fix(ci): Run crash.yml first so next flow verifies post-crash recovery

Moving crash.yml to last made the "implicit recovery" argument hollow —
no subsequent flow existed to prove the app can restart. Running it
first ensures the next flow's launchTestAppClear.yml (killApp +
launchApp + assertTestReady) acts as the post-crash recovery check.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: antonis

.github/workflows/e2e-v2.yml

@@ -526,8 +526,7 @@ jobs:
       - name: Run tests on iOS
         if: ${{ steps.platform-check.outputs.skip != 'true' && matrix.platform == 'ios' }}
         env:
-          # Increase timeout for Maestro iOS driver startup from default 60s to 180s.
-          # Cirrus Labs Tart VMs can be slow to respond after simulator boot.
+          # Increase timeout for Maestro iOS driver startup (default is 60s, some CI runners need more time)
           MAESTRO_DRIVER_STARTUP_TIMEOUT: 180000
         run: ./dev-packages/e2e-tests/cli.mjs ${{ matrix.platform }} --test
 

dev-packages/e2e-tests/cli.mjs

@@ -294,10 +294,12 @@ if (actions.includes('test')) {
     const flowFiles = fs.readdirSync(maestroDir)
       .filter(f => f.endsWith('.yml') && !f.startsWith('utils'))
       .sort((a, b) => {
-        // Run crash.yml last — it kills the app, and running it last avoids
-        // any potential simulator instability affecting subsequent flows.
-        if (a === 'crash.yml') return 1;
-        if (b === 'crash.yml') return -1;
+        // Run crash.yml first — it kills the app via nativeCrash(), and the
+        // NEXT flow's launchTestAppClear.yml verifies post-crash recovery
+        // (killApp + launchApp + assertTestReady). Per-flow process isolation
+        // ensures the crashed Maestro session doesn't affect other flows.
+        if (a === 'crash.yml') return -1;
+        if (b === 'crash.yml') return 1;
         return a.localeCompare(b);
       });
 

dev-packages/e2e-tests/maestro/crash.yml

@@ -4,8 +4,8 @@ jsEngine: graaljs
 - runFlow: utils/launchTestAppClear.yml
 - tapOn: "Crash"
 
-# Post-crash app relaunch is not tested here — Maestro's connection to the
-# app is broken after a native crash, making in-session launchApp unreliable
-# on Cirrus Labs Tart VMs. With per-flow process isolation, each subsequent
-# flow starts a fresh Maestro session that launches the app from scratch,
-# which implicitly verifies post-crash recovery.
+# Post-crash recovery is verified by the NEXT flow in the run order.
+# Each flow runs in its own Maestro process, so the next flow's
+# launchTestAppClear.yml (killApp + launchApp + assertTestReady)
+# proves the app can start after a native crash. crash.yml is sorted
+# to run first to ensure at least one flow follows it.