:
- `login-token.service.spec.ts`, `transient-token.service.spec.ts`,
`workspace-agnostic-token.service.spec.ts`,
`application-token.service.spec.ts`, `api-key.service.spec.ts`,
`connection-provider-oauth-flow.service.spec.ts`.
### Integration (`jwt-key-rotation.integration-spec.ts`)
- Existing ACCESS coverage (current key, legacy HS256 fallback,
rotated-out key, revoked key, unknown kid) is preserved.
- New `it.each` assertion: `REFRESH`, `WORKSPACE_AGNOSTIC`, and `LOGIN`
tokens emitted by the real signUp → signUpInNewWorkspace →
getAuthTokensFromLoginToken pipeline are ES256 with a `kid` matching the
current signing key — proves end-to-end that the migration didn't
regress those flows.
## Open question (separate decision)
This PR keeps the legacy HS256 verification fallback **forever**. We may
eventually want to sunset it for `API_KEY` once telemetry shows
pre-migration tokens are gone, but that's a separate product/security
decision and not part of this change.
## Test plan
- [ ] CI green
- [ ] `npx nx lint:diff-with-main twenty-server` passes
- [ ] `npx nx typecheck twenty-server` passes
- [ ] `jwt-key-rotation` integration suite passes (new + existing
assertions)
- [ ] Manually verify: signing in issues an ES256 ACCESS / REFRESH
token, generating an API key issues an ES256 token with `kid`, signed
file URL JWT is ES256 with `kid`
- [ ] Pre-existing HS256 tokens still authenticate (covered by
integration test, but worth a manual check with a token from `main`)
Automated daily sync of `ai-providers.json` from
[models.dev](https://models.dev).
This PR updates pricing, context windows, and model availability based
on the latest data.
New models meeting inclusion criteria (tool calling, pricing data,
context limits) are added automatically.
Deprecated models are detected based on cost-efficiency within the same
model family.
**Please review before merging** — verify no critical models were
incorrectly deprecated.
Co-authored-by: FelixMalfait <6399865+FelixMalfait@users.noreply.github.com>
## Context
Calling `POST /rest/views` (and other metadata mutations) currently
returns a generic `500` for user-input failures:
Ex:
1. **Invalid `objectMetadataId`** —
`resolveEntityRelationUniversalIdentifiers` throws
`FlatEntityMapsException(RELATION_UNIVERSAL_IDENTIFIER_NOT_FOUND)`.
Should be `404`.
2. **Missing required field** (e.g. `icon`) — Postgres raises a `NOT
NULL` violation, wrapped as
`WorkspaceMigrationRunnerException(EXECUTION_FAILED)` carrying a
`QueryFailedError`. Should be `400`.
Neither was caught by `ViewRestApiExceptionFilter`, so both fell through
to `UnhandledExceptionFilter` and were emitted as `500`s without
reaching Sentry.
Same gap existed on most metadata GraphQL resolvers — only
`page-layout*` and `role` resolvers covered
`WorkspaceMigrationRunnerException` via
`WorkspaceMigrationGraphqlApiExceptionInterceptor`.
## Changes
### New filters
REST (`HttpExceptionHandlerService` + Sentry-aware):
- `FlatEntityMapsRestApiExceptionFilter` — maps
`RELATION_UNIVERSAL_IDENTIFIER_NOT_FOUND` / `ENTITY_NOT_FOUND` → `404`,
`ENTITY_ALREADY_EXISTS` → `409`, others → `500`.
- `WorkspaceMigrationRunnerRestApiExceptionFilter` — for
`EXECUTION_FAILED`, unwraps the underlying `metadata` /
`workspaceSchema` / `actionTranspilation` error; if it's a
`QueryFailedError` it gets remapped to `400` via
`HttpExceptionHandlerService`. `APPLICATION_NOT_FOUND` → `404`,
`DDL_LOCKED` → `503`, otherwise `500`.
GraphQL (graphql-errors + existing formatter):
- `FlatEntityMapsGraphqlApiExceptionFilter` — kept as the GraphQL-shaped
counterpart (`NotFoundError` / `InternalServerError`).
- `WorkspaceMigrationRunnerGraphqlApiExceptionFilter` — reuses
`workspaceMigrationRunnerExceptionFormatter` for parity with the
existing interceptor.
### Wiring
Filters are now declared **per controller / resolver** via `@UseFilters`
(no global `APP_FILTER` registration) so they participate in the normal
NestJS filter chain instead of being preempted by
`UnhandledExceptionFilter`.
REST:
- `view.controller.ts` — adds `FlatEntityMapsRestApiExceptionFilter` and
`WorkspaceMigrationRunnerRestApiExceptionFilter`.
GraphQL (14 resolvers, all that mutate flat entities):
- `FlatEntityMapsGraphqlApiExceptionFilter` added to: `view`,
`view-field`, `view-field-group`, `view-sort`, `view-group`,
`view-filter`, `view-filter-group`, `page-layout`, `page-layout-tab`,
`page-layout-widget`, `role`, `object-metadata`, `field-metadata`,
`index-metadata`.
- `WorkspaceMigrationRunnerGraphqlApiExceptionFilter` added to the same
list **except** the four already covered by
`WorkspaceMigrationGraphqlApiExceptionInterceptor` (`page-layout`,
`page-layout-tab`, `page-layout-widget`, `role`) — to avoid
double-handling.
## Why per-resolver / per-controller instead of global
Earlier attempt to register the filters globally via `APP_FILTER`
regressed: NestJS reverses the global filter list and
`selectExceptionFilterMetadata` is first-match-wins, so
`UnhandledExceptionFilter` (registered last via `app.useGlobalFilters`
in `main.ts`) ended up first in the iteration order and preempted every
domain-specific filter. The per-resolver / per-controller approach is
explicit and predictable.
## Before
<img width="953" height="450" alt="Screenshot 2026-05-12 at 15 31 40"
src="https://github.com/user-attachments/assets/3c3bc6a8-f6bc-4032-97d0-7243540cfb90"
/>
## After
<img width="1050" height="598" alt="Screenshot 2026-05-12 at 15 31 17"
src="https://github.com/user-attachments/assets/c66c9ce5-d1ea-4f1d-b2fe-07979e2261f7"
/>
<img width="1068" height="503" alt="Screenshot 2026-05-12 at 15 31 09"
src="https://github.com/user-attachments/assets/ddd9eed8-812b-47d6-96cb-b019b807991b"
/>
## Context
Today every JWT issued by Twenty (access, refresh, login, file, etc.) is
HMAC-signed with a per-token-type secret derived from the global
`APP_SECRET`. Rotating that secret invalidates **every** active token at
once and there is no way to scope a leak to a subset of tokens.
This PR is the first slice of a broader effort to **decouple stateful
encryption (`APP_SECRET`-derived secrets) from stateless encryption
(JWTs)**. It introduces an asymmetric (private/public key) signing path
for `ACCESS` and `REFRESH` tokens and a signing-key registry to enable
**safe rotation**: leaked keys can be revoked by flipping
`revokedAt`/`isCurrent` on the matching row without invalidating tokens
issued by other keys.
> Out of scope (intentionally): swapping stateful encryption for
`APP_SECRET`, asymmetric signing for token types other than
`ACCESS`/`REFRESH`, an admin-panel rotation UI, and an enterprise
re-encryption command. Those will land in follow-up PRs.
## What changes
- **New `core.signingKey` table** (instance command `2.5.0` /
`1778550000000`) storing both the public key (PEM, in clear) and the
private key (PEM, encrypted with `APP_SECRET` via
`SecretEncryptionService`). One row is marked `isCurrent = true`
(enforced by a partial unique index). The row's UUID `id` is used
directly as the JWT `kid`.
- When a key is rotated out, its `privateKey` is nulled (we never keep
historical private keys) but the `publicKey` row stays so previously
issued tokens can still be verified.
- **`JwtKeyManagerService`** lazily loads-or-generates the current
signing key on first use:
- If a row with `isCurrent = true` exists → decrypts and uses it.
- Otherwise → generates a fresh EC P-256 keypair, encrypts the private
key, inserts the row (UUID id = kid). Handles concurrent insert races
via the unique constraint.
- **`JwtWrapperService.signAsync()`** signs `ACCESS`/`REFRESH` payloads
with `ES256` and a `kid` header. Falls back to `HS256` if no signing key
is available (boot-time DB error, transient failure).
- **Dual-path verification** in both `JwtWrapperService.verifyJwtToken`
and the Passport `JwtAuthStrategy.secretOrKeyProvider`:
- JWT with a `kid` header → resolve the public key PEM by id and verify
with `ES256`,
- otherwise → fall back to the existing `APP_SECRET`-derived `HS256`
path (unchanged).
- **`AccessTokenService` / `RefreshTokenService`** now sign through
`signAsync` (single public surface; the routing detail stays internal to
the wrapper).
- **Public key cache**: a new `SigningKeyEntityCacheProviderService`
plugs into `CoreEntityCacheService` (`signingKeyPublicKey` namespace)
and serves PEMs by id, with the standard local-memo + Redis layering.
- **PEM strings end-to-end**: `jsonwebtoken` accepts PEM strings
directly for both sign and verify, so the manager never converts to a
Node `KeyObject` and the cache hands the PEM straight to `jwt.verify`.
## Why ES256 (and not EdDSA / RS256)
- `@nestjs/jwt` is backed by `jsonwebtoken`, which does **not** support
EdDSA today.
- ES256 keys are tiny (~120 bytes vs 1.6 kB for RS256), signatures are
short (~64 bytes), and signing/verification is fast — important since
JWT verification runs on every authenticated request.
- ES256 is widely supported and standardized (RFC 7518), with mature
ecosystem support.
## Why store the private key in DB (not env)
- No new secret to provision: existing instances already have
`APP_SECRET`, which we reuse to encrypt the private key at rest.
- Self-healing: a fresh instance auto-generates its first signing key on
first boot. Nothing to copy/paste.
- Rotation is a SQL operation against `core.signingKey`, not a redeploy
+ env mutation.
## Backward compatibility
- All previously-issued tokens (no `kid`) keep verifying through the
legacy HS256 path with their existing `APP_SECRET`-derived secret. No
forced re-login.
- Token types not in scope (`WORKSPACE_AGNOSTIC`, `API_KEY`, `FILE`,
`LOGIN`, `EMAIL_VERIFICATION`, etc.) keep their current HS256 behavior
unchanged — they still go through the synchronous
`JwtWrapperService.sign(payload, options)` with a caller-supplied
secret.
- `signWithAppSecret` is kept intentionally as the HS256 fallback path;
it will be deprecated in a follow-up PR.
- If the DB lookup/generation fails for any reason, the wrapper logs the
error and falls back to HS256 — no startup crash, no silent regression.
## Rotation story
1. Bootstrap: first signing call lazily inserts a row in
`core.signingKey` with `isCurrent = true`, `privateKey =
encrypt(pem_A)`. New tokens carry `kid_A`.
2. Rotate: `UPDATE core."signingKey" SET "isCurrent" = false,
"privateKey" = NULL WHERE id = '<kid_A>';` then insert a new row with
`isCurrent = true`. New tokens carry `kid_B`. Tokens still in flight
with `kid_A` keep verifying because the public-key row for `kid_A` is
still there.
3. Revoke: `UPDATE core."signingKey" SET "revokedAt" = now() WHERE id =
'<kid_A>';`. All tokens with `kid_A` now fail verification cleanly with
`UNAUTHENTICATED` (no 500).
4. Tokens with no `kid` (legacy) are unaffected throughout.
## Test plan
- [x] Unit: `JwtWrapperService` dual-path verification (HS256 no-kid vs
ES256 with-kid), unknown-kid → `UNAUTHENTICATED`, `signAsync` happy path
+ `null` when no key, `signAsync` rejection for non-rotatable types.
- [x] Unit: `JwtAuthStrategy` `secretOrKeyProvider` dual-path resolution
and algorithm validation.
- [x] All existing JWT/auth/application unit tests adjusted to the
renamed public method.
- [x] Integration (`jwt-key-rotation.integration-spec.ts`):
- **Happy path**: signed-up user's `ACCESS` token has `alg=ES256` +
correct UUID `kid`, the `isCurrent=true` row exists in
`core.signingKey`, `getCurrentUser` resolves.
- **Legacy fallback**: hand-crafted no-kid HS256 token verifies via the
legacy `APP_SECRET`-derived path.
- **Previous-key rotation**: token signed by a hardcoded *previous* key
whose row is pre-inserted with `privateKey = NULL` (rotated-out) still
verifies — proves the leaked-key revocation flow works in both
directions.
- **Unknown kid**: token signed with an orphan UUID `kid` is cleanly
rejected (no 500).
- [x] `npx nx typecheck twenty-server`
- [x] `npx nx test twenty-server`
- [x] `npx nx run twenty-server:lint`
## Summary
On kanban cards, the title was being truncated even when the checkbox
wasn't displayed. The checkbox is hidden via `opacity: 0` on the card's
non-hovered state, which keeps it in flex flow and still reserves its
~24px of width — so the title's flex item was shrinking unnecessarily.
This change collapses the checkbox container's `max-width` to `0` (with
`overflow: hidden`) while it's hidden, and expands it back to the
checkbox's natural size (`spacing[6]` = 24px) on hover or when selected.
The existing `transition: all ease-in-out 160ms` animates the title
expanding into the freed space.
### Before
Title truncates with ellipsis even though the checkbox slot is empty:
<img width="350" alt="before"
src="https://i.imgur.com/placeholder-before.png" />
### After
Title uses the full row width when not hovered; the checkbox slides in
on hover (or when the card is selected) and the title reflows.
### Tooltip
The full title is already exposed on hover when truncated — `RecordChip`
→ `Chip` already wraps the label in `OverflowingTextWithTooltip`, which
detects overflow (`scrollWidth > clientWidth`) and renders an
`AppTooltip` with the full text. No additional wiring needed.
## Test plan
- [ ] On a kanban board, verify a long record title now uses the full
card width when the card is not hovered (no ellipsis if the title fits).
- [ ] Hover the card: the checkbox slides in smoothly (animated), and
the title reflows (may now truncate if it doesn't fit).
- [ ] Hover the (now-truncated) title: tooltip with the full title
appears.
- [ ] Select the card via the checkbox: checkbox stays visible (and
title stays in its hover-state width) without hovering.
- [ ] Compact view (eye icon) still renders correctly.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
## Summary
Two related cleanups, following the same pattern as #19916 and #19074.
### Dead feature flags
Drops four feature flags whose only references are the enum entry and
the generated GraphQL/SDK files:
- `IS_COMMAND_MENU_ITEM_ENABLED` — never read anywhere.
- `IS_DATASOURCE_MIGRATED` — already commented `@deprecated`. Zero
non-generated consumers.
- `IS_RICH_TEXT_V1_MIGRATED` — the 1-19 migration that gated it was
removed in #19074; the flag became dead at that point.
- `IS_CONNECTED_ACCOUNT_MIGRATED` — only read by the 1-21
`migrate-messaging-infrastructure-to-metadata` command as an
early-return guard, but the flag was never written anywhere in the
codebase, so the guard never fired (and that workspace command is now
removed entirely — see below).
Generated GraphQL/SDK schemas and the `workspace-entity-manager` test
mock are trimmed to match.
### 1-21 workspace commands
Same pattern as #19074 (which removed workspace commands ≤ 1.18). Twenty
is now on 2-5; the 1-21 workspace commands have long since run on every
active workspace and are dead code.
Removes:
- All 14 workspace commands under `upgrade-version-command/1-21/`
(compose-email menu item, key-value-pair index, datasource backfill,
message-thread backfill, dedup engine commands, select-all fixes, AI
response format migration, edit-layout label, drop messaging FKs, folder
parent-id migration, messaging-infra-to-metadata, navigation refactor,
message-thread label fix, search-menu-item label).
- The `1-21-upgrade-version-command.module.ts` registration and the
`V1_21_UpgradeVersionCommandModule` import from
`WorkspaceCommandProviderModule`.
**Kept** (intentionally): the 3 `1-21-instance-command-fast-*` files.
Unlike workspace commands (which mutate data), instance commands carry
**schema deltas** still required by current entity definitions
(`AddViewFieldGroupIdIndex`, `MigrateMessagingCalendarToCore`,
`AddEmailThreadWidgetType`). They remain registered in
`INSTANCE_COMMANDS` and `'1.21.0'` stays in `TWENTY_PREVIOUS_VERSIONS`.
They will fold away naturally on a future version bump when a
`CoreMigrationCheck`-style snapshot picks them up.
## Test plan
- [x] `npx nx typecheck twenty-shared`
- [x] `npx nx typecheck twenty-server`
- [x] `npx nx typecheck twenty-front`
- [x] `npx prettier --check` on the changed files
- [x] `npx oxlint` on the changed server files
- [x] `npx jest feature-flag` (4 suites, 20 tests pass)
- [x] `npx jest workspace-entity-manager` (1 suite, 5 tests pass)
## Summary
Follow-up to #20464. That PR added `--light` to the preview env seed
command but left the `--` between `yarn command:prod` and the script
args. After yarn strips its own `--`, nest-commander still sees `argv:
[..., '--', 'workspace:seed:dev', '--light']`. Commander.js treats `--`
as the end-of-options marker, so `--light` is parsed as a positional arg
and silently ignored — the seed runs in full mode (Apple + YCombinator +
Empty3 + Empty4) and Empty4 still ends up as the default workspace.
## Evidence
In the preview run on `f706cc052b` (which had #20464's `--light` flag),
the seed step took only ~40s but the `GqlTypeGenerator` log emits four
regenerations across two workspaces with custom objects:
- 28 standard → 28 + 5 custom (`rocket, surveyResult, employmentHistory,
petCareAgreement, pet`) — matches Apple
- 28 standard → 28 + 1 custom (`surveyResult`) — matches YCombinator
With `--light` actually applied, `getLightConfig` returns `{ objects:
[], fields: [] }` so no custom objects should be generated.
The working `twenty-app-dev` invocation in
`packages/twenty-docker/twenty-app-dev/rootfs/etc/s6-overlay/scripts/init-db.sh:66`
is `yarn command:prod workspace:seed:dev --light` — no `--`. Matching
that fixes it.
## Test plan
- [ ] Trigger the preview-app label on a PR, confirm only the Apple
workspace is created and `tim@apple.dev` signs in there
- [ ] Confirm the seed step still passes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
## Summary
The `GraphQL and OpenAPI Breaking Changes Detection` workflow has been
posting graphql-inspector stack traces as PR comments — see [#20445
comment](https://github.com/twentyhq/twenty/pull/20445#issuecomment-4421142635)
for an example.
### Root cause
- The wait step probed readiness with `curl -s URL > /dev/null 2>&1`,
which exits 0 for **any** HTTP response — including 5xx and GraphQL
error JSON. NestJS opens the HTTP listener before the workspace schema
cache is fully populated, so the wait often completed while the server
still served auth/metadata error JSON.
- The introspection download therefore wrote a small (~154-byte) error
payload instead of the real schema. `jq empty` in the validation step
only checks JSON *syntax*, so `{"errors":[...]}` passed validation.
- `graphql-inspector diff` then failed with `Unable to read JSON file:
... Not valid JSON content`, the workflow swallowed the error into the
diff markdown, and the bot posted that stack trace verbatim on the PR.
In the failing run, the main-branch files were 154 B (GraphQL) and 112 B
(REST 500); the current-branch files in the same run were 600 KB–2.8 MB.
### Fix
- Wait steps now POST an authenticated introspection (`{ __schema {
queryType { name } } }`) and require `.data.__schema` plus a 2xx
response from `/rest/open-api/core` (`curl -f`) before declaring the
server ready.
- Validation step now checks for the expected shape (`.data.__schema`
for GraphQL, `.openapi`/`.swagger` for OpenAPI) and includes the first
200 bytes of any bad payload in the warning, so when something genuinely
goes wrong the next debugger has a real lead instead of a generic stack
trace.
## Test plan
- [ ] CI runs against this branch — the workflow's own readiness probes
are now exercised against the real server, so a green run validates the
new check.
- [ ] If the readiness probe still passes but downloads regress, the
strengthened validation step will surface the payload in the workflow
logs instead of posting a graphql-inspector stack trace on the PR.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Adds an optional label field to logic-function input schema properties
(InputSchemaProperty and InputJsonSchema). When set, the workflow
builder renders the label in place of the raw property key for both leaf
inputs and nested sections; when unset, it falls back to the key.
jsonSchemaToInputSchema propagates the label so app authors can declare
it in their JSON schema. Payload paths, the variable picker, and saved
workflow inputs continue to use the property key — labels are
display-only.
## Summary
Closes#20382.
`lint:diff-with-main` can load `.oxlintrc.json` files that reference
`../twenty-oxlint-rules/dist/oxlint-plugin.mjs`, but the diff-lint
targets did not build `twenty-oxlint-rules` first. On fresh clones, that
generated plugin file is missing and oxlint fails before linting.
This PR adds `twenty-oxlint-rules:build` before diff lint for:
- the root `lint:diff-with-main` target default
- the custom `twenty-front:lint:diff-with-main` target
- the custom `twenty-server:lint:diff-with-main` target
It also adds regression coverage for:
- the default diff-lint target dependency
- the custom front/server diff-lint target dependencies
- preserving `twenty-website-new` custom dependencies because it does
not load the Twenty oxlint plugin
## Tests
- `npx vitest run --config
packages/twenty-oxlint-rules/vitest.config.mts
workspace/lint-diff-with-main-targets.spec.ts`
- `node_modules/.bin/nx test twenty-oxlint-rules`
- `node_modules/.bin/nx typecheck twenty-oxlint-rules`
- `node_modules/.bin/nx build twenty-oxlint-rules`
- `node_modules/.bin/nx lint:diff-with-main twenty-server`
- `node_modules/.bin/nx lint:diff-with-main twenty-front`
- `npx oxlint -c packages/twenty-oxlint-rules/.oxlintrc.json
packages/twenty-oxlint-rules/workspace/lint-diff-with-main-targets.spec.ts`
- `git diff --check`
## Notes
- `twenty-website-new:lint:diff-with-main` dependency shape remains
unchanged. Full local execution is blocked by missing local `unzip`,
which the existing `check-lottie-frames` script requires.
## Docs / Changelog
No docs or manual changelog update needed. This fixes Nx task wiring for
an existing documented command.
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
- Pass `--light` to `workspace:seed:dev` in the preview env keepalive
workflow so only the Apple workspace is created
- Avoids `Empty4` being picked as the default workspace at sign-in
(which has no users), making the prefilled `tim@apple.dev` credentials
land on a useful workspace
## Why
`workspace:seed:dev` (no flag) seeds Apple + YCombinator + Empty3 +
Empty4. Preview envs run in single-workspace mode
(`IS_MULTIWORKSPACE_ENABLED=false`), so
[`WorkspaceDomainsService.getDefaultWorkspace`](https://github.com/twentyhq/twenty/blob/main/packages/twenty-server/src/engine/core-modules/domain/workspace-domains/services/workspace-domains.service.ts)
returns the most recently created workspace — Empty4 — which has no
users. Users hitting the preview URL therefore see "Welcome, Empty4."
and can't sign in. Same failure mode #19822 fixed for `twenty-app-dev`.
## Test plan
- [ ] Trigger the `preview-app` label on a PR and confirm the preview
URL signs in to the Apple workspace, not Empty4
- [ ] Confirm the seed step still passes (no `Empty3`/`Empty4`
references break it)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
## Summary
`<input type=\"file\">` inside front-components was silently
non-functional:
- The host-side `serializeEvent` did not read `target.files`, so the
worker received an empty `onChange` detail.
- `SerializedEventData` had no `files` field.
- The `html-input` schema in `AllowedHtmlElements` exposed neither
`accept`, `multiple`, nor `capture` — the worker could not even
configure the picker.
This PR forwards file metadata (`name`, `size`, `type`, `lastModified`)
through the existing serialized event detail and accepts the missing
attributes on the `html-input` remote element. A new Storybook play test
guards the regression by uploading single and multiple files via
`userEvent.upload`.
Reading file contents inside the worker is intentionally out of scope
here and will need a separate host API bridge (the host has the `File`
objects on the real input element; passing bytes through `postMessage`
is a bigger design call).
## Summary
`bore.pub`'s public server has been increasingly unreliable: tunnels
register fine on the runner side (our `Create Tunnel` step always
succeeds), but the bore.pub side later stops accepting inbound traffic,
leaving the preview environment unreachable for the rest of the 5h
keep-alive window with no signal back to the runner. Recent symptom:
`curl http://bore.pub:50422` → `Couldn't connect to server`, while the
corresponding action keeps sleeping.
This PR replaces the `codetalkio/expose-tunnel` action with a direct
invocation of `cloudflared` running an account-less [Cloudflare quick
tunnel](https://developers.cloudflare.com/cloudflare-one/connections/connect-networks/do-more-with-tunnels/trycloudflare/).
The tunnel is served from Cloudflare's edge so reliability is materially
better, and the URL is HTTPS by default (`https://*.trycloudflare.com`),
which also eliminates the mixed-content issues we'd hit when
`SERVER_URL` was `http://bore.pub:port`.
## What changes
- `Create Tunnel` step now:
- Downloads a pinned `cloudflared` binary (`2026.3.0`)
- Starts `cloudflared tunnel --url http://localhost:3000` in the
background, logging to `$RUNNER_TEMP/cloudflared.log`
- Polls the log for `https://<name>.trycloudflare.com` (up to 2
minutes), failing fast if the process exits
- Writes the URL to the `tunnel-url` step output — same name as before,
so no downstream changes needed
- `Cleanup` step kills the `cloudflared` process for hygiene
## What stays the same
- `SERVER_URL` plumbing through `.env` → `docker compose up`
- `tunnel-url` artifact
- `$GITHUB_STEP_SUMMARY` formatting
- PR-comment dispatch (`twentyhq/ci-privileged`)
- 5h keep-alive sleep
## Trade-offs
- Quick tunnels are explicitly labelled by Cloudflare for
"testing/development" use without an SLA. For our preview-env use case
(ephemeral, per-PR) that fits, but if we ever need stable URLs on a
custom domain we'd move to *named* tunnels — same `cloudflared` binary,
plus a free Cloudflare account + delegated domain + a service token
stored as a repo secret. Strictly additive when we want it.
- `cloudflared` is pinned to `2026.3.0` to avoid surprise breakage from
upstream releases. Bumping is a one-line change.
## Testing
**Locally (macOS) — verified end-to-end:**
- `cloudflared tunnel --url http://localhost:18080` against a `python3
-m http.server`
- Regex `https://[a-zA-Z0-9-]+\.trycloudflare\.com` correctly extracts
the URL from the log
- `curl $URL/` returns the upstream server's response (HTTP 200, ~0.5s)
- Process supervision: if `cloudflared` dies mid-wait, the step fails
fast instead of hitting the 2-min timeout
**Validation:**
- `actionlint` passes (the remaining shellcheck warnings are in
pre-existing steps, not my changes)
- `shellcheck` on the new Create Tunnel script: clean
**What's not testable from a PR (and why):**
- The full keep-alive workflow runs on `repository_dispatch`, which
always uses the workflow file from `main`. So the cloudflared logic only
runs against PR contents *after* merge.
- I'll trigger a one-off Ubuntu-runner test of just the install + URL
extraction logic via a throwaway branch (`workflow_dispatch`-only) and
link the run here before this merges.
## Test plan
- [ ] Throwaway run validates: cloudflared installs on `ubuntu-latest`,
prints the URL, regex matches, tunnel is reachable from outside the
runner.
- [ ] After merge, the next PR's preview environment uses
`*.trycloudflare.com` instead of `bore.pub:port`, and the URL stays
reachable for the full 5h window.
- [ ] PR-comment bot still posts the preview URL correctly (link should
now be `https://*.trycloudflare.com`).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
## Problem
In the front-component sandbox, typing in the middle of a pre-filled
`<input>` or `<textarea>` caused the caret to jump to the end on every
keystroke. Characters appeared at the correct position, but editing
mid-string was effectively broken.
Root cause: the remote-DOM bridge round-trips every keystroke through
the
worker. By the time the updated `value` prop arrives back at the host,
React applies it by setting `inputElement.value = X` directly, which
browsers always reset the caret to the end.
Typing at the end was unaffected, which is why this went unnoticed in
search fields and similar append-only inputs.
## Fix
For text-like `<input>` types and `<textarea>`, the `value` prop is now
applied imperatively through a ref callback instead of being passed as a
React controlled prop:
- If the DOM value already matches the incoming prop, the assignment is
skipped entirely.
- If a write is needed and the element is focused, `selectionStart` and
`selectionEnd` are captured before the assignment and restored
afterwards with `setSelectionRange`.
Non-text input types (checkbox, radio, file, color, range) and all other
host elements are unaffected.
## Testing
Drop the repro from the issue into any front-component, click between
two
characters in the pre-filled value, and type — the caret should now stay
at the insertion point.
Fixes#20409
---------
Co-authored-by: Charles Bochet <charles@twenty.com>
## Summary
In the admin panel workspace detail page, the **Upgrade Status > Last
error** row was rendering the raw `errorMessage` string directly. Long
messages (typically full stack traces) overflowed the table cell and
overlapped neighbouring rows, breaking the layout.
The `Last command` row in the same table already uses
`OverflowingTextWithTooltip` (the helper used elsewhere in settings
tables) to clamp long values to a single line and reveal the full text
in a tooltip on hover.
This PR applies the same treatment to the `Last error` row, with
`isTooltipMultiline` so newlines in the stack trace are preserved when
the tooltip opens.
## Test plan
- [ ] Open Admin Panel > a workspace with a failed upgrade and verify
the `Last error` row stays on a single line with an ellipsis
- [ ] Hover the row and verify the full multi-line error is shown in the
tooltip
- [ ] Verify other rows (Last command, Last updated, etc.) and the
workspace info section are unaffected
# Introduction
Encrypt the `connectedAccount` `accessToken` and `refreshToken` using
`APP_SECRET` in order to mitigate potential data leak or `core` table
compromise
## Decrypt
Temporary allow already plain text stored token to be retrieve without
decryption until the slow instance has been passed
Will uncomment the invariant check in a patch when the instance slow has
fully be run
## Standards
- Token are encrypted as quickly as possible
- A token cannot be written in database non encrypted by mistake using a
custom constraint ( `enc:` prefix )
## What's next
We should standardize not managing secret as is in the the services and
layer, they should be encrypted on the flight the earliest and should
never be logged
Will create a dedicated pattern afterwards for `applicationVariables`
secrets too
---------
Co-authored-by: Charles Bochet <charlesBochet@users.noreply.github.com>
## Summary
Adds a new `twenty-claude-skills` workspace package under `packages/`
for Claude skills related to Twenty.
## Changes
- Registers `packages/twenty-claude-skills` in the root Yarn workspace
list.
- Adds package metadata for `twenty-claude-skills`.
- Adds a README documenting the multi-skill layout.
- Adds the `twenty-record-presentation` skill under
`skills/twenty-record-presentation/SKILL.md`.
## Impact
This gives Claude-specific Twenty skills a dedicated package location
while preserving the skill metadata from the provided skill bundle.
## Validation
- Parsed the root `package.json` and
`packages/twenty-claude-skills/package.json` with Node.
- Compared the imported skill content against the source `.skill`
archive; the only difference is a trailing newline at EOF.
# Introduction
Restructure the RelayState and avoid asserting the idp identifier from
this opaque blob
Inferring the id from the secured validated and signed request params
