fix(ai): tolerate null plan tasks in the readiness scan

AUTONOMOUS_STATE.json

@@ -7,7 +7,7 @@
     "asset-curator",
     "qa-playtester"
   ],
-  "last_cycle": 177,
+  "last_cycle": 178,
   "completed_targets": [
     {
       "cycle": 75,
@@ -1036,6 +1036,12 @@
       "target": "Actionable hints for HTTP 429 and 502/503 provider failures",
       "metric_delta": "The objective names 429/502/timeout as errors to surface gracefully; 429 and 5xx had no guidance (a 429 said only 'AI Provider Error (429): ...'). Added status hints: 429 -> rate limiting / wait-and-retry, 502|503 -> temporarily unavailable / retry or check base URL. Deliberately left 500 untouched so the existing 'unmatched provider error' test still holds as evidence of pass-through. 2 new tests red pre-fix; mutation-checked (neutralizing the hint fails both). One wording fix: the first 429 hint said 'too frequent' while the test matched /rate/, so the hint now says 'rate limiting requests'. 1599/1599 across 80 suites (was 1597), eslint+prettier clean, divergence clean.",
       "commit": "feat(ai): hint at rate limiting and provider outages"
+    },
+    {
+      "cycle": 178,
+      "target": "Null-tolerate the plan readiness scan (model-authored plan tasks)",
+      "metric_delta": "PlanStore.areTaskDependenciesSatisfied and getNextReadyTask read otherTask.status / task.status over plan tasks, which come from model-authored function_call_output JSON and persisted output - neither shape-validated per element. A null entry crashed the readiness scan while the studio finalized a child (the exact pass that flips the next task to in_progress). Both scans now skip non-objects. 1 test covering both, each guard mutation-checked individually (reverting either reproduces the TypeError). 1600/1600 across 80 suites (was 1599), eslint+prettier clean, divergence clean.",
+      "commit": "fix(ai): tolerate null plan tasks in the readiness scan"
     }
   ],
   "backlog": [

newIDE/app/src/AiGeneration/Studio/PlanStore.js

@@ -71,7 +71,14 @@ export const areTaskDependenciesSatisfied = (
 ): boolean => {
   const doneIds = new Set(
     tasks
-      .filter(otherTask => otherTask.status === 'done')
+      // A null/scalar entry is reachable: plan tasks come from model-authored
+      // JSON and persisted output, neither shape-validated per element.
+      .filter(
+        otherTask =>
+          otherTask &&
+          typeof otherTask === 'object' &&
+          otherTask.status === 'done'
+      )
       .map(otherTask => otherTask.id)
   );
   return (task.dependsOn || []).every(dependencyId =>
@@ -85,7 +92,10 @@ export const getNextReadyTask = (
 ): ?StudioPlanTask =>
   tasks.find(
     task =>
-      task.status === 'pending' && areTaskDependenciesSatisfied(task, tasks)
+      task &&
+      typeof task === 'object' &&
+      task.status === 'pending' &&
+      areTaskDependenciesSatisfied(task, tasks)
   ) || null;
 
 /**

newIDE/app/src/AiGeneration/Studio/PlanStore.spec.js

@@ -129,6 +129,20 @@ describe('PlanStore', () => {
       expect(readyA && readyA.id).toBe('a');
     });
 
+    it('tolerates a null hole in the task list', () => {
+      // Plan tasks come from model-authored JSON (function_call_output) and
+      // from persisted output, neither shape-validated per element: a null
+      // entry crashed the readiness scan while the studio finalized a child.
+      const ready = makeTask({ id: 'b', status: 'pending' });
+      expect(getNextReadyTask(([null, ready]: any))).toBe(ready);
+      expect(
+        areTaskDependenciesSatisfied(
+          makeTask({ dependsOn: ['a'] }),
+          ([null, { ...makeTask({ id: 'a' }), status: 'done' }]: any)
+        )
+      ).toBe(true);
+    });
+
     it('returns null when nothing is ready', () => {
       expect(
         getNextReadyTask([