fix: clean up maestro test reporter output (#935)

* fix: clean up maestro test reporter output * chore: enable expo build disk cache * refactor: simplify replay progress detail formatting * fix: surface replay runner recovery hints * fix: prioritize ios runner recovery hint * fix: avoid trailing punctuation in runner state hint * fix: keep internal cleanup scripts out of runner hints * chore: remove redundant maestro test app open flag * fix: make test app maestro flow self-contained * fix: simplify maestro test duration output * fix: refine maestro test summary output * fix: dim maestro live progress counters * test: clear test app state before maestro flow * test: update maestro reporter progress expectations * chore: remove maestro app open flag handling * fix: apply maestro reporter cleanup to default reporter
2026-09-14 20:06:34 +08:00 · 2026-06-29 21:02:30 +02:00
parent 8cc6bb8b92
commit 6ae0612ebc
25 changed files with 449 additions and 87 deletions
@@ -127,18 +127,15 @@ server from its launcher.
 The Maestro prototype suite lives in `examples/test-app/maestro` and runs through
 `agent-device replay --maestro`:

-```bash
-pnpm test-app:maestro:ios -- --open "Agent Device Tester"
-pnpm test-app:maestro:android -- --open "Agent Device Tester"
-```
-
-When the development build is already open and connected to Metro, omit
-`--open` and run the suite against the existing session:
-
 ```bash
 pnpm test-app:maestro:ios
+pnpm test-app:maestro:android
 ```

+The Maestro flow includes `launchApp`, so the suite launches the app inside each
+test attempt. Start Metro first when the installed development build needs the
+local bundle.
+
 The suite intentionally covers the compat layer syntax used by public Maestro suites:
 `runFlow` file/inline blocks, `when.platform`, config hooks, deterministic `repeat.times`,
 flow `env`, selectors, input, assertions, and swipe.
@@ -5,6 +5,9 @@
    "version": "1.0.0",
    "orientation": "default",
    "userInterfaceStyle": "automatic",
+    "buildCacheProvider": {
+      "plugin": "expo-build-disk-cache"
+    },
    "plugins": ["expo-router"],
    "ios": {
      "supportsTablet": true,
@@ -2,8 +2,10 @@ appId: com.callstack.agentdevicelab
 env:
  CHECKOUT_NAME: Ada Lovelace
  CHECKOUT_EMAIL: ada@example.com
-  PICKUP_TAPS: "2"
+  PICKUP_TAPS: '2'
 onFlowStart:
+  - launchApp:
+      clearState: true
  - assertVisible: Agent Device Tester
 onFlowComplete:
  - assertVisible: Delivery choices
@@ -1,4 +1,7 @@
 ---
+- scrollUntilVisible:
+    element:
+      id: home-open-form
 - tapOn:
    id: home-open-form
 - assertVisible: Checkout form
@@ -26,6 +26,7 @@
  },
  "devDependencies": {
    "@types/react": "~19.2.2",
+    "expo-build-disk-cache": "^0.7.4",
    "typescript": "~6.0.3"
  }
 }
@@ -59,6 +59,9 @@ importers:
      '@types/react':
        specifier: ~19.2.2
        version: 19.2.14
+      expo-build-disk-cache:
+        specifier: ^0.7.4
+        version: 0.7.4(@expo/cli@56.1.16(@expo/dom-webview@56.0.5)(@expo/metro-runtime@56.0.15)(expo-constants@56.0.18)(expo-font@56.0.7)(expo-router@56.2.11)(expo@56.0.12)(react-native@0.85.3(@babel/core@7.29.0)(@react-native/metro-config@0.86.0(@babel/core@7.29.0))(@types/react@19.2.14)(react@19.2.3))(react@19.2.3)(typescript@6.0.3))(@expo/config@56.0.9(typescript@6.0.3))(typescript@6.0.3)
      typescript:
        specifier: ~6.0.3
        version: 6.0.3
@@ -1216,6 +1219,10 @@ packages:
    resolution: {integrity: sha512-/Nf7TyzTx6S3yRJObOAV7956r8cr2+Oj8AC5dt8wSP3BQAoeX58NoHyCU8P8zGkNXStjTSi6fzO6F0pBdcYbEg==}
    engines: {node: '>= 0.8'}

+  callsites@3.1.0:
+    resolution: {integrity: sha512-P8BjAsXvZS+VIDUI11hHCQEv74YT67YUi5JJFNWIqL235sBmjX4+qx9Muvls5ivyNENctx46xQLQ3aTuE7ssaQ==}
+    engines: {node: '>=6'}
+
  camelcase@6.3.0:
    resolution: {integrity: sha512-Gmy6FhYlCY7uOElZUSbxo2UCDH8owEk996gkbrpsgGtrJLM3J7jGxl9Ic7Qwwj4ivOE5AWZWRMecDdF7hqGjFA==}
    engines: {node: '>=10'}
@@ -1314,6 +1321,15 @@ packages:
  core-js-compat@3.49.0:
    resolution: {integrity: sha512-VQXt1jr9cBz03b331DFDCCP90b3fanciLkgiOoy8SBHy06gNf+vQ1A3WFLqG7I8TipYIKeYK9wxd0tUrvHcOZA==}

+  cosmiconfig@9.0.2:
+    resolution: {integrity: sha512-gtTZxTDau1wL7Y7zifc2dd8jHSK/k6BTx/2Xp/BpdlAdnlYWFVt7qhJqgwi7637yRwRQ3qL4ZidbB4I8tA5VOg==}
+    engines: {node: '>=14'}
+    peerDependencies:
+      typescript: '>=4.9.5'
+    peerDependenciesMeta:
+      typescript:
+        optional: true
+
  cross-spawn@7.0.6:
    resolution: {integrity: sha512-uV2QOWP2nWzsy2aMp8aRibhi9dlzF5Hgh5SHaB9OiTGEyDTiJJyx0uy51QXdyWbtAHNua4XJzUKca3OzKUd3vA==}
    engines: {node: '>= 8'}
@@ -1405,6 +1421,13 @@ packages:
    resolution: {integrity: sha512-Q0n9HRi4m6JuGIV1eFlmvJB7ZEVxu93IrMyiMsGC0lrMJMWzRgx6WGquyfQgZVb31vhGgXnfmPNNXmxnOkRBrg==}
    engines: {node: '>= 0.8'}

+  env-paths@2.2.1:
+    resolution: {integrity: sha512-+h1lkLKhZMTYjog1VEpJNG7NZJWcuc2DDk/qsqSTRRCOXiLjeQ1d1/udrUGhqMxUgAlwKNZ0cf2uqan5GLuS2A==}
+    engines: {node: '>=6'}
+
+  error-ex@1.3.4:
+    resolution: {integrity: sha512-sqQamAnR14VgCr1A618A3sGrygcpK+HEbenA/HiEAkkUwcZIIB/tgWqHFxWgOyDh4nB4JCRimh79dR5Ywc9MDQ==}
+
  error-stack-parser@2.1.4:
    resolution: {integrity: sha512-Sk5V6wVazPhq5MhpO+AUxJn5x7XSXGl1R93Vn7i+zS15KDVxQijejNCrz8340/2bgLBjR9GtEG8ZVKONDjcqGQ==}

@@ -1438,6 +1461,13 @@ packages:
      react: '*'
      react-native: '*'

+  expo-build-disk-cache@0.7.4:
+    resolution: {integrity: sha512-mU1NaPC2kAwUZ0bVjbuRHQ+FBKnlFvNwu6BR1ZKPRV9c7McvL7Ir1V5B08VvzTDtEMpzYjhyITPSU6loO9Bsew==}
+    engines: {node: '>=18.0.0'}
+    peerDependencies:
+      '@expo/cli': '*'
+      '@expo/config': '*'
+
  expo-constants@56.0.18:
    resolution: {integrity: sha512-8AMtbDGl/WVPnWlmbpGmvcdnNCy9E4PFnwdVwj600vljkMDPSxcAcjw8GVXEPk3PpZ+ngTqsrkltWyj0UKYAxw==}
    peerDependencies:
@@ -1735,6 +1765,10 @@ packages:
    engines: {node: '>=16.x'}
    hasBin: true

+  import-fresh@3.3.1:
+    resolution: {integrity: sha512-TR3KfrTZTYLPB6jUjfx6MF9WcWrHL9su5TObK4ZkYgBdWKPOFoSoQIdEuTuR82pmtxH2spWG9h6etwfr1pLBqQ==}
+    engines: {node: '>=6'}
+
  indent-string@4.0.0:
    resolution: {integrity: sha512-EdDDZu4A2OyIK7Lr/2zG+w5jmbuk1DVBnEwREQvBzspBJkCEbRa8GxU1lghYcaGJCnRWibjDXlq779X1/y5xwg==}
    engines: {node: '>=8'}
@@ -1745,6 +1779,9 @@ packages:
  invariant@2.2.4:
    resolution: {integrity: sha512-phJfQVBuaJM5raOpJjSfkiD6BpbCE4Ns//LaXl6wGYtUBY83nWS6Rf9tXm2e8VaK60JEjYldbPif/A2B1C2gNA==}

+  is-arrayish@0.2.1:
+    resolution: {integrity: sha512-zz06S8t0ozoDXMG+ube26zeCTNXcKIPJZJi8hBrF4idCLms4CG9QtK7qBl1boi5ODzFpjswb5JPmHCbMpjaYzg==}
+
  is-arrayish@0.3.4:
    resolution: {integrity: sha512-m6UrgzFVUYawGBh1dUsWR5M2Clqic9RVXC/9f8ceNlv2IcO9j9J/z8UoCLPqtsPBFNzEpfR3xftohbfqDx8EQA==}

@@ -1806,6 +1843,9 @@ packages:
    engines: {node: '>=6'}
    hasBin: true

+  json-parse-even-better-errors@2.3.1:
+    resolution: {integrity: sha512-xyFwyhro/JEof6Ghe2iz2NcXoj2sloNsWr/XsERDK/oiPCfaNhl5ONfp+jQdAZRQQ0IJWNzH9zIZF7li91kh2w==}
+
  json5@2.2.3:
    resolution: {integrity: sha512-XmOWe7eyHYH14cLdVPoyg+GOH3rYX++KpzrylJwSW98t3Nk+U8XOl8FWKOgwtzdb8lXGf6zYwDUzeHMWfxasyg==}
    engines: {node: '>=6'}
@@ -1900,6 +1940,9 @@ packages:
    resolution: {integrity: sha512-NXYBzinNrblfraPGyrbPoD19C1h9lfI/1mzgWYvXUTe414Gz/X1FD2XBZSZM7rRTrMA8JL3OtAaGifrIKhQ5yQ==}
    engines: {node: '>= 12.0.0'}

+  lines-and-columns@1.2.4:
+    resolution: {integrity: sha512-7ylylesZQ/PV29jhEDl3Ufjo6ZX7gCqJr5F7PKrqc93v7fzSymt1BpwEU8nAUXs8qzzvqhbjhK5QZg6Mt/HkBg==}
+
  lodash.debounce@4.0.8:
    resolution: {integrity: sha512-FT1yDzDYEoYWhnSGnpE/4Kj1fLZkDFyqRb7fNt6FdYOSxlUWAtp42Eh6Wb0rGIv/m9Bgo7x4GhQbm5Ys4SG5ow==}

@@ -2058,6 +2101,11 @@ packages:
    engines: {node: ^10 || ^12 || ^13.7 || ^14 || >=15.0.1}
    hasBin: true

+  nanoid@3.3.15:
+    resolution: {integrity: sha512-y7Wygv/7mEOvxTuEQDB8StXdMRBWf1kR/tlhAzBRUFkB2jfcLOAxO/SHmOO2zgz1pVgK29/kyupn059/bCHdjA==}
+    engines: {node: ^10 || ^12 || ^13.7 || ^14 || >=15.0.1}
+    hasBin: true
+
  negotiator@0.6.3:
    resolution: {integrity: sha512-+EUsqGPLsM+j/zdChZjsnX51g4XrHFOIXwfnCVPGlQk/k5giakcKsuxCObBRu6DSm9opw/O6slWbJdghQM4bBg==}
    engines: {node: '>= 0.6'}
@@ -2115,6 +2163,14 @@ packages:
    resolution: {integrity: sha512-eNwHudNbO1folBP3JsZ19v9azXWtQZjICdr3Q0TDPIaeBQ3mXLrh54wM+er0+hSp+dWKf+Z8KM58CYzEyIYxYg==}
    engines: {node: '>=6'}

+  parent-module@1.0.1:
+    resolution: {integrity: sha512-GQ2EWRpQV8/o+Aw8YqtfZZPfNRWZYkbidE9k5rpl/hC3vtHHBfGm2Ifi6qWV+coDGkrUKZAxE3Lot5kcsRlh+g==}
+    engines: {node: '>=6'}
+
+  parse-json@5.2.0:
+    resolution: {integrity: sha512-ayCKvm/phCGxOkYRSCM82iDwct8/EonSEgCSxWxD7ve6jHggsFl4fZVQBPRNgQoKiuV/odhFrGzQXZwbifC8Rg==}
+    engines: {node: '>=8'}
+
  parse-png@2.1.0:
    resolution: {integrity: sha512-Nt/a5SfCLiTnQAjx3fHlqp8hRgTL3z7kTQZzvIMS9uCAepnCyjpdEc6M/sz69WqMBdaDBw9sF1F1UaHROYzGkQ==}
    engines: {node: '>=10'}
@@ -2343,6 +2399,10 @@ packages:
    resolution: {integrity: sha512-fGxEI7+wsG9xrvdjsrlmL22OMTTiHRwAMroiEeMgq8gzoLC/PQr7RsRDSTLUg/bZAZtF+TVIkHc6/4RIKrui+Q==}
    engines: {node: '>=0.10.0'}

+  resolve-from@4.0.0:
+    resolution: {integrity: sha512-pb/MYmXstAkysRFx8piNI1tGFNQIFA3vkE3Gq4EuA1dF6gHp/+vgZqsCGJapvy8N3Q+4o7FwvquPJcnZ7RYy4g==}
+    engines: {node: '>=4'}
+
  resolve-from@5.0.0:
    resolution: {integrity: sha512-qYg9KP24dD5qka9J47d0aVky0N+b4fTU89LN9iDnjB5waksiC49rvMB0PrUJQGoTmH50XPiqOvAjDfaijGxYZw==}
    engines: {node: '>=8'}
@@ -2727,6 +2787,9 @@ packages:
  zod@3.25.76:
    resolution: {integrity: sha512-gzUt/qt81nXsFGKIFcC3YnfEAx5NkunCfnDlvuBSSFS02bcXu4Lmea0AFIUwbLWxWPx3d9p8S5QoaujKcNQxcQ==}

+  zod@4.4.3:
+    resolution: {integrity: sha512-ytENFjIJFl2UwYglde2jchW2Hwm4GJFLDiSXWdTrJQBIN9Fcyp7n4DhxJEiWNAJMV1/BqWfW/kkg71UDcHJyTQ==}
+
 snapshots:

  '@adobe/css-tools@4.5.0': {}
@@ -4233,6 +4296,8 @@ snapshots:

  bytes@3.1.2: {}

+  callsites@3.1.0: {}
+
  camelcase@6.3.0: {}

  caniuse-lite@1.0.30001787: {}
@@ -4346,6 +4411,15 @@ snapshots:
    dependencies:
      browserslist: 4.28.2

+  cosmiconfig@9.0.2(typescript@6.0.3):
+    dependencies:
+      env-paths: 2.2.1
+      import-fresh: 3.3.1
+      js-yaml: 4.1.1
+      parse-json: 5.2.0
+    optionalDependencies:
+      typescript: 6.0.3
+
  cross-spawn@7.0.6:
    dependencies:
      path-key: 3.1.1
@@ -4402,6 +4476,12 @@ snapshots:

  encodeurl@2.0.0: {}

+  env-paths@2.2.1: {}
+
+  error-ex@1.3.4:
+    dependencies:
+      is-arrayish: 0.2.1
+
  error-stack-parser@2.1.4:
    dependencies:
      stackframe: 1.3.4
@@ -4429,6 +4509,16 @@ snapshots:
      - supports-color
      - typescript

+  expo-build-disk-cache@0.7.4(@expo/cli@56.1.16(@expo/dom-webview@56.0.5)(@expo/metro-runtime@56.0.15)(expo-constants@56.0.18)(expo-font@56.0.7)(expo-router@56.2.11)(expo@56.0.12)(react-native@0.85.3(@babel/core@7.29.0)(@react-native/metro-config@0.86.0(@babel/core@7.29.0))(@types/react@19.2.14)(react@19.2.3))(react@19.2.3)(typescript@6.0.3))(@expo/config@56.0.9(typescript@6.0.3))(typescript@6.0.3):
+    dependencies:
+      '@expo/cli': 56.1.16(@expo/dom-webview@56.0.5)(@expo/metro-runtime@56.0.15)(expo-constants@56.0.18)(expo-font@56.0.7)(expo-router@56.2.11)(expo@56.0.12)(react-native@0.85.3(@babel/core@7.29.0)(@react-native/metro-config@0.86.0(@babel/core@7.29.0))(@types/react@19.2.14)(react@19.2.3))(react@19.2.3)(typescript@6.0.3)
+      '@expo/config': 56.0.9(typescript@6.0.3)
+      cosmiconfig: 9.0.2(typescript@6.0.3)
+      env-paths: 2.2.1
+      zod: 4.4.3
+    transitivePeerDependencies:
+      - typescript
+
  expo-constants@56.0.18(expo@56.0.12)(react-native@0.85.3(@babel/core@7.29.0)(@react-native/metro-config@0.86.0(@babel/core@7.29.0))(@types/react@19.2.14)(react@19.2.3)):
    dependencies:
      '@expo/env': 2.3.0
@@ -4755,6 +4845,11 @@ snapshots:
    dependencies:
      queue: 6.0.2

+  import-fresh@3.3.1:
+    dependencies:
+      parent-module: 1.0.1
+      resolve-from: 4.0.0
+
  indent-string@4.0.0: {}

  inherits@2.0.4: {}
@@ -4763,6 +4858,8 @@ snapshots:
    dependencies:
      loose-envify: 1.4.0

+  is-arrayish@0.2.1: {}
+
  is-arrayish@0.3.4: {}

  is-core-module@2.16.1:
@@ -4820,6 +4917,8 @@ snapshots:

  jsesc@3.1.0: {}

+  json-parse-even-better-errors@2.3.1: {}
+
  json5@2.2.3: {}

  kleur@3.0.3: {}
@@ -4884,6 +4983,8 @@ snapshots:
      lightningcss-win32-arm64-msvc: 1.32.0
      lightningcss-win32-x64-msvc: 1.32.0

+  lines-and-columns@1.2.4: {}
+
  lodash.debounce@4.0.8: {}

  lodash.throttle@4.1.1: {}
@@ -5129,6 +5230,8 @@ snapshots:

  nanoid@3.3.11: {}

+  nanoid@3.3.15: {}
+
  negotiator@0.6.3: {}

  negotiator@0.6.4: {}
@@ -5182,6 +5285,17 @@ snapshots:
      strip-ansi: 5.2.0
      wcwidth: 1.0.1

+  parent-module@1.0.1:
+    dependencies:
+      callsites: 3.1.0
+
+  parse-json@5.2.0:
+    dependencies:
+      '@babel/code-frame': 7.29.0
+      error-ex: 1.3.4
+      json-parse-even-better-errors: 2.3.1
+      lines-and-columns: 1.2.4
+
  parse-png@2.1.0:
    dependencies:
      pngjs: 3.4.0
@@ -5213,7 +5327,7 @@ snapshots:

  postcss@8.5.12:
    dependencies:
-      nanoid: 3.3.11
+      nanoid: 3.3.15
      picocolors: 1.1.1
      source-map-js: 1.2.1

@@ -5447,6 +5561,8 @@ snapshots:

  require-directory@2.1.1: {}

+  resolve-from@4.0.0: {}
+
  resolve-from@5.0.0: {}

  resolve-workspace-root@2.0.1: {}
@@ -5757,3 +5873,5 @@ snapshots:
      yargs-parser: 21.1.1

  zod@3.25.76: {}
+
+  zod@4.4.3: {}
@@ -11,7 +11,6 @@ const options = {
  platform: 'ios',
  session: 'test-app-maestro',
  flowDir: path.join(repoRoot, 'examples', 'test-app', 'maestro'),
-  openTarget: '',
  close: false,
  passthrough: [],
 };
@@ -37,11 +36,6 @@ for (let index = 2; index < process.argv.length; index += 1) {
    index += 1;
    continue;
  }
-  if (arg === '--open' && process.argv[index + 1]) {
-    options.openTarget = process.argv[index + 1];
-    index += 1;
-    continue;
-  }
  if (arg === '--close') {
    options.close = true;
    continue;
@@ -67,14 +61,9 @@ function runAgentDevice(args) {
  });
 }

-if (options.openTarget) {
-  runAgentDevice(['open', options.openTarget, '--platform', options.platform, ...options.passthrough]);
-  runAgentDevice(['wait', 'Agent Device Tester', '30000', '--platform', options.platform, ...options.passthrough]);
-}
-
 runAgentDevice([
  'test',
-  options.flowDir,
+  ...flows,
  '--maestro',
  '--platform',
  options.platform,
@@ -142,10 +142,11 @@ test('test command prints suite summary and exits non-zero on failures', async (
  assert.doesNotMatch(result.stderr, /Running replay suite\.\.\./);
  assert.doesNotMatch(result.stdout, /✓ 01-pass\.ad \(0\.01s\)/);
  assert.doesNotMatch(result.stdout, /⨯ "Checkout failure" in 02-fail\.ad/);
+  assert.match(result.stdout, /Failures:\n  Checkout failure\n    file: 02-fail\.ad/);
  assert.match(result.stdout, /Replay failed at step 1 \(open Demo\): boom/);
  assert.match(result.stdout, /artifacts: \/tmp\/test-artifacts\/02-fail/);
  assert.doesNotMatch(result.stdout, /SKIP \/tmp\/03-skip\.ad/);
-  assert.match(result.stdout, /Test summary: 1 passed, 1 failed in 0\.025s/);
+  assert.match(result.stdout, /Test summary: 1 passed \(3\), 1 failed in 0\.025s/);
 });

 test('test command --verbose prints all test statuses', async () => {
@@ -158,7 +159,44 @@ test('test command --verbose prints all test statuses', async () => {
  assert.doesNotMatch(result.stderr, /Running replay suite\.\.\./);
  assert.doesNotMatch(result.stdout, /✓ 01-pass\.ad \(0\.01s\)/);
  assert.doesNotMatch(result.stdout, /SKIP 03-skip\.ad/);
-  assert.match(result.stdout, /Test summary: 1 passed, 1 failed in 0\.025s/);
+  assert.match(result.stdout, /Test summary: 1 passed \(3\), 1 failed in 0\.025s/);
+});
+
+test('test command colors suite summary segments when color is enabled', async () => {
+  const result = await runCliCapture(
+    ['test', './suite'],
+    async () => ({
+      ok: true,
+      data: {
+        total: 1,
+        executed: 1,
+        passed: 1,
+        failed: 0,
+        skipped: 0,
+        notRun: 0,
+        durationMs: 25,
+        failures: [],
+        tests: [
+          {
+            file: '/tmp/01-pass.ad',
+            session: 'default:test:suite:1',
+            status: 'passed',
+            durationMs: 25,
+            attempts: 1,
+          },
+        ],
+      },
+    }),
+    { env: { FORCE_COLOR: '1', NO_COLOR: undefined } },
+  );
+
+  assert.equal(result.code, null);
+  assert.ok(
+    result.stdout.includes(
+      'Test summary: \u001B[32m1 passed\u001B[39m \u001B[2m(1)\u001B[22m in \u001B[33m0.025s\u001B[39m',
+    ),
+  );
+  assert.doesNotMatch(result.stdout, /0 failed/);
 });

 test('test command --verbose omits step telemetry for passing tests without debug mode', async () => {
@@ -388,13 +426,13 @@ test('test command reports flaky passed-on-retry cases in the default summary',
  assert.doesNotMatch(result.stdout, /FLAKY/);
  assert.doesNotMatch(
    result.stdout,
-    /^✓ "Authentication flow" in auth-flow\.yml \(passed attempt 17\.5s, total 112\.2s\)$/m,
+    /^✓ Authentication flow \(passed attempt 17\.5s, total 112\.2s\)$/m,
  );
-  assert.match(result.stdout, /Test summary: 1 passed, 0 failed, 1 flaky in 0\.025s/);
+  assert.match(result.stdout, /Test summary: 1 passed \(1\), 1 flaky in 0\.025s/);
  assert.match(result.stdout, /Flaky tests:/);
  assert.match(
    result.stdout,
-    /✓ "Authentication flow" in auth-flow\.yml after 2 attempts \(passed attempt 17\.5s, total 112\.2s\)/,
+    /✓ Authentication flow after 2 attempts \(passed attempt 17\.5s, total 112\.2s\)/,
  );
  assert.match(
    result.stdout,
@@ -1,6 +1,9 @@
 import { test } from 'vitest';
 import assert from 'node:assert/strict';
-import { formatReplayTestProgressEvent } from '../cli-test-progress.ts';
+import {
+  createReplayTestProgressRenderer,
+  formatReplayTestProgressEvent,
+} from '../cli-test-progress.ts';
 import type { RequestProgressEvent } from '../daemon/request-progress.ts';

 function withStreamTty<T>(stream: NodeJS.WriteStream, isTTY: boolean, run: () => T): T {
@@ -70,7 +73,7 @@ test('formatReplayTestProgressEvent renders pass, retry, fail, and skip cases',
        maxAttempts: 2,
        durationMs: 12_345,
      },
-      expected: /^✓ 01-login\.ad \(12\.3s\)$/,
+      expected: /^✓ 01-login\.ad 12\.3s$/,
    },
    {
      event: {
@@ -91,6 +94,7 @@ test('formatReplayTestProgressEvent renders pass, retry, fail, and skip cases',
      event: {
        type: 'replay-test',
        file: '/tmp/03-payment.ad',
+        title: 'Payment flow',
        status: 'fail',
        index: 3,
        total: 3,
@@ -98,11 +102,29 @@ test('formatReplayTestProgressEvent renders pass, retry, fail, and skip cases',
        maxAttempts: 2,
        durationMs: 9_876,
        message: 'assertVisible failed',
+        hint: 'Stop the owning daemon and retry',
        session: 'maestro-test:test:suite:3:attempt-2',
        artifactsDir: '/tmp/replay-suite/payment',
      },
      expected:
-        /^⨯ 03-payment\.ad \(9\.88s\)\n    failed at: assertVisible failed\n    session: maestro-test:test:suite:3:attempt-2\n    artifacts: \/tmp\/replay-suite\/payment$/,
+        /^⨯ Payment flow 9\.88s\n    file: 03-payment\.ad\n    failed at: assertVisible failed\n    hint: Stop the owning daemon and retry\n    session: maestro-test:test:suite:3:attempt-2\n    artifacts: \/tmp\/replay-suite\/payment$/,
+    },
+    {
+      event: {
+        type: 'replay-test',
+        file: '/tmp/05-sharded.ad',
+        title: 'Sharded flow',
+        status: 'pass',
+        index: 5,
+        total: 6,
+        attempt: 1,
+        durationMs: 100,
+        shardIndex: 0,
+        shardCount: 2,
+        deviceId: 'emulator-5554',
+        deviceName: 'Pixel 8',
+      },
+      expected: /^✓ Sharded flow \[1\/2 Pixel 8\] 0\.1s$/,
    },
    {
      event: {
@@ -142,7 +164,37 @@ test('formatReplayTestProgressEvent colors stderr progress rows when stdout is p
      ),
    );

-    assert.equal(line, '\u001B[32m✓\u001B[39m 01-pass.ad (0.01s)');
+    assert.equal(line, '\u001B[32m✓\u001B[39m 01-pass.ad \u001B[33m0.01s\u001B[39m');
+  } finally {
+    if (typeof originalForceColor === 'string') process.env.FORCE_COLOR = originalForceColor;
+    else delete process.env.FORCE_COLOR;
+    if (typeof originalNoColor === 'string') process.env.NO_COLOR = originalNoColor;
+    else delete process.env.NO_COLOR;
+  }
+});
+
+test('createReplayTestProgressRenderer dims live step progress when color is enabled', () => {
+  const originalForceColor = process.env.FORCE_COLOR;
+  const originalNoColor = process.env.NO_COLOR;
+  process.env.FORCE_COLOR = '1';
+  delete process.env.NO_COLOR;
+  try {
+    const renderer = createReplayTestProgressRenderer({ liveProgress: true });
+    const rendered = renderer.render({
+      type: 'replay-test',
+      file: '/tmp/checkout.yaml',
+      title: 'Checkout flow',
+      status: 'progress',
+      index: 1,
+      total: 1,
+      stepIndex: 3,
+      stepTotal: 20,
+    });
+
+    assert.deepEqual(rendered, {
+      text: '\r\u001B[2K⊙ Checkout flow\u001B[2m [3/20]\u001B[22m',
+      newline: false,
+    });
  } finally {
    if (typeof originalForceColor === 'string') process.env.FORCE_COLOR = originalForceColor;
    else delete process.env.FORCE_COLOR;
@@ -175,7 +227,7 @@ test('formatReplayTestProgressEvent colors completed result markers when color i
        attempt: 1,
        durationMs: 10,
      }),
-      '\u001B[32m✓\u001B[39m 01-pass.ad (0.01s)',
+      '\u001B[32m✓\u001B[39m 01-pass.ad \u001B[33m0.01s\u001B[39m',
    );
    assert.equal(
      formatReplayTestProgressEvent({
@@ -188,7 +240,7 @@ test('formatReplayTestProgressEvent colors completed result markers when color i
        attempt: 2,
        durationMs: 30,
      }),
-      '\u001B[33m✓\u001B[39m "Retry flow" in 02-flaky.yml (0.03s)',
+      '\u001B[33m✓\u001B[39m Retry flow \u001B[33m0.03s\u001B[39m',
    );
    const failedLine = formatReplayTestProgressEvent({
      type: 'replay-test',
@@ -201,7 +253,7 @@ test('formatReplayTestProgressEvent colors completed result markers when color i
      durationMs: 5,
      message: 'boom',
    });
-    assert.ok(failedLine?.startsWith('\u001B[31m⨯\u001B[39m "Checkout failure" in 03-fail.ad'));
+    assert.ok(failedLine?.startsWith('\u001B[31m⨯\u001B[39m Checkout failure'));
  } finally {
    if (typeof originalForceColor === 'string') process.env.FORCE_COLOR = originalForceColor;
    else delete process.env.FORCE_COLOR;
@@ -123,7 +123,7 @@ test('readDaemonSocketProgressResponse parses split progress lines before respon
    assert.deepEqual(await responsePromise, { ok: true, data: { via: 'socket-progress' } });
    assert.equal(socket.encoding, 'utf8');
    assert.equal(socket.ended, true);
-    assert.match(stderr, /✓ "Login flow" in 01-login\.ad \(1\.23s\)/);
+    assert.match(stderr, /✓ Login flow 1\.23s/);
  } finally {
    process.stderr.write = originalStderrWrite;
  }
@@ -237,11 +237,9 @@ test('readDaemonSocketProgressResponse rewrites live progress and clears it for
    socket.emit('data', `${progress(3)}\n${progress(4)}\n${pass}\n${responseLine}\n`);

    assert.deepEqual(await responsePromise, { ok: true, data: { via: 'socket-progress' } });
-    assert.ok(stderr.includes('\r\u001B[2K⊙ "Tab View - Co..." in tab-view-coverflow.yml [3/10]'));
-    assert.ok(stderr.includes('\r\u001B[2K⊙ "Tab View - Co..." in tab-view-coverflow.yml [4/10]'));
-    assert.ok(
-      stderr.includes('\r\u001B[2K✓ "Tab View - Coverflow" in tab-view-coverflow.yml (17.8s)\n'),
-    );
+    assert.ok(stderr.includes('\r\u001B[2K⊙ Tab View - Coverflow [3/10]'));
+    assert.ok(stderr.includes('\r\u001B[2K⊙ Tab View - Coverflow [4/10]'));
+    assert.ok(stderr.includes('\r\u001B[2K✓ Tab View - Coverflow 17.8s\n'));
  } finally {
    if (typeof originalCi === 'string') process.env.CI = originalCi;
    else delete process.env.CI;
@@ -308,7 +306,7 @@ test('readDaemonSocketProgressResponse suppresses live progress outside interact
    socket.emit('data', `${progress}\n${pass}\n${responseLine}\n`);

    assert.deepEqual(await responsePromise, { ok: true, data: { via: 'socket-progress' } });
-    assert.equal(stderr, '✓ "Tab View - Coverflow" in tab-view-coverflow.yml (17.8s)\n');
+    assert.equal(stderr, '✓ Tab View - Coverflow 17.8s\n');
  } finally {
    process.stderr.write = originalStderrWrite;
  }
@@ -88,15 +88,15 @@ function formatReplayTestLiveProgressLine(
 ): string {
  const title = event.title?.trim();
  const file = path.basename(event.file);
-  const shardSuffix = formatReplayTestProgressShardSuffix(event);
-  const stepIndex = event.stepIndex ?? 0;
-  const stepTotal = event.stepTotal ?? 0;
-  const suffix = `${shardSuffix} [${stepIndex}/${stepTotal}]`;
+  const useColor = supportsColor(process.stderr);
+  const shardSuffix = formatReplayTestProgressShardSuffix(event, { useColor });
+  const stepSuffix = formatReplayTestLiveProgressStepSuffix(event, { useColor });
+  const suffix = `${shardSuffix}${stepSuffix}`;
  const prefix = '⊙ ';
  if (!title) return trimToColumns(`${prefix}${file}${suffix}`, options.columns);

-  const titlePrefix = `${prefix}"`;
-  const titleSuffix = `" in ${file}${suffix}`;
+  const titlePrefix = prefix;
+  const titleSuffix = suffix;
  const availableTitleColumns = Math.max(
    0,
    resolveColumns(options.columns) - titlePrefix.length - titleSuffix.length,
@@ -105,35 +105,79 @@ function formatReplayTestLiveProgressLine(
  return trimToColumns(`${titlePrefix}${formattedTitle}${titleSuffix}`, options.columns);
 }

+function formatReplayTestLiveProgressStepSuffix(
+  event: ReplayTestCaseProgressEvent,
+  options: { useColor?: boolean } = {},
+): string {
+  const stepIndex = event.stepIndex ?? 0;
+  const stepTotal = event.stepTotal ?? 0;
+  const suffix = ` [${stepIndex}/${stepTotal}]`;
+  return options.useColor ? colorizeProgressMarker(suffix, 'dim') : suffix;
+}
+
 function addReplayTestCaseDetailLines(
  lines: string[],
  event: ReplayTestCaseProgressEvent,
  options: ReplayTestProgressFormatOptions,
 ): void {
-  if (options.verbose && event.status === 'fail') return;
+  if (shouldSuppressReplayTestCaseDetailLines(event, options)) return;
+  const fileLine = replayTestProgressFailureFileLine(event);
+  const messageLine = replayTestProgressMessageLine(event);
+  if (fileLine) lines.push(fileLine);
+  if (messageLine) lines.push(messageLine);
+  appendReplayTestProgressHintLine(lines, event);
+  lines.push(...replayTestProgressFailureContextLines(event));
+}
+
+function shouldSuppressReplayTestCaseDetailLines(
+  event: ReplayTestCaseProgressEvent,
+  options: ReplayTestProgressFormatOptions,
+): boolean {
+  return options.verbose === true && event.status === 'fail';
+}
+
+function replayTestProgressFailureFileLine(event: ReplayTestCaseProgressEvent): string | undefined {
+  return event.status === 'fail' && event.title?.trim()
+    ? `    file: ${path.basename(event.file)}`
+    : undefined;
+}
+
+function replayTestProgressMessageLine(event: ReplayTestCaseProgressEvent): string | undefined {
  const message = event.message?.replace(/\s+/g, ' ').trim();
-  if (message) {
-    lines.push(`    ${event.status === 'fail' ? `failed at: ${message}` : message}`);
-  }
-  if (event.status === 'fail' && !event.retrying) {
-    if (event.session) lines.push(`    session: ${event.session}`);
-    if (event.artifactsDir) lines.push(`    artifacts: ${event.artifactsDir}`);
-  }
+  if (!message) return undefined;
+  return `    ${event.status === 'fail' ? `failed at: ${message}` : message}`;
+}
+
+function appendReplayTestProgressHintLine(
+  lines: string[],
+  event: ReplayTestCaseProgressEvent,
+): void {
+  const hint = event.hint?.replace(/\s+/g, ' ').trim();
+  if (event.status === 'fail' && hint) lines.push(`    hint: ${hint}`);
+}
+
+function replayTestProgressFailureContextLines(event: ReplayTestCaseProgressEvent): string[] {
+  if (event.status !== 'fail' || event.retrying) return [];
+  const lines: string[] = [];
+  if (event.session) lines.push(`    session: ${event.session}`);
+  if (event.artifactsDir) lines.push(`    artifacts: ${event.artifactsDir}`);
+  return lines;
 }

 function formatReplayTestCaseSummaryLine(event: ReplayTestCaseProgressEvent): string {
+  const useColor = supportsColor(process.stderr);
  const statusLabel = formatReplayTestProgressStatusLabel(event);
  const name = formatReplayTestProgressName(event);
-  const shardSuffix = formatReplayTestProgressShardSuffix(event);
+  const shardSuffix = formatReplayTestProgressShardSuffix(event, { useColor });
  const durationSuffix =
-    event.durationMs !== undefined ? ` (${formatReplayProgressDuration(event)})` : '';
+    event.durationMs !== undefined ? ` ${formatReplayProgressDuration(event, { useColor })}` : '';
  return `${statusLabel} ${name}${shardSuffix}${durationSuffix}`;
 }

 function formatReplayTestProgressName(event: ReplayTestCaseProgressEvent): string {
  const title = event.title?.trim();
  const file = path.basename(event.file);
-  return title ? `${JSON.stringify(title)} in ${file}` : file;
+  return title ? title : file;
 }

 function formatReplayTestProgressStatusLabel(event: ReplayTestCaseProgressEvent): string {
@@ -151,15 +195,30 @@ function colorizeProgressMarker(text: string, format: Parameters<typeof colorize
  return colorize(text, format, { validateStream: false });
 }

-function formatReplayTestProgressShardSuffix(event: ReplayTestCaseProgressEvent): string {
+function formatReplayTestProgressShardSuffix(
+  event: ReplayTestCaseProgressEvent,
+  options: { useColor?: boolean } = {},
+): string {
  if (typeof event.shardIndex !== 'number') return '';
  const shardCount = typeof event.shardCount === 'number' ? event.shardCount : '?';
-  const device = typeof event.deviceId === 'string' ? ` ${event.deviceId}` : '';
-  return ` [shard ${event.shardIndex + 1}/${shardCount}${device}]`;
+  const device = replayTestProgressShardDeviceName(event);
+  const suffix = ` [${event.shardIndex + 1}/${shardCount}${device ? ` ${device}` : ''}]`;
+  return options.useColor ? colorizeProgressMarker(suffix, 'dim') : suffix;
 }

-function formatReplayProgressDuration(event: ReplayTestCaseProgressEvent): string {
-  return formatDurationSeconds(event.durationMs ?? 0);
+function replayTestProgressShardDeviceName(event: ReplayTestCaseProgressEvent): string | undefined {
+  const name = event.deviceName?.trim();
+  if (name) return name;
+  const id = event.deviceId?.trim();
+  return id || undefined;
+}
+
+function formatReplayProgressDuration(
+  event: ReplayTestCaseProgressEvent,
+  options: { useColor?: boolean } = {},
+): string {
+  const duration = formatDurationSeconds(event.durationMs ?? 0);
+  return options.useColor ? colorizeProgressMarker(duration, 'yellow') : duration;
 }

 function isReplayTestCompletionProgressEvent(event: ReplayTestCaseProgressEvent): boolean {
@@ -13,6 +13,7 @@ import {
  replayErrorHintLine,
  replayErrorLogLine,
  replayTestDisplayNameWithFile,
+  replayTestFailureFileLine,
  type FailedReplayTestResult,
  type PassedReplayTestResult,
 } from './format.ts';
@@ -37,9 +38,29 @@ function renderReplayTestSummary(

 function formatReplayTestSummaryLine(data: ReplaySuiteResult, flakyCount: number): string {
  const durationMs = typeof data.durationMs === 'number' ? data.durationMs : undefined;
+  const useColor = supportsColor();
+  const passed = formatReplaySummaryPassed(data.passed, { useColor });
+  const total = formatReplaySummaryTotal(data.total, { useColor });
+  const failedSuffix = data.failed > 0 ? `, ${data.failed} failed` : '';
  const flakySuffix = flakyCount > 0 ? `, ${flakyCount} flaky` : '';
-  const durationSuffix = durationMs !== undefined ? ` in ${formatDurationSeconds(durationMs)}` : '';
-  return `Test summary: ${data.passed} passed, ${data.failed} failed${flakySuffix}${durationSuffix}`;
+  const durationSuffix =
+    durationMs !== undefined ? ` in ${formatReplayDuration(durationMs, { useColor })}` : '';
+  return `Test summary: ${passed} ${total}${failedSuffix}${flakySuffix}${durationSuffix}`;
+}
+
+function formatReplaySummaryPassed(passed: number, options: { useColor?: boolean } = {}): string {
+  const text = `${passed} passed`;
+  return options.useColor ? colorize(text, 'green') : text;
+}
+
+function formatReplaySummaryTotal(total: number, options: { useColor?: boolean } = {}): string {
+  const text = `(${total})`;
+  return options.useColor ? colorize(text, 'dim') : text;
+}
+
+function formatReplayDuration(durationMs: number, options: { useColor?: boolean } = {}): string {
+  const duration = formatDurationSeconds(durationMs);
+  return options.useColor ? colorize(duration, 'yellow') : duration;
 }

 function replayFlakyStatusIcon(): string {
@@ -97,6 +118,8 @@ function renderReplayFailureBody(
  context: ReplayTestReporterContext,
  indent: string,
 ): void {
+  const fileLine = replayTestFailureFileLine(result);
+  if (fileLine) context.writeStdout(`${indent}${fileLine}\n`);
  context.writeStdout(`${indent}${result.error?.message ?? 'Unknown test failure'}\n`);
  for (const line of replayFailureConsoleLines(result)) {
    context.writeStdout(`${indent}${line}\n`);
@@ -24,7 +24,7 @@ export function isFlakyReplayTestResult(
 export function replayTestDisplayNameWithFile(result: ReplaySuiteTestResult): string {
  const title = replayTestTitle(result);
  const filename = path.basename(result.file);
-  const base = title && title.length > 0 ? `${JSON.stringify(title)} in ${filename}` : filename;
+  const base = title && title.length > 0 ? title : filename;
  return `${base}${formatReplayTestShardSuffix(result)}`;
 }

@@ -41,6 +41,10 @@ export function replayArtifactsLine(
    : undefined;
 }

+export function replayTestFailureFileLine(result: FailedReplayTestResult): string | undefined {
+  return replayTestTitle(result) ? `file: ${path.basename(result.file)}` : undefined;
+}
+
 export function replayErrorHintLine(error: ReplayTestError): string | undefined {
  return error.hint ? `hint: ${error.hint}` : undefined;
 }
@@ -96,6 +100,10 @@ export function appendReplayTestShardMetadata(
    lines,
    typeof result.deviceId === 'string' ? `deviceId: ${result.deviceId}` : undefined,
  );
+  appendOptionalLine(
+    lines,
+    typeof result.deviceName === 'string' ? `deviceName: ${result.deviceName}` : undefined,
+  );
 }

 export function replayTestWarningLines(result: ReplaySuiteTestResult): string[] {
@@ -132,6 +140,13 @@ function replayTestTitle(result: ReplaySuiteTestResult): string | undefined {
 export function formatReplayTestShardSuffix(result: ReplaySuiteTestResult): string {
  if (!('shardIndex' in result) || typeof result.shardIndex !== 'number') return '';
  const shardCount = typeof result.shardCount === 'number' ? result.shardCount : '?';
-  const device = typeof result.deviceId === 'string' ? ` ${result.deviceId}` : '';
-  return ` [shard ${result.shardIndex + 1}/${shardCount}${device}]`;
+  const device = replayTestShardDeviceName(result);
+  return ` [${result.shardIndex + 1}/${shardCount}${device ? ` ${device}` : ''}]`;
+}
+
+function replayTestShardDeviceName(result: ReplaySuiteTestResult): string | undefined {
+  const name = 'deviceName' in result ? result.deviceName?.trim() : undefined;
+  if (name) return name;
+  const id = 'deviceId' in result ? result.deviceId?.trim() : undefined;
+  return id || undefined;
 }
@@ -32,7 +32,7 @@ const prepareCliSchema = {
  usageOverride: 'prepare ios-runner --platform ios|macos [--timeout <ms>]',
  listUsageOverride: 'prepare',
  helpDescription:
-    'Prepare platform helper infrastructure. ios-runner builds/reuses, starts, and health-checks the XCTest runner so later Apple snapshots and interactions do not pay first-use startup cost. In CI, run it after boot/install and before replay/test; if replay/test starts a separate daemon, run clean:daemon after prepare to release the prepared runner lease. It is not a recovery step for "runner already owned by another agent-device daemon"; stop or clean the owning daemon on the Mac with simulator access instead. Runner build/start output is written to the session runner.log; daemon.log is for daemon lifecycle/startup issues.',
+    'Prepare platform helper infrastructure. ios-runner builds/reuses, starts, and health-checks the XCTest runner so later Apple snapshots and interactions do not pay first-use startup cost. In CI, run it after boot/install and before replay/test; if replay/test starts a separate daemon, stop the prepare daemon before replay/test so it does not keep the prepared runner lease. It is not a recovery step for "runner already owned by another agent-device daemon"; stop the owning daemon on the Mac with simulator access instead. Runner build/start output is written to the session runner.log; daemon.log is for daemon lifecycle/startup issues.',
  summary:
    'Pre-warm platform helpers, especially the iOS/macOS XCTest runner before Apple automation',
  positionalArgs: ['ios-runner'],
@@ -740,7 +740,9 @@ test('parseMaestroReplayFlow parses the test-app Maestro suite fixture', () => {
  assert.deepEqual(
    parsed.actions.map((entry) => entry.command),
    [
+      'open',
      '__maestroAssertVisible',
+      '__maestroScrollUntilVisible',
      '__maestroTapOn',
      '__maestroAssertVisible',
      '__maestroTapOn',
@@ -760,4 +762,5 @@ test('parseMaestroReplayFlow parses the test-app Maestro suite fixture', () => {
      '__maestroAssertVisible',
    ],
  );
+  assert.equal(parsed.actions[0]?.flags.clearAppState, true);
 });
@@ -509,6 +509,12 @@ test('test --shard-all runs each runnable entry on each selected device', async
    'emulator-5556',
    'emulator-5556',
  ]);
+  expect(tests.map((entry) => entry.deviceName)).toEqual([
+    'Pixel 8',
+    'Pixel 8',
+    'Pixel 8 Pro',
+    'Pixel 8 Pro',
+  ]);
  expect(tests.map((entry) => entry.artifactsDir)).toEqual([
    expect.stringContaining(`${path.sep}shard-1${path.sep}01-login`),
    expect.stringContaining(`${path.sep}shard-1${path.sep}02-pay`),
@@ -567,6 +573,9 @@ test('test --shard-split distributes runnable entries by modulo and keeps skips
  expect(tests.filter((entry) => entry.status === 'passed').map((entry) => entry.deviceId)).toEqual(
    ['emulator-5554', 'emulator-5554', 'emulator-5556'],
  );
+  expect(
+    tests.filter((entry) => entry.status === 'passed').map((entry) => entry.deviceName),
+  ).toEqual(['Pixel 8', 'Pixel 8', 'Pixel 8 Pro']);
  expect(invoked.map((req) => req.flags?.serial).sort()).toEqual([
    'emulator-5554',
    'emulator-5554',
@@ -243,6 +243,7 @@ function emitReplayTestRetryProgress(
    durationMs: attempt.durationMs,
    retrying: true,
    message: attempt.response.error.message,
+    hint: attempt.response.error.hint,
    session: attempt.sessionName,
    artifactsDir: context.testArtifactsDir,
    ...replayTestProgressShardMetadata(params.shard),
@@ -322,6 +323,7 @@ function buildReplayTestFailedResult(
    session: outcome.finalSessionName,
    artifactsDir: context.testArtifactsDir,
    message: error.message,
+    hint: error.hint,
    ...replayTestProgressShardMetadata(shard),
  });
  return {
@@ -379,18 +381,19 @@ function readReplayResponseSnapshotDiagnostics(response: DaemonResponse | undefi

 function replayTestShardResultMetadata(
  shard: ReplayTestShardContext | undefined,
-): Pick<ReplaySuiteTestFailed, 'shardIndex' | 'shardCount' | 'deviceId'> {
+): Pick<ReplaySuiteTestFailed, 'shardIndex' | 'shardCount' | 'deviceId' | 'deviceName'> {
  return replayTestProgressShardMetadata(shard);
 }

 function replayTestProgressShardMetadata(
  shard: ReplayTestShardContext | undefined,
-): Pick<ReplaySuiteTestFailed, 'shardIndex' | 'shardCount' | 'deviceId'> {
+): Pick<ReplaySuiteTestFailed, 'shardIndex' | 'shardCount' | 'deviceId' | 'deviceName'> {
  return shard
    ? {
        shardIndex: shard.shardIndex,
        shardCount: shard.shardCount,
        deviceId: shard.device.id,
+        deviceName: shard.device.name,
      }
    : {};
 }
@@ -230,6 +230,7 @@ function buildUnexpectedShardFailure(
    shardIndex: shard.shardIndex,
    shardCount: shard.shardCount,
    deviceId: shard.device.id,
+    deviceName: shard.device.name,
  };
 }

@@ -25,11 +25,13 @@ export type ReplayTestProgressEvent = {
  durationMs?: number;
  retrying?: boolean;
  message?: string;
+  hint?: string;
  session?: string;
  artifactsDir?: string;
  shardIndex?: number;
  shardCount?: number;
  deviceId?: string;
+  deviceName?: string;
 };

 export type CommandProgressEvent = {
@@ -76,6 +76,7 @@ export type ReplaySuiteTestPassed = {
  shardIndex?: number;
  shardCount?: number;
  deviceId?: string;
+  deviceName?: string;
  snapshotDiagnostics?: SnapshotDiagnosticsSummary;
 };

@@ -91,6 +92,7 @@ export type ReplaySuiteTestFailed = {
  shardIndex?: number;
  shardCount?: number;
  deviceId?: string;
+  deviceName?: string;
  snapshotDiagnostics?: SnapshotDiagnosticsSummary;
 };

@@ -772,10 +772,26 @@ test('runner session startup rejects live foreign runner lease', async () => {
      String((thrown as { details?: Record<string, unknown> }).details?.hint),
      /Do not run prepare ios-runner/,
    );
+    assert.match(
+      String((thrown as { details?: Record<string, unknown> }).details?.hint),
+      /^If it is stuck, stop the owning agent-device daemon for AGENT_DEVICE_STATE_DIR='\/tmp\/agent-device-owner' and retry/,
+    );
+    assert.doesNotMatch(
+      String((thrown as { details?: Record<string, unknown> }).details?.hint),
+      /pnpm|clean:daemon/,
+    );
    assert.match(
      String((thrown as { details?: Record<string, unknown> }).details?.hint),
      /PID \d+ with AGENT_DEVICE_STATE_DIR=\/tmp\/agent-device-owner/,
    );
+    assert.doesNotMatch(
+      String((thrown as { details?: Record<string, unknown> }).details?.hint),
+      /AGENT_DEVICE_STATE_DIR=\/tmp\/agent-device-owner\./,
+    );
+    assert.doesNotMatch(
+      String((thrown as { details?: Record<string, unknown> }).details?.hint),
+      /Current daemon state dir/,
+    );
    assert.equal(mockRunCmdBackground.mock.calls.length, 0);
    assert.equal(
      mockRunAppleToolCommand.mock.calls.some((call) => call[0] === 'pkill'),
@@ -825,10 +841,19 @@ test('runner session busy error includes logical lease context after admission',
    deviceKey: device.id,
  });
  assert.match(String(thrown.details?.hint), /five-minute inactivity lease expires/);
+  assert.match(
+    String(thrown.details?.hint),
+    /^If it is stuck, stop the owning agent-device daemon for AGENT_DEVICE_STATE_DIR='\/tmp\/agent-device-owner' and retry/,
+  );
+  assert.doesNotMatch(String(thrown.details?.hint), /pnpm|clean:daemon/);
  assert.match(
    String(thrown.details?.hint),
    /Runner owner: PID \d+ with AGENT_DEVICE_STATE_DIR=\/tmp\/agent-device-owner/,
  );
+  assert.doesNotMatch(
+    String(thrown.details?.hint),
+    /AGENT_DEVICE_STATE_DIR=\/tmp\/agent-device-owner\./,
+  );
  assert.equal(mockRunCmdBackground.mock.calls.length, 0);
 });

@@ -188,26 +188,44 @@ function buildBusyRunnerLeaseHint(
  lease: RunnerLease,
  logicalLeaseContext?: RunnerLogicalLeaseContext,
 ): string {
-  const owner = `PID ${lease.ownerPid}`;
-  const stateDir = lease.ownerStateDir ? ` with AGENT_DEVICE_STATE_DIR=${lease.ownerStateDir}` : '';
-  const currentStateDir = readCurrentStateDir();
-  const current =
-    currentStateDir && currentStateDir !== lease.ownerStateDir
-      ? ` Current daemon state dir is ${currentStateDir}.`
-      : '';
+  const owner = buildRunnerOwnerHint(lease);
+  const cleanup = buildBusyRunnerLeaseCleanupHint(lease);
  if (logicalLeaseContext) {
    return [
-      `The device is busy because another active device lease owns it, or the runner is owned by another daemon/process after lease admission. Runner owner: ${owner}${stateDir}.${current}`,
+      cleanup,
+      owner,
+      'The device is busy because another active device lease owns it, or the runner is owned by another daemon/process after lease admission.',
      'Retry after the owning session closes or after the five-minute inactivity lease expires.',
-      'If this persists after expiry, inspect the runner owner details and clean the stale daemon state on the machine with simulator access.',
    ].join(' ');
  }
  return [
-    `Runner owner details: ${owner}${stateDir}.${current} Retry after the owning runner finishes.`,
-    'Do not run prepare ios-runner from another daemon/client to recover this; a live foreign runner lease cannot be released by the remote client.',
+    cleanup,
+    owner,
+    'If the runner is still active, wait for it to finish. Do not run prepare ios-runner from another daemon/client to recover this.',
  ].join(' ');
 }

+function buildRunnerOwnerHint(lease: RunnerLease): string {
+  const owner = `Runner owner: PID ${lease.ownerPid}`;
+  if (lease.ownerStateDir) return `${owner} with AGENT_DEVICE_STATE_DIR=${lease.ownerStateDir}`;
+  return `${owner}.`;
+}
+
+function buildBusyRunnerLeaseCleanupHint(lease: RunnerLease): string {
+  if (lease.ownerStateDir) {
+    return `If it is stuck, stop the owning agent-device daemon for ${formatEnvAssignment('AGENT_DEVICE_STATE_DIR', lease.ownerStateDir)} and retry.`;
+  }
+  return 'If it is stuck, stop the owning agent-device daemon and retry.';
+}
+
+function formatEnvAssignment(name: string, value: string): string {
+  return `${name}=${shellQuote(value)}`;
+}
+
+function shellQuote(value: string): string {
+  return `'${value.replaceAll("'", "'\\''")}'`;
+}
+
 export async function cleanupOwnedRunnerLease(
  deviceId: string,
  cleanup: RunnerLeaseCleanupAdapter,
@@ -1391,7 +1391,8 @@ test('usageForCommand documents prepare ios-runner', () => {
  assert.match(help, /--timeout <ms>/);
  assert.match(help, /XCTest runner/);
  assert.match(help, /separate daemon/);
-  assert.match(help, /clean:daemon after prepare/);
+  assert.match(help, /stop the prepare daemon before replay\/test/);
+  assert.doesNotMatch(help, /clean:daemon|pnpm/);
  assert.match(
    help,
    /not a recovery step for "runner already owned by another agent-device daemon"/,
@@ -558,8 +558,8 @@ test('sendToDaemon prints replay test progress before the socket response', asyn
    });

    assert.deepEqual(response, { ok: true, data: { via: 'socket' } });
-    assert.match(stderr, /✓ "Login flow" in 01-login\.ad \(1\.23s\)/);
-    assert.equal(stderr.match(/✓ "Login flow" in 01-login\.ad \(1\.23s\)/g)?.length, 1);
+    assert.match(stderr, /✓ Login flow 1\.23s/);
+    assert.equal(stderr.match(/✓ Login flow 1\.23s/g)?.length, 1);
    assert.match(stderr, /steps \(attempt 2\):/);
    assert.match(stderr, /open "Demo" \(line 3, 0\.25s\)/);
    assert.match(stderr, /assertVisible "text=\\"Home\\"" "3000" \(line 4, 0\.75s\)/);
@@ -663,7 +663,7 @@ test('sendToDaemon prints replay test progress before the HTTP NDJSON response',
      assert.deepEqual(response, { ok: true, data: { via: 'http-progress' } });
    });
    assert.deepEqual(seenPaths, ['GET /agent-device/health', 'POST /agent-device/rpc']);
-    assert.match(stderr, /✓ "Payments flow" in 02-payments\.ad \(2\.50s\)/);
+    assert.match(stderr, /✓ Payments flow 2\.50s/);
  } finally {
    (http as unknown as { request: typeof http.request }).request = originalHttpRequest;
    process.stderr.write = originalStderrWrite;
@@ -155,7 +155,7 @@ Bootstrap:
  agent-device prepare ios-runner --platform ios --timeout 240000
  If app id is unknown, plan devices, apps, then open <discovered-app-id>. Discovery is not enough when the task asks to open/start the app.
  Install arguments are app/package id then artifact path. If the task says install, use install; use reinstall only when explicitly requested. Fresh runtime state is open --relaunch after install.
-  In Apple CI, run prepare ios-runner after boot/install and before replay/test. prepare ios-runner builds/reuses the XCTest runner, health-checks it with a lightweight command, and retries one stuck/non-connecting runner launch before the first snapshot pays that setup cost. It is not a recovery step for "runner already owned by another agent-device daemon"; stop or clean the owning daemon on the Mac with simulator access instead. If the replay/test step starts a separate daemon, run clean:daemon after prepare so the prepared runner does not keep a live lease owned by the prepare daemon.
+  In Apple CI, run prepare ios-runner after boot/install and before replay/test. prepare ios-runner builds/reuses the XCTest runner, health-checks it with a lightweight command, and retries one stuck/non-connecting runner launch before the first snapshot pays that setup cost. It is not a recovery step for "runner already owned by another agent-device daemon"; stop the owning daemon on the Mac with simulator access instead. If the replay/test step starts a separate daemon, stop the prepare daemon before replay/test so the prepared runner does not keep a live lease owned by that daemon.
  CI may cache ~/.agent-device/ios-runner/derived with an exact key that includes the agent-device package and Xcode version. Avoid broad restore-key fallbacks; prepare ios-runner already recovers bad restored runner artifacts and one retryable non-connecting runner launch. Runner build/start output is written to the session's runner.log; daemon.log is for daemon lifecycle/startup issues.
  Do not open artifact paths or invent package ids. If apps lookup misses the target and no URL/artifact is provided, ask or stop.