# Why
When session ID is omitted and the `observe:session` command is running interactively, walk the user through picking an event whose session they want to inspect. Non-interactive without a session ID still errors as before.
The goal is to support a flow similar to that already available on our website, where a user can select an event from a displayed list, and navigate to the session timeline that contains that event.
# How
Add new flags `--event-name` and `--sort-order`. If these are not provided, user will be prompted first to select from a list of event names (including custom events, if applicable), then select a sort order. The command also now takes the same time period flags as the other commands (`--days`, `--start`, `--end`).
Once the user has made the above selections, the user can choose from the list of events shown (up to 25), and the session for that event will be shown.
# Test Plan
- Tested manually with an SDK 57 project
- New unit tests added
* [eas-cli] Show accurate build overage count in pre-build usage warning
The overage count shown to paid users ("N builds beyond your included
credits") summed the value of every overageMetrics row, including rows
without build metadata (e.g. rollup/summary rows), which could inflate the
number. Filter to genuine per-worker build overage rows (those carrying
billingResourceClass + platform metadata) before summing.
Fetch that metadata on overageMetrics in the warning query so the rows can
be distinguished.
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
* [eas-cli] Fix changelog PR reference for build overage warning entry
Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
---------
Co-authored-by: Sarah Lane <[email protected]>
Co-authored-by: Claude Opus 4.8 (1M context) <[email protected]>
I found it sometimes isn't printed.
---
## Summary
- Stop waiting for `serve-sim` to emit a `Stream:` URL before reporting startup success.
- Upload only `previewUrl` in the serve-sim remote config payload.
- Remove `streamUrl` from the simulator session query selection and generated query result type.
- Simplify the serve-sim URL parser to match only the tunnel URL.
## Why
The stream URL is no longer part of the expected serve-sim output path. Keeping it in the startup contract made the build step wait for and upload a value that is not needed by the CLI display flow.
## Test plan
- `yarn --cwd packages/build-tools typecheck`
- `yarn --cwd packages/eas-cli typecheck`
- `yarn --cwd packages/build-tools jest-unit src/steps/utils/__tests__/remoteDeviceRunSession.test.ts`
- `yarn --cwd packages/eas-cli test src/commands/simulator/__tests__/get.test.ts src/commands/simulator/__tests__/start.test.ts --runInBand`
- `git diff --check`
## Summary
- include a shortened simulator UDID and first-frame timestamp in each screen recording artifact name
- format the timestamp as a readable, stable UTC value
- add `__eas_type: "screen-recording"` metadata for typed artifact detection
- cover recordings from a rebooted simulator and simultaneous same-model simulators
Artifact names now use this format:
```
iPhone 16 screen recording (01234567-…, started at Jul 10, 2026, 10:00:00.000 UTC)
```
## Why
A simulator can reboot during one device run session, producing multiple recordings with the same device name and UDID. A session can also contain two same-model simulators whose recordings start at precisely the same time. Device name alone, or device name plus timestamp alone, therefore does not reliably distinguish every recording.
The first eight UDID characters distinguish same-model simulators without putting the full UUID in the user-facing name. The millisecond first-frame timestamp distinguishes subsequent recordings after a reboot. The full UDID and canonical ISO timestamp remain in artifact metadata for exact identity, parsing, and ordering.
The new `__eas_type` marker provides a descriptive discriminator for screen recordings. The existing `__eas_screen_recording: "1"` marker remains temporarily so currently deployed consumers continue recognizing these artifacts during migration.
The artifact metadata also provides `udid`, `deviceName`, `runtimeDisplayName`, `firstFrameAt`, and encoded `width` and `height`.
## Validation
- focused screen recording upload tests covering rebooted and simultaneous same-model simulators
- build-tools TypeScript typecheck
## Summary
- set `NODE_OPTIONS=--max-old-space-size=3072` for build child processes when the worker has at least 6 GiB of constrained or physical memory and its effective V8 heap limit is below 3 GiB
- preserve existing worker `NODE_OPTIONS`; project-provided environment values retain their later override position
- leave workers with less memory or an existing heap limit of at least 3 GiB unchanged
## Why
V8 defaults to a roughly 2 GiB heap on our smaller build workers. Dependency installation and Expo export can exhaust that heap even though the VM still has enough memory. Raising the ceiling to 3 GiB gives those child processes additional headroom without preallocating memory or changing larger workers that already receive a higher V8 default.
## Validation
- `yarn build`
- `yarn workspace @expo/worker test:unit --runInBand src/__unit__/env.test.ts`
- `yarn workspace @expo/worker typecheck`
- focused Oxlint and Oxfmt checks
# Why
Supersedes #3968.
Dependabot alert 354 requires `lodash` and `lodash-es` 4.18.1. The original PR upgraded the worker's direct dependency and enforced the patched transitive version with a root resolution. That resolution is no longer needed: this PR upgrades the EAS-owned declarations and the GraphQL Code Generator packages that brought in the remaining restrictive `lodash ~4.17.0` dependency.
# How
- Upgrade direct Lodash declarations in build-tools, local-build-plugin, template-file, and worker to 4.18.1.
- Upgrade the scripts project's `lodash-es` declaration and its independent lockfile to 4.18.1.
- Upgrade the complete GraphQL Code Generator family used by EAS CLI:
- `@graphql-codegen/cli` 5.0.0 → 6.3.1
- `@graphql-codegen/introspection` 4.0.0 → 5.0.2
- `@graphql-codegen/typescript` 4.0.1 → 5.0.10
- `@graphql-codegen/typescript-operations` 4.0.1 → 5.1.0
- Regenerate the checked-in GraphQL TypeScript output.
- Keep the root free of `resolutions` and regenerate both affected lockfiles.
These are the newest aligned Codegen releases before `typescript-operations` v6. That v6 upgrade changes generated operation type semantics (including nullable response-field presence) and causes broad fixture/type failures across EAS CLI, so it should be handled as a separate GraphQL contract migration rather than bundled into this security dependency update. The selected packages use `@graphql-codegen/plugin-helpers` 6.3.x, which no longer depends on Lodash.
The root dependency graph now reports only `[email protected].1`, selected through ordinary declared dependency ranges. The separate scripts project reports only `[email protected].1`. There is no resolution or override.
# Test Plan
- `yarn install`
- `yarn why lodash` — only `[email protected].1`; no Codegen consumer
- `cd scripts && yarn install && yarn why lodash-es` — only `[email protected].1`
- `yarn build` — all 13 projects passed
- `yarn workspace eas-cli verify-graphql-code` — generated output compiles against the current live schema
- `cd scripts && yarn typecheck`
- `yarn lint-changelog`
- template-file tests — 7 passed
- worker tests — 51 passed
- `yarn fmt`
## Summary
- increase the graceful `record-sim` shutdown timeout from 30 to 70 seconds
- record readable finalization milestones after a termination signal
- include the recorder's recent stdout/stderr when build-tools must force-stop it
## Why
A production device run session finalized two simulator recordings concurrently. One recording completed normally, while the other was force-killed after exactly 30 seconds and was therefore unable to write its manifest or upload its MP4.
The parent timeout was shorter than the recorder's own valid finalization window: `record-sim` can spend up to 5 seconds appending its final frame and up to 60 seconds waiting for `AVAssetWriter.finishWriting()`. Killing it after 30 seconds could therefore interrupt legitimate finalization.
The existing logs also could not distinguish between a process that failed to observe `SIGINT` and one that observed it but stalled later. The new milestones identify whether the recorder:
1. observed the termination signal,
2. stopped screen capture and began saving the video,
3. saved the video and began writing recording details.
These messages stay buffered during successful recordings. They are added to the user-visible forced-stop warning only when finalization exceeds 70 seconds, keeping normal job logs concise while making failures diagnosable.
## Validation
- built the release `record-sim` executable
- ran the build-tools TypeScript typecheck
- ran all 88 build-tools unit test suites (831 tests)
## Summary
- remove the temporary Datadog log that compared old and new updateEnv merge-order results
- keep the new updateEnv merge order behavior unchanged
- remove the log-specific test assertions
## Test plan
- yarn fmt packages/build-tools/src/context.ts packages/build-tools/src/__tests__/context.test.ts
- yarn --cwd packages/build-tools jest-unit src/__tests__/context.test.ts
- yarn --cwd packages/build-tools typecheck
This is intended to be merged after we confirm the migration looks good in Datadog.
# Why
Supersedes #3967.
Dependabot alert 372 requires `@xmldom/xmldom` 0.8.13. The original PR used a root resolution to force 0.8.13 through parents that declared the incompatible `~0.7.7` range. This replacement upgrades those parent dependencies instead, so xmldom can be updated without overriding any package's declared contract.
# How
- Upgrade EAS CLI's direct `@expo/plist` dependency from 0.2.0 to 0.3.5.
- Upgrade build-tools' `@expo/plist` range from `^0.2.0` to `^0.3.5`.
- Upgrade EAS CLI's `@expo/prebuild-config` from 8.0.17 to 9.0.12, moving its config-plugin chain to a version that uses the newer plist package.
- Upgrade the transitive lockfile entry to `@xmldom/xmldom` 0.8.13.
After these changes, every installed plist consumer declares `@xmldom/xmldom ^0.8.8`, and `yarn why @xmldom/xmldom` reports a single natural 0.8.13 installation. No root resolution is added.
The relevant `getPrebuildConfigAsync` iOS path used by EAS CLI for entitlement inspection is unchanged across the prebuild-config upgrade; the direct implementation change is confined to passing `projectRoot` to the Android plugin path.
# Test Plan
- `yarn install`
- `yarn why @xmldom/xmldom`
- `yarn build` — all 13 projects passed with xmldom 0.8.13 installed
- Four build-tools plist/configuration suites — 19 tests and 14 snapshots passed
- EAS CLI `UpdatesModule` iOS suite — 2 tests passed
- `yarn fmt`
- `yarn fmt:check`
- `yarn lint` — 0 errors; 13 pre-existing warnings also present on `main`
## Why
The image URL test only needs to serve a single PNG over HTTP. Using Express for that adds an otherwise-unused framework and its dependency tree to EAS CLI test dependencies.
This supersedes #3969. That PR upgrades Express to address the vulnerable transitive `path-to-regexp` version, but after this change EAS CLI no longer depends on Express, so the affected dependency path is removed instead of upgraded.
## What changed
- replace the Express static server with Node's built-in `http` server
- bind to port `0` so the operating system selects an available port
- remove `express` and `@types/express` from `packages/eas-cli`
- regenerate the lockfile, removing the associated Express 4 dependency tree
## Impact
This only changes test infrastructure. The test continues to exercise the real remote-URL fetch and PNG transparency validation path. There is no EAS CLI runtime behavior change.
## Validation
- `yarn workspace eas-cli test src/utils/__tests__/image-test.ts --runInBand`
- `yarn workspace eas-cli typecheck`
- formatting check
- `git diff --check`
* [eas-cli] Add `eas browse` command
Add `eas browse [PAGE]` to open the project page on expo.dev, with an
optional whitelisted page argument (e.g. `build`, `hosting`, `cicd`) for
project subpages and a `--no-browser` flag to print the URL instead.
`eas open` becomes a thin alias that opens the project dashboard.
* [eas-cli] Update changelog PR link
* [eas-cli] Add `observe` page to `eas browse`
* [eas-cli] Support --json and --non-interactive in eas browse, hide eas open
* Fix CHANGELOG
---------
Co-authored-by: Douglas Lowder <[email protected]>
* [build-tools] Add eas/posthog_capture_event workflow function
* [build-tools] address PostHog client review nits
* [build-tools] report which PostHog credentials are missing
* [build-tools] add poll helpers, credentials-error factory, query abort signal
* [build-tools] move credential catalog and error into PosthogUtils
* [build-tools] accept an abort signal on all PostHog client requests
* [build-tools] Add eas/posthog_flag_rollout workflow function
* [build-tools] use PosthogUtils namespace in flag rollout
* [build-tools] name missing credentials in flag rollout
* [build-tools] use credentials-error factory in flag rollout
* [build-tools] use PosthogUtils.missingCredentialsError in flag rollout
* [build-tools] thread abort signal through flag rollout requests
* [build-tools] Add eas/posthog_wait_for_metric workflow function
* [build-tools] respect abort signal while waiting for metric
* [build-tools] name missing credentials when waiting for metric
* [build-tools] share poll helpers, harden operator guard, thread signal (metric)
* [build-tools] use PosthogUtils.missingCredentialsError when waiting for metric
* [build-tools] Add eas/posthog_upload_sourcemaps workflow function
* [build-tools] use PosthogUtils namespace in sourcemap upload
* [build-tools] name missing credentials in sourcemap upload
* [build-tools] use credentials-error factory in sourcemap upload
* [build-tools] use PosthogUtils.missingCredentialsError in sourcemap upload
* [build-tools] pass abort signal to the sourcemap upload subprocess
* [build-tools] Add eas/posthog_annotation workflow function
* [build-tools] use PosthogUtils namespace in annotation
* [build-tools] name missing credentials in annotation
* [build-tools] use credentials-error factory in annotation
* [build-tools] use PosthogUtils.missingCredentialsError in annotation
* [build-tools] thread abort signal through annotation request
* [build-tools] Add eas/posthog_wait_for_query workflow function
* [build-tools] respect abort signal while waiting for query
* [build-tools] name missing credentials when waiting for query
* [build-tools] share poll helpers and thread signal (query)
* [build-tools] use PosthogUtils.missingCredentialsError when waiting for query
* [build-tools] Add eas/posthog_capture_event workflow function
* [build-tools] address PostHog client review nits
* [build-tools] report which PostHog credentials are missing
* [build-tools] add poll helpers, credentials-error factory, query abort signal
* [build-tools] move credential catalog and error into PosthogUtils
* [build-tools] accept an abort signal on all PostHog client requests
* [build-tools] Add eas/posthog_flag_rollout workflow function
* [build-tools] use PosthogUtils namespace in flag rollout
* [build-tools] name missing credentials in flag rollout
* [build-tools] use credentials-error factory in flag rollout
* [build-tools] use PosthogUtils.missingCredentialsError in flag rollout
* [build-tools] thread abort signal through flag rollout requests
* [build-tools] Add eas/posthog_wait_for_metric workflow function
* [build-tools] respect abort signal while waiting for metric
* [build-tools] name missing credentials when waiting for metric
* [build-tools] share poll helpers, harden operator guard, thread signal (metric)
* [build-tools] use PosthogUtils.missingCredentialsError when waiting for metric
* [build-tools] Add eas/posthog_upload_sourcemaps workflow function
* [build-tools] use PosthogUtils namespace in sourcemap upload
* [build-tools] name missing credentials in sourcemap upload
* [build-tools] use credentials-error factory in sourcemap upload
* [build-tools] use PosthogUtils.missingCredentialsError in sourcemap upload
* [build-tools] pass abort signal to the sourcemap upload subprocess
* [build-tools] Add eas/posthog_annotation workflow function
* [build-tools] use PosthogUtils namespace in annotation
* [build-tools] name missing credentials in annotation
* [build-tools] use credentials-error factory in annotation
* [build-tools] use PosthogUtils.missingCredentialsError in annotation
* [build-tools] thread abort signal through annotation request
* [build-tools] Add eas/posthog_capture_event workflow function
* [build-tools] address PostHog client review nits
* [build-tools] report which PostHog credentials are missing
* [build-tools] add poll helpers, credentials-error factory, query abort signal
* [build-tools] move credential catalog and error into PosthogUtils
* [build-tools] accept an abort signal on all PostHog client requests
* [build-tools] Add eas/posthog_flag_rollout workflow function
* [build-tools] use PosthogUtils namespace in flag rollout
* [build-tools] name missing credentials in flag rollout
* [build-tools] use credentials-error factory in flag rollout
* [build-tools] use PosthogUtils.missingCredentialsError in flag rollout
* [build-tools] thread abort signal through flag rollout requests
* [build-tools] Add eas/posthog_wait_for_metric workflow function
* [build-tools] respect abort signal while waiting for metric
* [build-tools] name missing credentials when waiting for metric
* [build-tools] share poll helpers, harden operator guard, thread signal (metric)
* [build-tools] use PosthogUtils.missingCredentialsError when waiting for metric
* [build-tools] Add eas/posthog_upload_sourcemaps workflow function
* [build-tools] use PosthogUtils namespace in sourcemap upload
* [build-tools] name missing credentials in sourcemap upload
* [build-tools] use credentials-error factory in sourcemap upload
* [build-tools] use PosthogUtils.missingCredentialsError in sourcemap upload
* [build-tools] pass abort signal to the sourcemap upload subprocess
* [build-tools] Add eas/posthog_capture_event workflow function
* [build-tools] address PostHog client review nits
* [build-tools] report which PostHog credentials are missing
* [build-tools] add poll helpers, credentials-error factory, query abort signal
* [build-tools] move credential catalog and error into PosthogUtils
* [build-tools] accept an abort signal on all PostHog client requests
* [build-tools] Add eas/posthog_flag_rollout workflow function
* [build-tools] use PosthogUtils namespace in flag rollout
* [build-tools] name missing credentials in flag rollout
* [build-tools] use credentials-error factory in flag rollout
* [build-tools] use PosthogUtils.missingCredentialsError in flag rollout
* [build-tools] thread abort signal through flag rollout requests
* [build-tools] Add eas/posthog_wait_for_metric workflow function
* [build-tools] respect abort signal while waiting for metric
* [build-tools] name missing credentials when waiting for metric
* [build-tools] share poll helpers, harden operator guard, thread signal (metric)
* [build-tools] use PosthogUtils.missingCredentialsError when waiting for metric
* [build-tools] Add eas/posthog_capture_event workflow function
* [build-tools] address PostHog client review nits
* [build-tools] report which PostHog credentials are missing
* [build-tools] add poll helpers, credentials-error factory, query abort signal
* [build-tools] move credential catalog and error into PosthogUtils
* [build-tools] accept an abort signal on all PostHog client requests
* [build-tools] Add eas/posthog_flag_rollout workflow function
* [build-tools] use PosthogUtils namespace in flag rollout
* [build-tools] name missing credentials in flag rollout
* [build-tools] use credentials-error factory in flag rollout
* [build-tools] use PosthogUtils.missingCredentialsError in flag rollout
* [build-tools] thread abort signal through flag rollout requests
## Summary
- include the simulator device name and human-readable runtime name in Device Run Session screen recording metadata
- carry those fields through the finish-step output and construct the artifact display name only during upload
- normalize CoreSimulator runtime identifiers such as `iOS-18-6` to `iOS 18.6`
- add focused coverage for runtime normalization and the uploaded metadata payload
## Why
The website needs device and runtime details to identify and present EAS-generated simulator screen recordings without parsing artifact names or CoreSimulator identifiers.
## Validation
- `yarn workspace @expo/build-tools typecheck`
- `yarn workspace @expo/build-tools jest-unit --runInBand src/utils/__tests__/IosSimulatorUtils.test.ts src/steps/functions/__tests__/uploadDeviceRunSessionScreenRecordings.test.ts`
## Summary
Observe the detached Android emulator process promise immediately after startup so expected exits during retry cleanup do not surface as unhandled promise rejections.
The emulator is managed through adb after launch. When a startup attempt times out and cleanup kills the emulator, the underlying SpawnPromise can reject while the retry loop is handling the readiness failure. Observing the promise keeps Sentry from capturing that cleanup path as an unhandled SIGKILL while preserving the returned promise for callers that still await it later.
## Test plan
- yarn run -T oxfmt packages/build-tools/src/utils/AndroidEmulatorUtils.ts
- yarn workspace @expo/build-tools typecheck
# Why
Argent now writes per-install tool-server state files such as `~/.argent/tool-server-<hash>.json`. The remote session step only watched the legacy `tool-server.json`, so it could time out even when the newly launched tool server was running.
# How
- expose the PID of detached processes
- add a process ancestry helper that works on macOS and Linux
- run Argent in the foreground beneath the detached `bunx` process and pass `--force`
- scan legacy and hashed Argent state files, selecting only the state whose PID descends from the launched `bunx` process
- validate the state schema includes the tool-server PID
- cover legacy, hashed, matching, and unrelated state files
# Test Plan
- `yarn workspace @expo/build-tools jest-unit packages/build-tools/src/steps/functions/__tests__/startArgentRemoteSession.test.ts --runInBand`
- `yarn workspace @expo/build-tools build`
- `yarn workspace @expo/worker build`
- `yarn workspace eas-cli build`
- `yarn workspace @expo/build-tools typecheck`
- `yarn lint`
- `yarn fmt:check`
## Summary
- add separate `eas/start_ios_simulator_recordings` and `eas/finish_ios_simulator_recordings` build functions
- poll booted iOS Simulators and record each session with `record-sim`
- retry failed recorder processes up to three times per simulator boot
- upload completed MP4 files as screen-recording device run session artifacts with recording metadata
- extend device run session artifact uploads to accept metadata
## Why
Device run sessions need simulator screen recordings that can start independently from remote sessions and be finalized and uploaded when the build workflow finishes.
## Impact
Workflows can explicitly start simulator recording polling and later finish and upload recordings. Missing `record-sim` binaries or device run session IDs disable recording or upload without failing the workflow.
## Validation
- `yarn workspace @expo/build-tools typecheck`
- `yarn workspace @expo/build-tools jest-unit steps/utils/__tests__/deviceRunSessionArtifacts.test.ts --runInBand`
- `yarn oxfmt --check` on all changed files
- `git diff --check origin/main...HEAD`
# Why
`syncCapabilitiesForEntitlementsAsync` in bundleIdCapabilities.ts wrapped `bundleId.updateBundleIdCapabilityAsync` in a try/catch whose block only rethrew for the specific error "bundle '…' cannot be deleted. Delete all the Apps". Every other Apple API error was silently swallowed while the function still returned the intended enabled/disabled lists — leading to undebuggable signing failures downstream. (Reported in issue #3986 against 18.12.3 and 20.5.1.)
# How
Rethrow every non-matching error with the same actionable context we already gave the "cannot be deleted" case — the Apple developer console URL and the EXPO_NO_CAPABILITY_SYNC=1 escape hatch — and preserve the underlying Apple error via `.cause` so EXPO_DEBUG=1 still surfaces the raw response. Log the failing request payload for all failures, not just the matched one, so future variants are debuggable from CLI output.
Also fix the "developer-mdn.apple.com" typo in the existing error message (the correct host is developer.apple.com).
Note that capability sync will not work correctly until the fix in https://github.com/expo/third-party/pull/161 is merged and released.
# Test Plan
- Expand the existing "throw useful error message" test to cover generic Apple errors
- Correct expected URL in test assertion
## Summary
Fixes the iOS Simulator `apsd` shutdown helper to use the launchd domain requested by iOS 26.5:
- disables and boots out `user/foreground/com.apple.apsd` first
- falls back to `system/com.apple.apsd` for compatibility
- verifies whether the service is still `state = running`
- treats `launchctl print` returning service-not-found as a successful stopped state
## Root cause
The previous implementation used `system/com.apple.apsd`. On iOS 26.5, `launchctl` warns to switch to `user/foreground/com.apple.apsd`, and workflow logs showed `bootout system/com.apple.apsd` exiting non-zero. That made a successful or partially successful disable look like a failure and could leave `apsd` running.
## Validation
- `yarn test packages/build-tools/src/utils/__tests__/IosSimulatorUtils.test.ts`
- `yarn typecheck` in the original checkout
- `git diff --check`
I also verified locally on a disposable iOS 26.5 simulator clone that `disable` + `bootout` with `user/foreground/com.apple.apsd` stops the service, and subsequent `launchctl print` returns service-not-found.
# Why
When running `observe:metrics` in non-interactive mode, navigation metric names could not be passed into the command; only app startup metrics were accepted.
# How
- Included the navigation metrics in the validation check
- Renamed the short names (keys) for the navigation metrics to avoid confusion with app startup metric short names
# Test Plan
- Tested locally
- Unit tests added and updated
# Why
`eas integrations:posthog:dashboard` should open the linked PostHog project already signed in, instead of a static URL that hits a login wall.
# How
Mints a one-time, signed-in link via the new `createPostHogDeepLink` mutation and opens it. The link carries a single-use token, so it isn't printed by default; pass `--show-link` to print it. `--json` / `--non-interactive` return the stable static URL, which minting failures also fall back to.
# Test Plan
Unit tests + verified end-to-end with `easd`
<img width="1548" height="484" alt="carbon (5)" src="https://github.com/user-attachments/assets/c00961e5-da49-456c-af1e-c4a2700eece4" />
Ports the Expo CLI agent telemetry integration from https://github.com/expo/expo/pull/47248 to EAS CLI.
Adds `agent-cli-detector` to EAS CLI telemetry so events recorded from known coding agents include lightweight agent context. The CLI uses `[email protected]`, which exposes a CommonJS entry point, so detection stays synchronous and does not participate in telemetry initialization.
When a known agent is detected, telemetry event context includes:
```ts
context.agent = {
id: string,
sessionId: string | undefined,
}
```
When no known agent is detected, `context.agent` is omitted to keep RudderStack and BigQuery payloads sparse. CLI version is already available in telemetry, so analysis can distinguish newer CLI versions with no detected agent from older CLI versions without this integration.
Tests cover detected and undetected telemetry payloads. Locally verified with `yarn workspace eas-cli test src/analytics/__tests__/AnalyticsManager-test.ts --runInBand`, `yarn workspace eas-cli typecheck`, `yarn lint-changelog`, and targeted `oxlint`.
## Summary
- replace the remote device session infinite waits with polling for the device run session to become `STOPPED`
- let agent-device, Argent, and serve-sim session steps return normally after the stop signal so the worker can report success and the launcher can close the VM
- make Argent artifact polling stoppable, drain one final artifact list on shutdown, and wait briefly for pending uploads
## Why
Stopping a device run session currently cancels the backing turtle job run, which tears down the VM immediately. That skips graceful shutdown work such as finishing recordings and uploading session artifacts. This change makes the worker observe the session stop state and finish cleanup before exiting.
## Validation
- `yarn oxfmt <modified files>`
- `yarn --cwd packages/build-tools test -- remoteDeviceRunSession`
- `yarn --cwd packages/build-tools test -- argentArtifacts`
- `yarn --cwd packages/build-tools typecheck`
- `git diff --check`
## Note
This relies on the backend/API behavior changing so `ensureDeviceRunSessionStopped` marks the device run session `STOPPED` without canceling the turtle job run, and allows artifact uploads while the stopped session is still cleaning up.