## I have read the CONTRIBUTING.md file.
YES
## What kind of change does this PR introduce?
Fix (CLI) — `twenty deploy` now detects an expired/invalid API key on
the active remote and offers an interactive re-auth flow (TTY only). In
non-TTY contexts the behavior is unchanged: a clear error and a non-zero
exit.
Fixes#20197
## What is the current behavior?
After a workspace DB reset, key revocation, or workspace deletion, both
`twenty deploy` and (effectively) `twenty dev` fail with:
```
Upload failed: Token has expired.
```
The message is technically correct but gives the user no way forward.
They have to know to mint a new key from **Settings → Developers** and
re-run `twenty remote add --local --api-key <NEW_KEY>`. This came up
while testing PR #20181 and is the same friction on any DB reset, key
revocation, or workspace deletion.
## What is the new behavior?
Two changes, layered:
### (1) Better error message + remediation hint
When the upload returns a 401 or its message matches a token-expired
pattern (`/token has expired|unauthori[sz]ed|invalid api key/i`),
`appDeploy` now prints:
```
Your API key for remote "local" is no longer valid
(the workspace may have been reset, or the key was revoked).
Re-authenticate with:
twenty remote:add --as local --api-key <NEW_KEY>
Generate a new key at: <SERVER_URL>/settings/developers
```
### (2) Interactive re-auth prompt (TTY only)
If the process is attached to a TTY, after the hint is printed the user
is prompted:
```
Re-authenticate now? (Y/n)
```
- **Yes** → re-validate the token (it may have been refreshed
externally), and if still invalid, instruct the user to re-run
`remote:add`. The original `appDeploy` is then retried once.
- **No** → the original `DEPLOY_FAILED` error is surfaced (same code,
better message).
- **Non-TTY (CI, scripts, redirects)** → the prompt is suppressed
entirely. The user gets the hint and a non-zero exit, preserving
scriptable behavior. **No change** to existing CI scripts.
## Acceptance criteria
| Scenario | Before | After |
|---|---|---|
| Happy path deploy | ✅ works | ✅ works (no change) |
| Deploy with expired key (TTY) | generic error, exit 1 | hint + prompt,
retry on Y, error on N |
| Deploy with expired key (CI / no-TTY) | generic error, exit 1 | hint +
exit 1 (no prompt, scriptable) |
| Deploy with unrelated error (e.g. 500) | generic error, exit 1 |
unchanged (no false positive on the matcher) |
## Reproduction
1. Spin up Twenty, mint an API key, run `twenty deploy` — confirm the
happy path.
2. Reset the DB (`core.appToken` cleared) and re-run `twenty deploy` —
confirm the new hint + prompt fire and the retry succeeds.
3. Repeat step 2 in a non-TTY context (e.g. `twenty deploy < /dev/null`
or via `script -qc ''`) — confirm the prompt is suppressed and the
scriptable exit-1 behavior is preserved.
## Implementation notes
- **`FileApi.uploadAppTarball`** now tags 401 responses with an
`isAuthError: true` flag on the failing `ApiResponse`. The existing
`error` string is still populated so callers that don't check the flag
continue to work — **additive, no breaking change**.
- **`FailingApiResponse<TError>`** gained an optional `isAuthError?:
boolean` field. The other `ApiResponse` call sites in the SDK don't need
to set it.
- **`@/cli/utilities/auth/reauth-helper.ts`** is new. It owns:
- `isTokenExpiredMessage(...)` — pure matcher, easy to unit-test, used
as a backstop if a non-401 message still says "expired" (GraphQL returns
200 with errors in some cases).
- `promptForReauthentication(remoteName)` — TTY-gated `inquirer.confirm`
prompt that re-validates the token and either returns
`'reauthenticated'`, `'declined'`, or `'non-interactive'`.
- **`@/cli/operations/deploy.ts`** is the single call site that wires
the helper. The helper is structured so it can be reused from the dev
orchestrator's upload step (a follow-up) without changes.
- **New unit test** at `__tests__/reauth-helper.test.ts` covers the
matcher: positive cases, negative cases, case-insensitivity, and nullish
input.
## Out of scope (per the issue)
- Long-lived dev tokens for `--local` remotes.
- Web-based OAuth login flow for the CLI (the existing
`authenticate(...)` flow in `remote.ts` is fine; the prompt here just
tells the user to re-run it).
## Files changed
```
packages/twenty-sdk/src/cli/operations/deploy.ts | 33 ++++++++
packages/twenty-sdk/src/cli/utilities/api/api-response-type.ts | 1 +
packages/twenty-sdk/src/cli/utilities/api/file-api.ts | 8 +++
packages/twenty-sdk/src/cli/utilities/auth/__tests__/reauth-helper.test.ts | 34 ++++++++++
packages/twenty-sdk/src/cli/utilities/auth/reauth-helper.ts | 61 ++++++++++++++++++
5 files changed, 137 insertions(+)
```
Happy to address feedback and split this into two PRs (hint-only first,
prompt-on-top) if the maintainers prefer a smaller first cut.
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: martmull <martmull@hotmail.fr>
Split out of #21240. Stacked on #21250 (review/merge that first).
`yarn twenty dev --once --dry-run` computes the migration plan and
prints the diff **without applying anything** (no migration, no
app-record update, no SDK generation). Also renders the diff on a normal
`dev --once` sync.
<img width="646" height="179" alt="image"
src="https://github.com/user-attachments/assets/59f3ddcd-2a5b-4b8a-b21a-c659abe16af0"
/>
sdk handle auth of one workspace per session -- but server could be
configured as multi or single -- hence for multi get subdomain -- and
for single the localhost fallback!
also: link includes applicationId so it opens the app detail page
directly (not the list)
## QA
multi workspace flag on -
<img width="2996" height="1712" alt="CleanShot 2026-05-22 at 18 21
31@2x"
src="https://github.com/user-attachments/assets/8499b9f3-b22e-45e2-8b97-4b27fadc3c94"
/>
multi workspace flag off -
<img width="3012" height="1734" alt="CleanShot 2026-05-22 at 18 14
37@2x"
src="https://github.com/user-attachments/assets/3af2f492-5e2d-4a4b-8251-c3343d79ae9e"
/>
## Summary
- Add `WorkspaceMigrationGraphqlApiExceptionInterceptor` to
`MarketplaceResolver` and `ApplicationInstallResolver` so validation
failures during app install return `METADATA_VALIDATION_FAILED` with
structured `extensions.errors` instead of generic
`INTERNAL_SERVER_ERROR`
- Update SDK `installTarballApp()` to pass the full GraphQL error object
(including extensions) through the install flow
- Add `formatInstallValidationErrors` utility to format structured
validation errors for CLI output
- Add integration test verifying structured error responses for invalid
navigation menu items and view fields
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
## Summary
- **Fix `createApplicationRegistration` flow**: The server's
`createApplicationRegistration` mutation returns a `clientSecret`, not
`accessToken`/`refreshToken` directly. The SDK now correctly requests
`clientSecret` and immediately performs an OAuth `client_credentials`
exchange to obtain `appAccessToken` and `appRefreshToken`, then stores
them in config.
- **New `exchangeCredentialsForTokens` helper**: Shared by both `dev`
and `dev --once` flows. Takes `clientId` + `clientSecret`, calls
`/oauth/token` with `client_credentials` grant, and persists the
resulting tokens.
- **Bump `twenty-sdk`, `twenty-client-sdk`, `create-twenty-app` to
`1.22.0-canary.2`**
## Context
The `1.22.0-canary.1` SDK release expected
`createApplicationRegistration` to return `accessToken`/`refreshToken`
directly, but the `v1.22.0` server returns `clientSecret`. This caused
`yarn twenty dev` and `yarn twenty dev --once` to fail with "No
registration found" errors.
## Summary
- **SDK (`dev` & `dev --once`)**: After app registration, the CLI now
obtains an `APPLICATION_ACCESS` token via `client_credentials` grant
using the app's own `clientId`/`clientSecret`, and uses that token for
CoreApiClient schema introspection — instead of the user's
`config.accessToken` which returns the full unscoped schema.
- **Config**: `oauthClientSecret` is now persisted alongside
`oauthClientId` in `~/.twenty/config.json` when creating a new app
registration, so subsequent `dev`/`dev --once` runs can obtain fresh app
tokens without re-registration.
- **CI action**: `spawn-twenty-app-dev-test` now outputs a proper
`API_KEY` JWT (signed with the seeded dev workspace secret) instead of
the previous hardcoded `ACCESS` token — giving consumers a real API key
rather than a user session token.
## Motivation
When developing Twenty apps, `yarn twenty dev` was using the CLI user's
OAuth token for GraphQL schema introspection during CoreApiClient
generation. This token (type `ACCESS`) has no `applicationId` claim, so
the server returns the **full workspace schema** — including all objects
— rather than the scoped schema the app should see at runtime (filtered
by `applicationId`).
This caused a discrepancy: the generated CoreApiClient contained fields
the app couldn't actually query at runtime with its `APPLICATION_ACCESS`
token.
By switching to `client_credentials` grant, the SDK now introspects with
the same token type the app will use in production, ensuring the
generated client accurately reflects the app's runtime capabilities.
## Summary
- **Config as source of truth**: `~/.twenty/config.json` is now the
single source of truth for SDK authentication — env var fallbacks have
been removed from the config resolution chain.
- **Test instance support**: `twenty server start --test` spins up a
dedicated Docker instance on port 2021 with its own config
(`config.test.json`), so integration tests don't interfere with the dev
environment.
- **API key auth for marketplace**: Removed `UserAuthGuard` from
`MarketplaceResolver` so API key tokens (workspace-scoped) can call
`installMarketplaceApp`.
- **CI for example apps**: Added monorepo CI workflows for `hello-world`
and `postcard` example apps to catch regressions.
- **Simplified CI**: All `ci-create-app-e2e` and example app workflows
now use a shared `spawn-twenty-app-dev-test` action (Docker-based)
instead of building the server from source. Consolidated auth env vars
to `TWENTY_API_URL` + `TWENTY_API_KEY`.
- **Template publishing fix**: `create-twenty-app` template now
correctly preserves `.github/` and `.gitignore` through npm publish
(stored without leading dot, renamed after copy).
## Test plan
- [x] CI SDK (lint, typecheck, unit, integration, e2e) — all green
- [x] CI Example App Hello World — green
- [x] CI Example App Postcard — green
- [x] CI Create App E2E minimal — green
- [x] CI Front, CI Server, CI Shared — green
## Summary
- The `createOneApplication` GraphQL mutation was removed from the
server during the application architecture refactor (#18432), but the
SDK CLI (`app:dev`, `app:build --sync`) still called it, causing
failures.
- Simplified the SDK to use `syncApplication` (which now internally
creates the `ApplicationEntity` via `ensureApplicationExists`) instead
of a separate create step.
- On first run (clean install), the orchestrator now runs an initial
sync before initializing the file uploader, so file uploads can proceed
(they require the `ApplicationEntity` to exist).
## Test plan
- [x] Typecheck passes for both `twenty-sdk` and `twenty-server`
- [x] `app:dev` tested locally with existing app (finds app, uploads,
syncs)
- [x] `app:dev` tested locally after `app:uninstall` (creates app via
sync, uploads, syncs)
- [x] SDK unit tests pass (23/26 files, 3 pre-existing failures
unrelated)
Made with [Cursor](https://cursor.com)
## Summary
- **Refactor frontend metadata loading architecture**: Split the
monolithic `EagerMetadataLoadEffect` into focused provider effects
(`UserMetadataProviderEffect`, `ObjectMetadataProviderEffect`,
`ViewMetadataProviderEffect`) orchestrated by `MetadataProviderEffects`.
Replaced `UserProvider` + `ObjectMetadataItemsProvider` with a single
`MetadataGater` that gates rendering on `isAppMetadataReadyState`. The
metadata store now validates view-object consistency before promoting
views, and `updateDraft` skips no-op updates via deep equality checks.
- **SDK CLI improvements**: Added `app:typecheck` command, improved
error handling in API sync (extracts GraphQL error messages), added
`serializeError` utility for human-readable error output, added `error`
file status to dev mode orchestrator with UI support, and fixed
ClickHouse migration/seed commands to use `transpile-only`.
## Add API client generation to SDK dev mode and refactor orchestrator
into step-based pipeline
### Why
The SDK dev mode lacked typed API client generation, forcing developers
to work without auto-generated GraphQL types when building applications.
Additionally, the orchestrator was a monolithic class that mixed watcher
management, token handling, and sync logic — making it difficult to
extend with new steps like client generation.
### How
- **Refactored the orchestrator** into a step-based pipeline with
dedicated classes: `CheckServer`, `EnsureValidTokens`,
`ResolveApplication`, `BuildManifest`, `UploadFiles`,
`GenerateApiClient`, `SyncApplication`, and `StartWatchers`. Each step
has typed input/output/status, managed by a new `OrchestratorState`
class.
- **Added `GenerateApiClientOrchestratorStep`** that detects
object/field schema changes and regenerates a typed GraphQL client (via
`@genql/cli`) into `node_modules/twenty-sdk/generated` for seamless
imports.
- **Replaced `checkApplicationExist`** with `findOneApplication` on both
server resolver and SDK API service, returning the entity data instead
of a boolean.
- **Added application token pair mutations**
(`generateApplicationToken`, `renewApplicationToken`) to the API
service, with the server now returning `ApplicationTokenPairDTO`
containing both access and refresh tokens.
- **Restructured the dev UI** into `dev/ui/components/` with dedicated
panel, section, and event log components.
- **Simplified `AppDevCommand`** from ~180 lines of watcher management
down to ~40 lines that delegate entirely to the orchestrator.
# Improve cross-entity duplicate detection in manifest validation
- Refactored findDuplicates to receive the full manifest, enabling
cross-entity duplicate checks
-ObjectExtensionEntityBuilder now validates that extension field IDs
don't conflict with object field IDs
- Renamed DuplicateId type to EntityIdWithLocation for clarity
- Updated all entity builders to use the new signature
Heavy Refactoring of the watcher, sorry about this one, I'll keep
iterating on it.
In a nutshell:
- app-dev.ts is maintaining 3 watchers in parallel: manifest, function
and frontComponent
Mani bharadwaj