test(skillgym): make SpringBoard widget case assertions sequence-aware

Presence-only matchers let an out-of-order or invented-ref plan pass:
one early unrelated snapshot, a selector-based longpress instead of
the required empty-space coordinate long-press, or a coordinate press
standing in for any ref-driven step could all satisfy the old case.

Add a small sequentialSteps mechanism to makeCase and bind each fresh
snapshot, ref (@e30/@e31/@e35), and coordinate fallback to its
documented SpringBoard step in the order the workflow actually
requires. Verified locally against six synthetic finalOutput plans,
including the reviewer's exact described bad pattern (early snapshot
+ selector-based longpress + invented refs + unrelated coordinate
action), which now fails as expected; two valid orderings (with and
without an extra leading snapshot) still pass.

Author: thymikee

test/skillgym/suites/agent-device-smoke-suite.ts

@@ -451,6 +451,43 @@ const IOS_TEST_APP_DEV_BUILD_OPEN = new RegExp(
   'i',
 );
 
+interface SequentialCommandStep {
+  label: string;
+  regex: RegExp;
+}
+
+/**
+ * Enforces that each step's pattern appears, in order, at or after the position where the
+ * previous step matched — unlike the presence-only OutputMatcher checks above, this catches a
+ * plan that satisfies every individual requirement out of order (for example an early,
+ * unrelated snapshot standing in for the one that must follow a specific action).
+ */
+function assertSequentialCommandSteps(
+  finalOutput: string,
+  steps: readonly SequentialCommandStep[],
+  caseLabel: string,
+) {
+  const lines = normalizedFinalOutput(finalOutput)
+    .split('\n')
+    .map((line) => line.trim())
+    .filter(Boolean);
+  let cursor = 0;
+  for (const step of steps) {
+    let matchIndex = -1;
+    for (let i = cursor; i < lines.length; i++) {
+      if (step.regex.test(lines[i]!)) {
+        matchIndex = i;
+        break;
+      }
+    }
+    assert.ok(
+      matchIndex !== -1,
+      `${caseLabel}: expected step "${step.label}" at or after position ${cursor}. Observed commands:\n${lines.join('\n')}`,
+    );
+    cursor = matchIndex + 1;
+  }
+}
+
 function makeCase(options: {
   id: string;
   contract: string[];
@@ -461,6 +498,7 @@ function makeCase(options: {
   strictFinalOutput?: boolean;
   allowOnlyLocalCliHelpCommands?: boolean;
   finalOutputInstructions?: string;
+  sequentialSteps?: readonly SequentialCommandStep[];
 }): Case {
   return {
     id: options.id,
@@ -484,6 +522,9 @@ function makeCase(options: {
       if (options.allowOnlyLocalCliHelpCommands) {
         assertOnlyLocalCliHelpCommands(report);
       }
+      if (options.sequentialSteps) {
+        assertSequentialCommandSteps(ctx.finalOutput(), options.sequentialSteps, options.id);
+      }
     },
   };
 }
@@ -2641,24 +2682,64 @@ Do not combine final commands with shell operators such as &&, ||, pipes, or sem
       'SpringBoard control labels vary by iOS version and locale; do not assume fixed English strings such as "Add Widget" or "Edit" — use the refs above, sourced from a fresh snapshot -i, instead',
     ],
     task: 'Plan the commands to open SpringBoard, enter edit mode, open the widget gallery with @e30, search it for "Calendar" using @e31, select the result using the documented coordinate fallback for the unlabeled search-result row, place the widget with @e35, reopen Agent Device Tester to resume app automation, and close the session.',
-    outputs: [
-      plannedCommand('open com.apple.springboard'),
-      /--platform ios/i,
-      /snapshot\b[^\n]*-i\b/i,
-      plannedCommand('longpress'),
-      /(?:^|\n)(?:agent-device\s+)?(?:press|click)\s+@e30\b/i,
-      /(?:^|\n)(?:agent-device\s+)?fill\s+@e31\b[^\n]*Calendar/i,
-      /(?:^|\n)(?:agent-device\s+)?screenshot\b/i,
-      RAW_COORDINATE_TARGET,
-      /(?:^|\n)(?:agent-device\s+)?(?:press|click)\s+@e35\b/i,
-      plannedCommand('open com.callstack.agentdevicelab'),
-      plannedCommand('close'),
-    ],
+    outputs: [/--platform ios/i],
     forbiddenOutputs: [
       plannedCommandAlternatives(['widget', 'widget add']),
       /widget\s+add/i,
       /label=["'](?:Add Widget|Edit|Done)["']/i,
     ],
+    // Presence-only matchers would let an out-of-order or invented-ref plan pass (an early
+    // unrelated snapshot, a selector-based longpress instead of the required empty-space
+    // coordinate long-press, or a coordinate press standing in for any of the ref-driven
+    // steps). Bind each fresh snapshot, ref, and coordinate fallback to its documented
+    // SpringBoard step in the order the workflow actually requires.
+    sequentialSteps: [
+      {
+        label: 'open SpringBoard',
+        regex: /^(?:agent-device\s+)?open\s+com\.apple\.springboard\b/i,
+      },
+      {
+        label: 'empty-space long-press to enter edit mode (raw coordinates, not a selector)',
+        regex: /^(?:agent-device\s+)?longpress\s+-?\d+(?:\.\d+)?\s+-?\d+(?:\.\d+)?\b/i,
+      },
+      {
+        label: 'fresh snapshot -i exposing @e30',
+        regex: /^(?:agent-device\s+)?snapshot\b.*-i\b/i,
+      },
+      {
+        label: 'press/click @e30 to open the widget gallery',
+        regex: /^(?:agent-device\s+)?(?:press|click)\s+@e30\b/i,
+      },
+      {
+        label: 'fresh snapshot -i exposing @e31',
+        regex: /^(?:agent-device\s+)?snapshot\b.*-i\b/i,
+      },
+      {
+        label: 'fill @e31 with "Calendar" to search the gallery',
+        regex: /^(?:agent-device\s+)?fill\s+@e31\b.*calendar/i,
+      },
+      {
+        label: 'screenshot before the coordinate fallback',
+        regex: /^(?:agent-device\s+)?screenshot\b/i,
+      },
+      {
+        label: 'raw-coordinate press on the unlabeled search-result row',
+        regex: /^(?:agent-device\s+)?(?:press|click)\s+-?\d+(?:\.\d+)?\s+-?\d+(?:\.\d+)?\b/i,
+      },
+      {
+        label: 'fresh snapshot -i exposing @e35',
+        regex: /^(?:agent-device\s+)?snapshot\b.*-i\b/i,
+      },
+      {
+        label: 'press/click @e35 to place the widget',
+        regex: /^(?:agent-device\s+)?(?:press|click)\s+@e35\b/i,
+      },
+      {
+        label: 'reopen Agent Device Tester',
+        regex: /^(?:agent-device\s+)?open\s+com\.callstack\.agentdevicelab\b/i,
+      },
+      { label: 'close the session', regex: /^(?:agent-device\s+)?close\b/i },
+    ],
   }),
   makeCase({
     id: 'replay-maintenance-update',