🤖 test: add regression tests for step providerMetadata and cache creation

Cover the fix to prevent regression:
- StreamingMessageAggregator: test step providerMetadata storage/retrieval/clear
- displayUsage: test cacheCreationInputTokens extraction from providerMetadata

_Generated with mux_

Author: ethanndickson

src/browser/utils/messages/StreamingMessageAggregator.test.ts

@@ -517,6 +517,40 @@ describe("StreamingMessageAggregator", () => {
       expect(aggregator.getActiveStreamCumulativeProviderMetadata("msg-1")).toBeUndefined();
     });
 
+    test("stores and retrieves step providerMetadata for cache creation display", () => {
+      const aggregator = new StreamingMessageAggregator(TEST_CREATED_AT);
+
+      aggregator.handleUsageDelta({
+        type: "usage-delta",
+        workspaceId: "ws-1",
+        messageId: "msg-1",
+        usage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
+        cumulativeUsage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
+        providerMetadata: {
+          anthropic: { cacheCreationInputTokens: 800 },
+        },
+      });
+
+      expect(aggregator.getActiveStreamStepProviderMetadata("msg-1")).toEqual({
+        anthropic: { cacheCreationInputTokens: 800 },
+      });
+    });
+
+    test("step providerMetadata is undefined when not provided", () => {
+      const aggregator = new StreamingMessageAggregator(TEST_CREATED_AT);
+
+      aggregator.handleUsageDelta({
+        type: "usage-delta",
+        workspaceId: "ws-1",
+        messageId: "msg-1",
+        usage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
+        cumulativeUsage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
+        // No providerMetadata
+      });
+
+      expect(aggregator.getActiveStreamStepProviderMetadata("msg-1")).toBeUndefined();
+    });
+
     test("clearTokenState clears all usage tracking (step, cumulative, metadata)", () => {
       const aggregator = new StreamingMessageAggregator(TEST_CREATED_AT);
 
@@ -526,18 +560,21 @@ describe("StreamingMessageAggregator", () => {
         messageId: "msg-1",
         usage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
         cumulativeUsage: { inputTokens: 1000, outputTokens: 50, totalTokens: 1050 },
+        providerMetadata: { anthropic: { cacheCreationInputTokens: 300 } },
         cumulativeProviderMetadata: { anthropic: { cacheCreationInputTokens: 500 } },
       });
 
       // All should be defined
       expect(aggregator.getActiveStreamUsage("msg-1")).toBeDefined();
+      expect(aggregator.getActiveStreamStepProviderMetadata("msg-1")).toBeDefined();
       expect(aggregator.getActiveStreamCumulativeUsage("msg-1")).toBeDefined();
       expect(aggregator.getActiveStreamCumulativeProviderMetadata("msg-1")).toBeDefined();
 
       aggregator.clearTokenState("msg-1");
 
       // All should be cleared
       expect(aggregator.getActiveStreamUsage("msg-1")).toBeUndefined();
+      expect(aggregator.getActiveStreamStepProviderMetadata("msg-1")).toBeUndefined();
       expect(aggregator.getActiveStreamCumulativeUsage("msg-1")).toBeUndefined();
       expect(aggregator.getActiveStreamCumulativeProviderMetadata("msg-1")).toBeUndefined();
     });

src/common/types/stream.ts

@@ -142,14 +142,14 @@ export interface UsageDeltaEvent {
   type: "usage-delta";
   workspaceId: string;
   messageId: string;
-  // This step's usage (inputTokens = current context size, for context window display)
+
+  // Step-level: this step only (for context window display)
   usage: LanguageModelV2Usage;
-  // Cumulative usage across all steps so far (for live cost display)
+  providerMetadata?: Record<string, unknown>;
+
+  // Cumulative: sum across all steps (for live cost display)
   cumulativeUsage: LanguageModelV2Usage;
-  // Cumulative provider metadata across all steps (for live cost display with cache tokens)
   cumulativeProviderMetadata?: Record<string, unknown>;
-  // This step's provider metadata (for context window cache display)
-  providerMetadata?: Record<string, unknown>;
 }
 
 export type AIServiceEvent =

src/common/utils/tokens/displayUsage.test.ts

@@ -276,4 +276,68 @@ describe("createDisplayUsage", () => {
     expect(result!.input.tokens).toBe(1000);
     expect(result!.cached.tokens).toBe(0);
   });
+
+  describe("Anthropic cache creation tokens from providerMetadata", () => {
+    // Cache creation tokens are Anthropic-specific and only available in
+    // providerMetadata.anthropic.cacheCreationInputTokens, not in LanguageModelV2Usage.
+    // This is critical for liveUsage display during streaming.
+
+    test("extracts cacheCreationInputTokens from providerMetadata", () => {
+      const usage: LanguageModelV2Usage = {
+        inputTokens: 1000,
+        outputTokens: 50,
+        totalTokens: 1050,
+      };
+
+      const result = createDisplayUsage(usage, "anthropic:claude-sonnet-4-20250514", {
+        anthropic: { cacheCreationInputTokens: 800 },
+      });
+
+      expect(result).toBeDefined();
+      expect(result!.cacheCreate.tokens).toBe(800);
+    });
+
+    test("cacheCreate is 0 when providerMetadata is undefined", () => {
+      const usage: LanguageModelV2Usage = {
+        inputTokens: 1000,
+        outputTokens: 50,
+        totalTokens: 1050,
+      };
+
+      const result = createDisplayUsage(usage, "anthropic:claude-sonnet-4-20250514");
+
+      expect(result).toBeDefined();
+      expect(result!.cacheCreate.tokens).toBe(0);
+    });
+
+    test("cacheCreate is 0 when anthropic metadata lacks cacheCreationInputTokens", () => {
+      const usage: LanguageModelV2Usage = {
+        inputTokens: 1000,
+        outputTokens: 50,
+        totalTokens: 1050,
+      };
+
+      const result = createDisplayUsage(usage, "anthropic:claude-sonnet-4-20250514", {
+        anthropic: { someOtherField: 123 },
+      });
+
+      expect(result).toBeDefined();
+      expect(result!.cacheCreate.tokens).toBe(0);
+    });
+
+    test("handles gateway Anthropic model with cache creation", () => {
+      const usage: LanguageModelV2Usage = {
+        inputTokens: 2000,
+        outputTokens: 100,
+        totalTokens: 2100,
+      };
+
+      const result = createDisplayUsage(usage, "mux-gateway:anthropic/claude-sonnet-4-5", {
+        anthropic: { cacheCreationInputTokens: 1500 },
+      });
+
+      expect(result).toBeDefined();
+      expect(result!.cacheCreate.tokens).toBe(1500);
+    });
+  });
 });

src/node/services/streamManager.ts

@@ -969,10 +969,12 @@ export class StreamManager extends EventEmitter {
               type: "usage-delta",
               workspaceId: workspaceId as string,
               messageId: streamInfo.messageId,
-              usage: finishStepPart.usage, // For context window display
-              cumulativeUsage: streamInfo.cumulativeUsage, // For live cost display
-              cumulativeProviderMetadata: streamInfo.cumulativeProviderMetadata, // For live cache costs
-              providerMetadata: finishStepPart.providerMetadata, // For context window cache display
+              // Step-level (for context window display)
+              usage: finishStepPart.usage,
+              providerMetadata: finishStepPart.providerMetadata,
+              // Cumulative (for live cost display)
+              cumulativeUsage: streamInfo.cumulativeUsage,
+              cumulativeProviderMetadata: streamInfo.cumulativeProviderMetadata,
             };
             this.emit("usage-delta", usageEvent);
             break;