[brentvatne]

Follows #4695 (merged), which added the Android shell quoting and the checks this PR moves earlier.

Why

On iOS, agent-device sessions validate launch inputs at the start of the step (validateServeSimLaunchOptions). On Android, launch inputs were only checked inside launchApplicationAsync, which runs after the emulator boots, the build downloads and the app installs. Input that can never work, such as a NUL character or a command too long for adb, therefore failed about a minute into a billed session.

Nothing in #4263 or #4324 says this was deliberate. #4324 made the up-front check iOS-only because it guards serve-sim's own limits, and Android never got an equivalent.

The check doesn't depend on the controller, so it belongs in @expo/eas-build-job. www can then run it when a session is created, for every session type and client, before any job starts.

How

• @expo/eas-build-job now exports, from deviceRunSession.ts:
  • quoteAndroidShellCommand: the quoting from [build-tools] Quote Android launch arguments for the device shell #4695, plus the NUL and 65,000-byte checks. It measures bytes with TextEncoder, so it doesn't need Node's Buffer.
  • validateAndroidLaunchOptions({ launchArgs, openUrl }): runs those checks before the launch component (<package>/<activity>) is known. It measures the command without the component, so it rejects only commands that can never launch, and the launch still checks the exact command.
• launchApplication.ts now uses quoteAndroidShellCommand from @expo/eas-build-job instead of its own copy. The launch still checks the exact command.
• startAgentDeviceSession calls validateAndroidLaunchOptions next to the iOS check, so agent-device sessions fail before the emulator boots.
• Follow-up in universe: once www's @expo/eas-build-job bump includes this, call validateAndroidLaunchOptions in validateApplicationLaunchOptions for Android sessions. That covers argent and Appium sessions too. They only receive launch inputs in the later eas/launch_application step, so until then they check at launch.

Test Plan

• New tests in eas-build-job's deviceRunSession.test.ts:
  • quoting;
  • NUL rejection;
  • the length check, accepting a command of exactly 65,000 bytes without the component and rejecting one byte more;
  • byte counting, with a multibyte value at the limit. Counting characters instead fails this test;
  • the URL check.
• New tests in startAgentDeviceSession.test.ts: a NUL argument, a NUL URL and an oversized command each fail before the emulator starts, the build downloads or the session starts. With the call removed, all three fail.
• The full @expo/build-tools unit suite passes (2,438 tests), and so do yarn typecheck, yarn lint and yarn fmt:check.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 75.31%. Comparing base (53ac936) to head (8aabf49).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4696      +/-   ##
==========================================
+ Coverage   75.30%   75.31%   +0.01%     
==========================================
  Files        1053     1053              
  Lines       52170    52181      +11     
  Branches    11356    11398      +42     
==========================================
+ Hits        39281    39293      +12     
+ Misses      11901    11900       -1     
  Partials      988      988              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

✅ Thank you for adding a changeset!

[gwdp]

Thank you!!