test: repair full-suite drift against current runtime contracts

Partial module mocks gain the attachGenAiAttributes export the chat terminals now call, and the api/runtime suites reset the module-global provider-health circuit between tests so one intentional failure burst cannot open the breaker for the rest of the file. The Anthropic transport tests pin maxRetries to 1 (the retry-era default re-reads a one-shot mock Response and masks the intended error) and the thinking-replay coverage now exercises both sides of the per-model split: strip on non-retaining models, verbatim replay on retaining ones. The OpenRouter session tests spy on the axios client seam instead of global fetch, mapOpenRouterUsage expectations carry inclusiveInputTokens, orchestration default-model pins move to claude-sonnet-4-6, assembleForPrompt expectations follow the maxTierRank option and embed dedup, the ExtensionLoader child_process mock exports execFile, and the container-repair fixture uses correctly escaped string-encoded JSON.

Author: jddunn

src/api/runtime/__tests__/agentPromptEngine.test.ts

@@ -3,6 +3,7 @@
  * Tests for PromptEngine, Memory, and Skills integration in agent().
  */
 import { describe, expect, it, vi, beforeEach } from 'vitest';
+import { globalLLMProviderHealth } from '../../../core/safety/LLMProviderHealthRegistry.js';
 
 const mockGenerateCompletion = vi.hoisted(() =>
   vi.fn().mockResolvedValue({
@@ -38,9 +39,16 @@ vi.mock('../../model.js', () => ({
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue({}),
 }));
 
+// The provider-health circuit is module-global state: one test's intentional
+// failure burst must not open the breaker for the rest of the file.
+beforeEach(() => {
+  globalLLMProviderHealth.reset();
+});
+
 vi.mock('../../../evaluation/observability/otel.js', () => ({
   withAgentOSSpan: vi.fn((_name: string, _attrs: unknown, fn?: Function) => {
     const callback = fn ?? _attrs;

src/api/runtime/__tests__/analyzeVideo.test.ts

@@ -66,6 +66,7 @@ vi.mock('../../../io/hearing/providers/OpenAIWhisperSpeechToTextProvider.js', ()
 // Mock observability to avoid OTel dependencies
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue(undefined),
 }));
 

src/api/runtime/__tests__/generateMusic.test.ts

@@ -82,6 +82,7 @@ vi.mock('../../../media/audio/index.js', () => {
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue(undefined),
 }));
 

src/api/runtime/__tests__/generateObject.test.ts

@@ -911,7 +911,7 @@ describe('string-encoded container repair', () => {
       ),
     });
     hoisted.generateCompletion.mockResolvedValue(
-      mockResponse('{"verdicts":"[{\"trackId\":\"t\",\"verdict\":\"purple\"}]"}'),
+      mockResponse('{"verdicts":"[{\\"trackId\\":\\"t\\",\\"verdict\\":\\"purple\\"}]"}'),
     );
 
     const err = await generateObject({

src/api/runtime/__tests__/generateSFX.test.ts

@@ -82,6 +82,7 @@ vi.mock('../../../media/audio/index.js', () => {
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue(undefined),
 }));
 

src/api/runtime/__tests__/generateVideo.test.ts

@@ -80,6 +80,7 @@ vi.mock('../../../io/media/video/index.js', () => {
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue(undefined),
 }));
 

src/api/runtime/__tests__/generationHooks.test.ts

@@ -3,6 +3,7 @@
  * Tests for generation lifecycle hooks on generateText/streamText.
  */
 import { describe, expect, it, vi, beforeEach } from 'vitest';
+import { globalLLMProviderHealth } from '../../../core/safety/LLMProviderHealthRegistry.js';
 
 const mockGenerateCompletion = vi.hoisted(() =>
   vi.fn().mockResolvedValue({
@@ -42,9 +43,16 @@ vi.mock('../../model.js', () => ({
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue({}),
 }));
 
+// The provider-health circuit is module-global state: one test's intentional
+// failure burst must not open the breaker for the rest of the file.
+beforeEach(() => {
+  globalLLMProviderHealth.reset();
+});
+
 vi.mock('../../../evaluation/observability/otel.js', () => ({
   withAgentOSSpan: vi.fn((_name: string, _attrs: unknown, fn?: Function) => {
     const callback = fn ?? _attrs;

src/api/runtime/__tests__/modelRouter.test.ts

@@ -3,6 +3,7 @@
  * Tests for ModelRouter integration into generateText/streamText/agent.
  */
 import { describe, expect, it, vi, beforeEach } from 'vitest';
+import { globalLLMProviderHealth } from '../../../core/safety/LLMProviderHealthRegistry.js';
 
 const mockGenerateCompletion = vi.hoisted(() =>
   vi.fn().mockResolvedValue({
@@ -50,9 +51,16 @@ vi.mock('../../model.js', () => ({
 
 vi.mock('../../observability.js', () => ({
   attachUsageAttributes: vi.fn(),
+  attachGenAiAttributes: vi.fn(),
   toTurnMetricUsage: vi.fn().mockReturnValue({}),
 }));
 
+// The provider-health circuit is module-global state: one test's intentional
+// failure burst must not open the breaker for the rest of the file.
+beforeEach(() => {
+  globalLLMProviderHealth.reset();
+});
+
 vi.mock('../../../evaluation/observability/otel.js', () => ({
   withAgentOSSpan: vi.fn((_name: string, _attrs: unknown, fn?: Function) => {
     const callback = fn ?? _attrs;

src/cognition/memory/__tests__/AgentMemory.mood-args.test.ts

@@ -51,12 +51,12 @@ describe('AgentMemory.getContext mood arg', () => {
   it('omitted → NEUTRAL passed to assembleForPrompt', async () => {
     const { mem, manager } = cognitiveMemoryWithManagerSpy();
     await mem.getContext('q', { tokenBudget: 1000 });
-    expect(manager.assembleForPrompt).toHaveBeenCalledWith('q', 1000, { valence: 0, arousal: 0, dominance: 0 });
+    expect(manager.assembleForPrompt).toHaveBeenCalledWith('q', 1000, { valence: 0, arousal: 0, dominance: 0 }, { maxTierRank: undefined });
   });
   it('supplied currentMood passes through; default budget 2000 preserved', async () => {
     const { mem, manager } = cognitiveMemoryWithManagerSpy();
     await mem.getContext('q', { currentMood: { valence: 0.4, arousal: 0.2, dominance: 0 } });
-    expect(manager.assembleForPrompt).toHaveBeenCalledWith('q', 2000, { valence: 0.4, arousal: 0.2, dominance: 0 });
+    expect(manager.assembleForPrompt).toHaveBeenCalledWith('q', 2000, { valence: 0.4, arousal: 0.2, dominance: 0 }, { maxTierRank: undefined });
   });
 });
 

src/cognition/memory/__tests__/assemble-for-prompt-concurrency.test.ts

@@ -118,17 +118,27 @@ describe('assembleForPrompt concurrency', () => {
       return { documents: [] };
     });
 
+    const checkSpy = vi.spyOn(
+      (manager as unknown as { prospective: { check: (...a: unknown[]) => Promise<unknown[]> } })
+        .prospective,
+      'check',
+    );
+
     const pending = manager.assembleForPrompt('what did we plan for friday', 2048, MOOD);
 
     // Let the event loop drain everything that isn't blocked on the latch.
     await new Promise((resolve) => setTimeout(resolve, 25));
 
     // Serial implementation: neither fires until retrieve() resolves.
-    // Concurrent implementation: both have already started.
+    // Concurrent implementation: both have already started. Identical query
+    // embeds can be served from the embedding cache (one manager hit shared
+    // by the store and prospective paths), so assert the prospective stage
+    // itself ran during the latch window instead of counting embed calls.
     expect(persistentRead).toHaveBeenCalledTimes(1);
+    expect(checkSpy).toHaveBeenCalledTimes(1);
     expect(
       mocks.mockEmbeddingManager.generateEmbeddings.mock.calls.length,
-    ).toBeGreaterThanOrEqual(2); // store's query embed + prospective's query embed
+    ).toBeGreaterThanOrEqual(1);
 
     releaseQuery();
     const out = await pending;