[brentvatne]

Why

• eas simulator:get on a failed session shows Status ERRORED and nothing about why, in both human and --json output.
• Failures without a specific error code reach the CLI only as UNKNOWN_ERROR / UNKNOWN_CUSTOM_BUILD_ERROR, "Unknown error. See logs for more information." Every failed adb or simctl call ends up there. The real cause is only in the job run log.

How

Shared lookup (src/simulator/jobRunErrors.ts, used by simulator:start and simulator:get):

• Job run errors: fetched with the failure-only query from [eas-cli] Show why a simulator session failed to start #4693.
• Unknown errors: when every job run error is an unknown error, the CLI downloads the job run log and prints:
  • for the first step that failed (marker: end-step, result: fail): the stderr line that names an error, exception or fatal error, the stderr line right after it, and the step's last error-level line;
  • for later failed steps, only their last error-level line;
  • at most four lines, each truncated to 500 characters.
  • If no step failed, it prints nothing from the log and keeps the job run's own message. That way an error a successful step ignored (for example "Failed to upload … Ignoring error.") is never shown as the cause.
• Masked text only: it prints the log line's msg, which the worker masks, and never err.message or err.stack, which it doesn't.
  • Unknown failures inside a build phase log an empty msg (context.ts:310). For those, the CLI keeps the job run's own message, which already names the phase.
  • A separate worker PR masks err too.
• Query bound: the same bounded, quiet, non-retried lookup as [eas-cli] Show why a simulator session failed to start #4693, plus logFileUrls.
• Download bound:
  • it goes through eas-cli's proxy-aware fetch, which throws on status ≥ 400;
  • it has a 20 MiB limit and its own 10s timeout, and Ctrl+C cancels it too;
  • an owned AbortController is aborted in finally, so the connection closes even after a size-limit rejection.
• Fallback: any failure falls back to the job run's own error messages. Debug output leaves out node-fetch's error message, which contains the signed log URL.

simulator:get: it looks up the errors only when the session or its job run is ERRORED. Human output gets a "Session errors:" section, and --json always includes errors: [{ errorCode, message, docsUrl? }].

Against a real Android session that failed with UNKNOWN_ERROR (a launch argument with a space):

Status       ERRORED
Session errors:
  Launch application: Error: Activity not started, unable to resolve Intent { act=android.intent.action.MAIN … pkg=launchArgProbe }
  Launch application: adb shell am start launchArgProbe hello world -n dev.brent.euxy/dev.brent.euxy.MainActivity exited with non-zero code: 1

Before, that printed "Unknown error. See logs for more information."

Test Plan

• 251 tests pass across src/commands/simulator, jobRunErrors.test.ts, DeviceRunSessionQuery-test.ts and createGraphqlClient.
• Log selection: tests use the shape of a real log, and cover:
  • a phase failure with a secret in err (nothing printed);
  • stderr from a step that didn't fail, and an error a successful step ignored (both left out);
  • a two-line stderr cause;
  • a cleanup step failing after the real failure (the first step's cause is kept).
• Transport: tests run against a real loopback server through the real src/fetch.ts, with https_proxy cleared. They cover a 404, an oversized log, a stalled download, Ctrl+C, and an already-aborted signal, and check that the server sees the connection close.
• Query cancellation: tested against a real urql client and a server that never answers.
• Mutation checks: each of these fails a test: removing the finally abort; removing the unsubscribe on abort; restoring the whole-log fallback; logging the full download error.
• yarn typecheck, oxlint and oxfmt --check pass.
• Ran simulator:get from this branch against the production session above and an iOS session with a bad --device.

🤖 Generated with Claude Code

[github-actions]

Subscribed to pull request

File Patterns	Mentions
packages/eas-cli/**	@douglowder

Generated by CodeMention

Warning: The preamble and epilogue options in commentConfiguration are deprecated. Use template instead.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 75.37%. Comparing base (f775b10) to head (e20caba).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4694      +/-   ##
==========================================
+ Coverage   75.33%   75.37%   +0.05%     
==========================================
  Files        1053     1054       +1     
  Lines       52257    52338      +81     
  Branches    11383    11446      +63     
==========================================
+ Hits        39362    39445      +83     
+ Misses      11907    11905       -2     
  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.

[gwdp]

LGTM, nice UX improvement!

Codex found one small P2 worth fixing while here (optional): strip terminal control characters from downloaded error messages and step names before printing them, so crafted log text cannot clear the terminal or spoof output.

— Codex

[brentvatne]

Done in 669c0eb. Job run error messages, step names and log lines now have escape sequences and control characters stripped before they're printed, with newlines collapsed so a message can't fake extra lines. That also fixed a related bug: a colored Error: stderr line wasn't being picked as the cause, because the pattern was matched against the raw text.

[github-actions]

✅ Thank you for adding a changeset!