## Summary
- [eas-cli] Upgrade `@expo/apple-utils` from `2.1.19` to `2.1.22`.
- Refresh the Yarn lockfile entry for the new package tarball.
- Add a changelog entry under the unreleased `main` section.
## Why
This picks up the latest published Apple API helpers, including recent App Store Connect pagination, SIWA, and review-related fixes from `@expo/apple-utils`.
## Validation
CI should pass.
## Summary
- fix main typecheck after `Env` was widened to allow undefined values
- align `getManagedTvBuildSettings` with the shared `Env` type instead of requiring `Record<string, string>`
## Test plan
CI should pass.
## Summary
Widens the shared build `Env` type to allow `undefined` values, matching Node/process env semantics where undefined entries can represent absent values.
## Details
- Changes `Env` to `Record<string, string | undefined>` in `@expo/eas-build-job`.
- Allows `ctx.reportError` extras to contain undefined values for diagnostic metadata.
- Updates strict string boundaries to explicitly require values where needed, such as GraphQL IDs and static context API URLs.
- Filters undefined values out of `displayRuntimeInfo` string-only display maps.
## Validation
CI should pass.
## Summary
- Add an opt-in `enable_accessibility_settings` boolean input to `eas/start_ios_simulator`.
- When enabled, call `IosSimulatorUtils.enableAccessibilitySettingsAsync` inline immediately before each `IosSimulatorUtils.startAsync` call, including cloned simulators.
- Reuse the existing `xcrun simctl list devices --json --no-escape-slashes available` wrapper so the `dataPath` from simctl JSON is used directly, then do an exact UDID/name equality check in JS. This avoids `simctl` search-term substring matches like `iPhone 17` matching `iPhone 17 Pro`.
- Treat invalid requested simulator state as `UserError`; wrap command/filesystem setup failures as `SystemError`.
## Why
These accessibility preferences help Argent and other AX-tree based tools inspect Simulator state more reliably. They can influence app/runtime accessibility behavior, so the setting is explicit and default-off rather than applied to every simulator startup.
## Maestro compatibility
`enable_accessibility_settings` defaults to `false`, so existing Maestro test runs and other `eas/start_ios_simulator` users keep the previous startup behavior unless they opt in. When enabled, the setup only writes preferences for an exact, available, shutdown simulator before `simctl bootstatus -b`. If the requested simulator is not available or not shutdown, the step fails with `UserError`; if the setup command/filesystem work fails, it fails with `SystemError`.
## Notes
- `plutil` is a standard macOS utility (`/usr/bin/plutil` on the local validation machine), so this does not add an external tool installation requirement.
## Validation
CI should pass.
# Why
Non interactive builds were attempting to incorrectly verify tvOS provisioning profiles with settings for the iOS platform.
# How
Recognize when EXPO_TV env variable is set, and in that case, set the target platform to tvOS.
Fixes#3905
# Test Plan
- Tested with a real SDK 56 TV project
- New unit tests added
## Why
The Expo dashboard route for device run sessions moved to the simulator sessions surface, so EAS CLI should send users to that page and use the simulator-session terminology in user-facing output.
## How
- Add a dashboard URL helper that points device run sessions at `/simulator-sessions/{id}`.
- Update `eas simulator:{start,get,list}` output and JSON to expose `deviceRunSessionUrl` using the simulator sessions route.
- Update user-facing simulator command copy from "device run session" to "simulator session".
## Test Plan
CI should pass.
## Summary
Updates the codemention GitHub Action pin from `v1.4.0` to a reviewed post-`v1.5.2` upstream commit:
- `tobyhs/codemention@ebac5877393ab8693fd73b42708c8552043e9a3b`
- keeps the action pinned to an immutable commit SHA because this workflow runs on `pull_request_target` with `pull-requests: write`
- avoids pinning to moving `main` or a mutable tag
## Security review
I reviewed upstream `tobyhs/codemention` changes from `v1.4.0` through `v1.5.2` and found that `v1.5.2` introduced `[email protected]`, which currently has critical/high production dependency advisories. I therefore did not pin to `v1.5.2`.
The selected post-release commit includes follow-up dependency fixes on upstream `main`, including `[email protected]`, `[email protected]`, and `[email protected].0`. `npm audit --omit=dev` on that upstream commit reported 0 vulnerabilities at review time.
## Validation
- `yarn fmt:check .github/workflows/codemention.yml`
- YAML parse check
- `git diff --check`
* [eas-cli] Add tiered build-credit usage warnings (approaching/at/over)
Lowers the warning threshold from 85% to 80% to match email
notifications. Adds two new tiers beyond the existing "approaching":
one when the user has reached their included credits, and one when
they've exceeded them on a paid plan. Free plans collapse the "over"
case into the "at" message since Free is blocked at the limit.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* [eas-cli] Add CHANGELOG entry for tiered usage warnings (#3882)
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Sarah Lane <[email protected]>
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
Syncs the generated app `AGENTS.md` and `CLAUDE.md` templates from `expo/llm-configs` during EAS CLI `prepack`, before the package is rebuilt for publishing. Keeps the synced templates bundled in EAS CLI so `eas project:new` still works offline and does not add a runtime `@expo/llm-configs` dependency.
Verified with `yarn workspace eas-cli sync-agent-templates`, byte-for-byte comparison against the upstream raw files, `yarn workspace eas-cli test src/commandUtils/new/__tests__/projectFiles.test.ts --runInBand`, `yarn workspace eas-cli typecheck-for-build`, and `yarn workspace eas-cli build`. Previously also ran a local `eas project:new` smoke test and confirmed generated agent files matched the shared templates.
`eas update --channel <name>` was failing with `Channel has no branches associated with it` when the channel existed but had no branch mapped.
This trips up fresh projects: `eas build` auto-creates the channel without a branch, so by the time you run `eas update` the channel already exists. The CLI only ever auto-created a branch when the channel was *missing* (`ChannelNotFoundError`), so the existing-but-empty channel fell through to the error instead.
Now the empty-channel case does the same thing as the missing-channel case: create a branch with the channel's name and map it.
## Test plan
- Added a unit test for the empty-channel case
- Ran `eas update --channel <empty channel>` against a real project: the old CLI fails with the error above, this build creates the branch and publishes the update
This was needed for running in development where `sudo` is NOT passwordless and waits for password input.
## Summary
- skip `sudo xcode-select` for remote session startup when `ENVIRONMENT=development`
- centralize Xcode developer directory selection for agent-device, Argent, and serve-sim remote sessions
- run `sudo xcode-select` with stdin ignored and stdout/stderr piped to the step logger outside development
## Test plan
- yarn workspace @expo/build-tools test src/steps/utils/__tests__/remoteDeviceRunSession.test.ts
- yarn workspace @expo/build-tools typecheck
Since https://github.com/software-mansion/argent/pull/347 Argent exposes a `GET /artifacts/` endpoint which returns a `size` for every artifact. This `size` is then being used to create a `www` upload session (which validates and includes the `size` as `Content-Length` of the signed upload URL).
The `size` is being calculated in Argent as `fs.stat()` which works great for single files, but for directories it returns 96 bytes. The actual archiving of a given directory (producing an artifact) is only happening when downloading the artifact with `GET /artifacts/:id`.
This pull request changes the artifacts behavior around Argent from:
- list artifacts
- trust `size`
- pipe `GET /artifacts/:id` directly to signed upload URL
to:
- list artifacts
- download from `GET /artifacts/:id` to temporary directory
- check the `size` of the actually downloaded file
- pipe it to signed upload URL.
## Test
You can see which traces have been created _with_ the fix and which _without_ it.
<img width="445" height="354" alt="Zrzut ekranu 2026-06-19 o 13 03 01" src="https://github.com/user-attachments/assets/fddecddf-419c-46a1-8876-599bc00ea8ce" />
## Summary
- bump the minimum supported Argent remote-session version to the latest published `@swmansion/argent` version (`0.12.0`)
- enable Argent's artifact inventory endpoint before starting the tool-server
- poll the local Argent tool-server every 5 seconds for new artifacts and stream each new artifact to the EAS device-run-session artifact upload backend
- add focused tests for version validation, artifact inventory listing, and signed upload streaming
## Testing
CI should pass.
# Why
New command for retrieving and analyzing events in a session.
```
USAGE
$ eas observe:session SESSIONID [--project-id <value>] [--json] [--non-interactive]
ARGUMENTS
SESSIONID Session ID to inspect
FLAGS
--json Enable JSON output, non-JSON messages will be printed to stderr. Implies --non-interactive.
--non-interactive Run the command in non-interactive mode.
--project-id=<value> EAS project ID (defaults to the project ID of the current directory)
DESCRIPTION
display the timeline of metric and log events for a specific session
```
# How
Query both metrics and logs, filtered by the session ID passed in.
Example results:
```
# Results for an app startup session (main session)
easd observe:session 2bb44b16-7650-4efd-8ec8-a1bd75ee81dc
App version: 1.1.0 (31)
Device: Pixel 7 · Android 13
First seen: May 11, 2026, 04:12:07.662 AM
Last seen: May 11, 2026, 04:13:04.886 AM
Offset Type Name Value Properties Severity
------ ------ ------------ ----- ----------------- --------
0.00s metric Bundle Load 0.77s - -
0.05s log app.open - cold_start=false INFO
0.15s metric Startup TTR 1.38s - -
0.45s metric Startup TTI 2.16s - -
0.55s metric Warm Launch 1.29s - -
0.95s log screen.view - screen=profile INFO
17.16s log habit.logged - habit_id=meditate INFO
streak_days=12
56.59s log screen.view - screen=streak INFO
57.22s log screen.view - screen=paywall INFO
# Results for a navigation session with routes
$ eas observe:session 7e57f6cd-f8e9-4fe0-b6a5-934856d61663
App version: 1.1.0 (31)
Device: iPhone16,1 · iOS 17.5
First seen: May 6, 2026, 06:54:35.820 AM
Last seen: May 6, 2026, 06:56:11.753 AM
Offset Type Name Value / Severity
------ ------ ----------------------------- ----------------
0.00s metric Nav Cold TTR · /task/[taskId] 0.51s
0.08s metric Nav TTI · /task/[taskId] 0.53s
4.78s metric Nav Warm TTR · /stats 0.55s
4.84s metric Nav TTI · /stats 0.57s
33.66s metric Nav Warm TTR · /stats 0.53s
33.72s metric Nav TTI · /stats 0.47s
53.96s metric Nav Warm TTR · /stats 0.24s
54.02s metric Nav TTI · /stats 0.82s
78.61s metric Nav Warm TTR · /new 0.21s
78.67s metric Nav TTI · /new 0.27s
95.87s metric Nav Warm TTR · /new 0.08s
95.93s metric Nav TTI · /new 0.20s
```
# Test Plan
- Tested against a project in staging.
- New unit tests added.
Fixes Argent remote simulator sessions getting stuck immediately after starting the Argent tool-server.
The latest Argent support change switched the startup path to `bunx @swmansion/argent server start ...` and awaited that process. When the server process stays in the foreground, the build step never advances to read the tool-server state file, start the ngrok tunnel, or report remote config back to the API.
This restores detached process launching for the Argent tool-server while keeping the new server/auth flow. If the state file is not created in time, the error now includes captured Argent process output to aid diagnosis.
Failing job run referenced while debugging: https://expo.dev/accounts/sjchmiela/projects/test-production/job-runs/019eda48-6467-789d-8c5e-a7ed5f4ec89d
## Why
Connect an Expo app to PostHog in one command. It provisions the PostHog project, asks which features you want, installs the SDK, and writes the env vars — so analytics / session replay / error tracking work after a build.
## How
Fast-forward order — provision first, then ask what to wire up:
1. **Provision** — reuses your account's PostHog org + this app's project, or creates them. Only a brand-new connection asks a question: **region** (US / EU), which sets data residency and can't be changed later.
2. **Pick features** — a multiselect, all preselected: Analytics, Session replay, Error tracking (source maps, _requires a personal API key_).
3. **Error tracking, if selected** — prompts you to paste a PostHog personal API key (we can't mint it — you create it in PostHog and paste it in).
It then:
- **Installs** `posthog-react-native` (+ `expo-file-system`, `expo-application`, `expo-device`, `expo-localization`; plus `posthog-react-native-session-replay` when replay is on) via `npx expo install`
- **Adds** the `posthog-react-native/expo` config plugin (prints manual steps if your app config is dynamic)
- **Writes env vars** to `.env.local` (local dev) and EAS (Production / Preview / Development):
- Analytics → `EXPO_PUBLIC_POSTHOG_API_KEY`, `EXPO_PUBLIC_POSTHOG_HOST` (public)
- Error tracking → `POSTHOG_CLI_API_KEY` (sensitive), `POSTHOG_CLI_PROJECT_ID`, `POSTHOG_CLI_HOST` (public)
- Source-map uploads on builds are covered by our guide; the command just sets the `POSTHOG_CLI_*` vars.
This PR also includes the PostHog GraphQL layer the command builds on — `PostHogQuery` / `PostHogMutation`, the connection/project types, and `commandUtils/posthog` (dashboard URL).
**Flags:** `--region`, `--session-replay` / `--no-session-replay`, `--error-tracking` / `--no-error-tracking`, `--posthog-cli-api-key`, `--overwrite`, plus `--json` / `--non-interactive`. Agent-ready: error tracking auto-skips without a key (or pass `--posthog-cli-api-key`); `--error-tracking` without a key fails before provisioning; `--region` is required non-interactively.
## Test Plan
- CI — 34 unit tests, full coverage of `connect.ts`
- Ran end-to-end against production: provisioned a PostHog project, selected all features, and confirmed the env vars land in `.env.local` + EAS and the SDK packages / config plugin are added
<img width="2048" height="2718" alt="carbon (1)" src="https://github.com/user-attachments/assets/073a5c41-7f07-48d6-b5b7-b1de3cff89fa" />
- v0.11 of Argent added official support for "remote mode", i.e. we don't need to start the MCP server for _it_ to start the tools-server, we can just start the tools-server.
- just to be sure I've added a min Argent version check.
- new Argent versions also guard the tools server by authorization, so added `authToken` support
- new Argent versions also check `Host` header so made ngrok rewrite it to `localhost:...` to fix error below
<img width="868" height="538" alt="Zrzut ekranu 2026-06-18 o 10 59 54" src="https://github.com/user-attachments/assets/6b3689a4-9359-4f8f-bba7-4e98e97ff0cd" />
* [eas-cli] Fix update:embedded topic description
Without an explicit topic entry, oclif derived the "update:embedded"
group description from a subcommand, so the topic listing showed
"delete an embedded update registered with EAS Update" — confusingly
describing only the delete subcommand. Add an explicit topic description.
* [eas-cli] Add changelog entry for update:embedded topic fix
* [eas-cli] add support for non-interactive update:rollback
* [eas-cli] Add CHANGELOG entry for non-interactive update:rollback
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>
* [worker] Resolve runtime cache settings from job env and fallback on npm cache failures
* [worker] Resolve runtime cache settings from job env and fallback on npm cache failures
* no longer url fallback
* use env over runtime stngs
* report log to sentry only
* proper subclass errs
* proper subclass errs
* proper subclass errs
* runtime setting override, rem set registry cmd
* continue support DISABLE flags, in line err tracker
## What changed
- Removed `gcsSignedUploadUrlForApplicationArchive` and `gcsSignedUploadUrlForBuildArtifacts` from worker runtime config and runtime config typing.
- Removed launcher signed-upload fallback behavior for managed application archive and build artifacts uploads.
- Classified managed artifact upload failures as `SystemError`s with internal tracking codes while keeping the public error code as `SERVER_ERROR`.
- Added retry options to worker upload API `turtleFetch` calls for upload-session creation and artifact metadata saving.
## Why
These signed upload URL fields are no longer expected, so keeping fallback paths made the worker handle an obsolete runtime contract. Upload failures in these managed paths are infrastructure/API/storage failures rather than user-actionable project issues.
## Validation
- `yarn oxfmt packages/worker/src/config.ts packages/worker/src/upload.ts packages/worker/src/external/turtle.ts packages/worker/src/__unit__/upload.test.ts`
- `yarn workspace @expo/worker test:unit src/__unit__/upload.test.ts --runInBand`
- `yarn workspace @expo/worker typecheck`
## Summary
Drops the temporary legacy cache URL fallback path now that the worker env-var PR is deployed.
This removes the fallback URL state from runtime settings, removes the old cache URL fields from worker runtime config parsing/types, and removes the now-obsolete config fallback test.
## Impact
Cache URL sourcing now relies on the runtime-settings cache gates plus the new worker cache URL environment variables only. Public build env names remain unchanged.
## Validation
CI should pass. Once deployed to staging I'll confirm the env vars and cache config is there.
# Why
Many projects already configure an App Store Connect API key in EAS credentials for submissions, but non-interactive iOS App Store and Enterprise builds (`eas build --non-interactive`) could not use that key when validating or repairing provisioning profiles on Apple servers. Instead, CI had to provide `EXPO_ASC_*` environment variables separately, even when a usable key was already stored on EAS.
This makes non-interactive credential setup more convenient by reusing the EAS-stored submission ASC API key for App Store and Enterprise profile validation and repair.
# How
- Added `tryAuthenticateAppStoreWithEasAscApiKeyAsync` in `AscApiKeyUtils.ts` — a best-effort, non-throwing helper that authenticates `ctx.appStore` in non-interactive mode using:
1. `EXPO_ASC_API_KEY_PATH` / `EXPO_ASC_KEY_ID` / `EXPO_ASC_ISSUER_ID` environment variables, or
2. The App Store Connect API key configured for submissions on the app in EAS credentials.
- In `SetUpProvisioningProfile.runAsync`, call this helper before Apple-side validation when running non-interactively without an existing auth context.
- Use actual authentication state (`ctx.appStore.authCtx`) instead of startup `defaultAuthenticationMode` when deciding whether Apple authentication is available in `SetUpProvisioningProfile`, `CreateProvisioningProfile`, and `ConfigureProvisioningProfile`.
- In non-interactive mode, treat unexpected Apple validation errors as best-effort: log a warning and continue rather than failing the build.
- Extracted shared `resolveAscApiKeyForAppCredentialsAsync` (previously duplicated in `SetUpAdhocProvisioningProfile`) for reuse across credential flows.
- Derive the `AppleTeamType` for profile operations from the distribution type (`IN_HOUSE` for enterprise, `COMPANY_OR_ORGANIZATION` otherwise), with `EXPO_APPLE_TEAM_TYPE` override.
- Updated error messages to mention both `EXPO_ASC_*` environment variables and EAS-stored ASC API keys.
Out of scope: ad-hoc provisioning profile refresh (opt-in, handled separately via `--refresh-ad-hoc-provisioning-profile`).
# Test Plan
- Added unit tests for auto-authentication, team type resolution, best-effort validation skip, and missing-auth paths in `SetUpProvisioningProfile-test.ts`.
- Tested manually:
- Run an interactive iOS production build and create a provisioning profile
- Confirm non-interactive build works with that profile
- Revoke the provisioning profile in the Apple Developer portal
- Confirm non-interactive build succeeds and the profile is repaired when the submission ASC API key is configured in app credentials (`easd build --platform ios --profile production --non-interactive`)
* [eas-cli] Regenerate GraphQL schema
* [eas-cli] Add --max-duration-minutes flag to eas simulator:start
Co-Authored-By: Claude Opus 4.7 <[email protected]>
* [eas-cli] Only reject negative --max-duration-minutes in simulator:start
Leave the upper-bound check to the server instead of capping it client-side.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 <[email protected]>
When launching serve-sim, the worker requests short-lived TURN ICE servers from www's turtle-job-runs route (same robot-token pattern as the project clone URL) and forwards them as --stun-url/--turn-url/--turn-username/--turn-credential. Covers all three serve-sim launchers (serve_sim, agent_device, argent).
Best-effort: on any failure serve-sim falls back to its built-in P2P/STUN behavior, so a TURN outage never blocks a session.
Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>