@vellumai/assistant 0.7.0 → 0.7.2

package/src/__tests__/conversation-surfaces-standalone.test.ts

@@ -1,4 +1,12 @@
-import { afterEach, beforeEach, describe, expect, test } from "bun:test";
+import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
+
+import type { ServerMessage } from "../daemon/message-protocol.js";
+import type { InteractiveUiResult } from "../runtime/interactive-ui.js";
+
+let broadcastImpl: (msg: ServerMessage) => void = () => {};
+mock.module("../runtime/assistant-event-hub.js", () => ({
+  broadcastMessage: (msg: ServerMessage) => broadcastImpl(msg),
+}));
 
 import {
   canShowInteractiveUi,
@@ -7,8 +15,6 @@ import {
   showStandaloneSurface,
   type SurfaceConversationContext,
 } from "../daemon/conversation-surfaces.js";
-import type { ServerMessage } from "../daemon/message-protocol.js";
-import type { InteractiveUiResult } from "../runtime/interactive-ui.js";
 
 // ── Helpers ──────────────────────────────────────────────────────────
 
@@ -28,6 +34,7 @@ function createMockContext(
   enqueuedMessages: Array<{ content: string; requestId: string }>;
 } {
   const sentMessages: ServerMessage[] = [];
+  broadcastImpl = (msg: ServerMessage) => sentMessages.push(msg);
   const enqueuedMessages: Array<{ content: string; requestId: string }> = [];
 
   return {
@@ -44,7 +51,6 @@ function createMockContext(
       emit: () => {},
     },
     sendToClient: (msg: ServerMessage) => sentMessages.push(msg),
-    broadcastToAllClients: (msg: ServerMessage) => sentMessages.push(msg),
     pendingSurfaceActions: new Map(),
     lastSurfaceAction: new Map(),
     surfaceState: new Map(),
@@ -58,8 +64,9 @@ function createMockContext(
     hasNoClient: overrides?.hasNoClient ?? false,
     isProcessing: () => false,
     enqueueMessage: (content, _attachments, _onEvent, requestId) => {
-      enqueuedMessages.push({ content, requestId });
-      return { queued: false, requestId };
+      const resolvedId = requestId ?? "mock-request-id";
+      enqueuedMessages.push({ content, requestId: resolvedId });
+      return { queued: false, requestId: resolvedId };
     },
     getQueueDepth: () => 0,
     processMessage: async () => "msg-id",
@@ -291,9 +298,8 @@ describe("showStandaloneSurface", () => {
 
   test("timeout still resolves when emit throws", async () => {
     const ctx = createMockContext();
-    // Override broadcastToAllClients to throw on ui_surface_complete
     let showEmitted = false;
-    ctx.broadcastToAllClients = (msg: ServerMessage) => {
+    broadcastImpl = (msg: ServerMessage) => {
       const type = (msg as unknown as Record<string, unknown>).type;
       if (type === "ui_surface_show") {
         showEmitted = true;
@@ -518,14 +524,13 @@ describe("standalone surface cleanup on conversation dispose", () => {
 
     // Simulate conversation dispose — cancel all pending with dismiss
     // notifications, matching the Conversation.dispose() implementation.
-    const emit = ctx.broadcastToAllClients ?? ctx.sendToClient.bind(ctx);
     for (const [surfaceId, entry] of ctx.pendingStandaloneSurfaces!) {
       clearTimeout(entry.timer);
-      emit({
+      broadcastImpl({
         type: "ui_surface_dismiss",
         conversationId: ctx.conversationId,
         surfaceId,
-      });
+      } as ServerMessage);
       entry.resolve({ status: "cancelled", surfaceId });
     }
     ctx.pendingStandaloneSurfaces!.clear();
@@ -568,14 +573,13 @@ describe("standalone surface cleanup on conversation dispose", () => {
     ctx.sentMessages.length = 0;
 
     // Simulate the dispose path (matching Conversation.dispose())
-    const emit = ctx.broadcastToAllClients ?? ctx.sendToClient.bind(ctx);
     for (const [surfaceId, entry] of ctx.pendingStandaloneSurfaces!) {
       clearTimeout(entry.timer);
-      emit({
+      broadcastImpl({
         type: "ui_surface_dismiss",
         conversationId: ctx.conversationId,
         surfaceId,
-      });
+      } as ServerMessage);
       entry.resolve({ status: "cancelled", surfaceId });
     }
     ctx.pendingStandaloneSurfaces!.clear();

package/src/__tests__/conversation-surfaces-state-update.test.ts

@@ -46,8 +46,9 @@ function makeContext(opts?: {
     currentTurnSurfaces: [],
     isProcessing: () => false,
     enqueueMessage: (content, _attachments, _onEvent, requestId) => {
-      enqueueCalls.push({ content, requestId });
-      return { queued: false, requestId };
+      const resolvedId = requestId ?? "mock-request-id";
+      enqueueCalls.push({ content, requestId: resolvedId });
+      return { queued: false, requestId: resolvedId };
     },
     getQueueDepth: () => 0,
     processMessage: async (content, _attachments, _onEvent, requestId) => {

package/src/__tests__/conversation-tool-setup-app-refresh.test.ts

@@ -25,6 +25,11 @@ import type { ToolExecutionResult } from "../tools/types.js";
 
 const refreshSpy = mock(() => {});
 const updatePublishedSpy = mock(() => Promise.resolve());
+const broadcastSpy = mock(() => {});
+
+mock.module("../runtime/assistant-event-hub.js", () => ({
+  broadcastMessage: broadcastSpy,
+}));
 
 // Mock session-surfaces so refreshSurfacesForApp is captured
 mock.module("../daemon/conversation-surfaces.js", () => ({
@@ -120,6 +125,7 @@ const noopLifecycleHandler = mock(() => {});
 describe("session-tool-setup app refresh side effects", () => {
   beforeEach(() => {
     refreshSpy.mockClear();
+    broadcastSpy.mockClear();
     updatePublishedSpy.mockClear();
   });
 
@@ -132,7 +138,6 @@ describe("session-tool-setup app refresh side effects", () => {
         content: '{"id":"app-1"}',
         isError: false,
       });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -140,7 +145,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_refresh", { app_id: "app-1" });
@@ -153,7 +157,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("broadcasts app_files_changed with correct appId", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "{}", isError: false });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -161,7 +164,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_refresh", { app_id: "app-42" });
@@ -183,7 +185,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        mock(() => {}),
       );
 
       await toolFn("app_refresh", { app_id: "app-publish" });
@@ -202,7 +203,6 @@ describe("session-tool-setup app refresh side effects", () => {
         content: "Error: not found",
         isError: true,
       });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -210,7 +210,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_refresh", { app_id: "app-err" });
@@ -223,7 +222,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("skips side effects when app_id is missing", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "{}", isError: false });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -231,7 +229,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_refresh", {});
@@ -250,7 +247,6 @@ describe("session-tool-setup app refresh side effects", () => {
         content: JSON.stringify({ id: "new-app-1", name: "My App" }),
         isError: false,
       });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -258,7 +254,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_create", { name: "My App", html: "<h1>hi</h1>" });
@@ -273,7 +268,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("skips side effects when app_create result is an error", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "Error", isError: true });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -281,7 +275,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_create", { name: "Bad", html: "" });
@@ -305,7 +298,6 @@ describe("session-tool-setup app refresh side effects", () => {
         }),
         isError: false,
       });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -313,7 +305,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_create", { name: "Busted", html: "" });
@@ -334,7 +325,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("broadcasts app_files_changed after app_delete", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "{}", isError: false });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -342,7 +332,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_delete", { app_id: "del-app-1" });
@@ -357,7 +346,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("skips side effects when app_delete result is an error", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "Error", isError: true });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -365,7 +353,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       await toolFn("app_delete", { app_id: "del-err" });
@@ -384,7 +371,6 @@ describe("session-tool-setup app refresh side effects", () => {
       // hooks as core tools.
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "{}", isError: false });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -392,12 +378,12 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       // Simulate calling app_refresh by name (as the agent loop does)
       for (const toolName of ["app_refresh"]) {
         refreshSpy.mockClear();
+    broadcastSpy.mockClear();
         broadcastSpy.mockClear();
         updatePublishedSpy.mockClear();
 
@@ -418,7 +404,6 @@ describe("session-tool-setup app refresh side effects", () => {
     test("other tool names do not trigger app refresh side effects", async () => {
       const ctx = makeCtx();
       const executor = makeFakeExecutor({ content: "{}", isError: false });
-      const broadcastSpy = mock(() => {});
 
       const toolFn = createToolExecutor(
         executor as unknown as ToolExecutor,
@@ -426,7 +411,6 @@ describe("session-tool-setup app refresh side effects", () => {
         noopSecretPrompter,
         ctx,
         noopLifecycleHandler,
-        broadcastSpy,
       );
 
       for (const toolName of [
@@ -439,6 +423,7 @@ describe("session-tool-setup app refresh side effects", () => {
         "app_file_write",
       ]) {
         refreshSpy.mockClear();
+    broadcastSpy.mockClear();
         broadcastSpy.mockClear();
         updatePublishedSpy.mockClear();
 
@@ -451,28 +436,5 @@ describe("session-tool-setup app refresh side effects", () => {
     });
   });
 
-  // ── broadcastToAllClients optional ──────────────────────────────────
-
-  describe("broadcastToAllClients is optional", () => {
-    test("side effects work without broadcastToAllClients callback", async () => {
-      const ctx = makeCtx();
-      const executor = makeFakeExecutor({ content: "{}", isError: false });
-
-      // No broadcast callback provided
-      const toolFn = createToolExecutor(
-        executor as unknown as ToolExecutor,
-        noopPrompter,
-        noopSecretPrompter,
-        ctx,
-        noopLifecycleHandler,
-      );
-
-      // Should not throw even though broadcastToAllClients is undefined
-      const result = await toolFn("app_refresh", { app_id: "app-no-bc" });
 
-      expect(result.isError).toBe(false);
-      expect(refreshSpy).toHaveBeenCalledTimes(1);
-      expect(updatePublishedSpy).toHaveBeenCalledTimes(1);
-    });
-  });
 });

package/src/__tests__/conversation-usage.test.ts

@@ -7,6 +7,8 @@ const updateConversationUsageCalls: Array<{
   estimatedCost: number;
 }> = [];
 
+let mockLlmConfig = createMockLlmConfig();
+
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
     new Proxy({} as Record<string, unknown>, {
@@ -16,9 +18,7 @@ mock.module("../util/logger.js", () => ({
 
 mock.module("../config/loader.js", () => ({
   getConfig: () => ({
-    llm: {
-      pricingOverrides: [],
-    },
+    llm: mockLlmConfig,
   }),
 }));
 
@@ -42,16 +42,67 @@ import { recordUsage } from "../daemon/conversation-usage.js";
 import { getDb } from "../memory/db-connection.js";
 import { initializeDb } from "../memory/db-init.js";
 import { listUsageEvents } from "../memory/llm-usage-store.js";
+import type { Provider, ProviderResponse } from "../providers/types.js";
+import { UsageTrackingProvider } from "../providers/usage-tracking.js";
 import type { PricingUsage } from "../usage/types.js";
 import { resolvePricingForUsageWithOverrides } from "../util/pricing.js";
 
 initializeDb();
 
+function createMockLlmConfig() {
+  return {
+    default: {
+      provider: "anthropic" as const,
+      model: "claude-opus-4-6",
+      maxTokens: 64_000,
+      effort: "max" as const,
+      speed: "standard" as const,
+      verbosity: "medium" as const,
+      temperature: null,
+      thinking: { enabled: true, streamThinking: true },
+      contextWindow: {
+        enabled: true,
+        maxInputTokens: 200_000,
+        targetBudgetRatio: 0.3,
+        compactThreshold: 0.8,
+        summaryBudgetRatio: 0.05,
+        overflowRecovery: {
+          enabled: true,
+          safetyMarginRatio: 0.05,
+          maxAttempts: 3,
+          interactiveLatestTurnCompression: "summarize" as const,
+          nonInteractiveLatestTurnCompression: "truncate" as const,
+        },
+      },
+      openrouter: { only: [] },
+    },
+    profiles: {
+      conversationProfile: {
+        provider: "openai" as const,
+        model: "gpt-4o",
+      },
+      summaryProfile: {
+        provider: "anthropic" as const,
+        model: "claude-haiku-3",
+      },
+    },
+    profileOrder: [],
+    callSites: {
+      conversationSummarization: {
+        profile: "summaryProfile",
+      },
+    },
+    activeProfile: undefined,
+    pricingOverrides: [],
+  };
+}
+
 describe("recordUsage", () => {
   beforeEach(() => {
     const db = getDb();
     db.run(`DELETE FROM llm_usage_events`);
     updateConversationUsageCalls.length = 0;
+    mockLlmConfig = createMockLlmConfig();
   });
 
   test("applies fast mode pricing when any response has speed: fast", () => {
@@ -211,4 +262,203 @@ describe("recordUsage", () => {
       },
     ]);
   });
+
+  test("manual provider usage tracking leaves conversation aggregate recording as the only ledger row", async () => {
+    const response: ProviderResponse = {
+      content: [{ type: "text", text: "ok" }],
+      model: "gpt-5.4-mini",
+      usage: {
+        inputTokens: 1_000,
+        outputTokens: 2_000,
+      },
+      stopReason: "end_turn",
+    };
+    const provider: Provider = {
+      name: "openai",
+      async sendMessage() {
+        return response;
+      },
+    };
+    const wrapped = new UsageTrackingProvider(provider);
+
+    await wrapped.sendMessage(
+      [{ role: "user", content: [{ type: "text", text: "Hello" }] }],
+      undefined,
+      undefined,
+      {
+        config: {
+          callSite: "mainAgent",
+          usageTracking: "manual",
+        },
+      },
+    );
+    expect(listUsageEvents()).toHaveLength(0);
+
+    const usageStats = {
+      inputTokens: 0,
+      outputTokens: 0,
+      estimatedCost: 0,
+    };
+
+    recordUsage(
+      {
+        conversationId: "conv-manual-1",
+        providerName: "openai",
+        usageStats,
+      },
+      response.usage.inputTokens,
+      response.usage.outputTokens,
+      response.model,
+      () => {},
+      "main_agent",
+      "req-manual-1",
+    );
+
+    const events = listUsageEvents();
+    expect(events).toHaveLength(1);
+    expect(events[0]).toMatchObject({
+      actor: "main_agent",
+      conversationId: "conv-manual-1",
+      requestId: "req-manual-1",
+      provider: "openai",
+      model: "gpt-5.4-mini",
+      inputTokens: 1_000,
+      outputTokens: 2_000,
+    });
+  });
+
+  test("persists resolved main-agent attribution from input without changing totals", () => {
+    const usageStats = {
+      inputTokens: 0,
+      outputTokens: 0,
+      estimatedCost: 0,
+    };
+
+    recordUsage(
+      {
+        conversationId: "conv-attrib-1",
+        providerName: "openai",
+        usageStats,
+      },
+      100,
+      20,
+      "gpt-4o",
+      () => {},
+      "main_agent",
+      "req-attrib-1",
+      0,
+      0,
+      undefined,
+      1,
+      undefined,
+      {
+        callSite: "mainAgent",
+        overrideProfile: "conversationProfile",
+      },
+    );
+
+    const events = listUsageEvents();
+    expect(events).toHaveLength(1);
+    expect(events[0]).toMatchObject({
+      actor: "main_agent",
+      provider: "openai",
+      model: "gpt-4o",
+      inputTokens: 100,
+      outputTokens: 20,
+      callSite: "mainAgent",
+      inferenceProfile: "conversationProfile",
+      inferenceProfileSource: "conversation",
+    });
+    expect(usageStats.inputTokens).toBe(100);
+    expect(usageStats.outputTokens).toBe(20);
+  });
+
+  test("persists compaction attribution using the summary call-site profile", () => {
+    const usageStats = {
+      inputTokens: 0,
+      outputTokens: 0,
+      estimatedCost: 0,
+    };
+
+    recordUsage(
+      {
+        conversationId: "conv-attrib-2",
+        providerName: "anthropic",
+        usageStats,
+      },
+      500,
+      80,
+      "claude-haiku-3",
+      () => {},
+      "context_compactor",
+      "req-attrib-2",
+      0,
+      0,
+      undefined,
+      1,
+      undefined,
+      {
+        callSite: "conversationSummarization",
+        overrideProfile: "conversationProfile",
+      },
+    );
+
+    const events = listUsageEvents();
+    expect(events).toHaveLength(1);
+    expect(events[0]).toMatchObject({
+      actor: "context_compactor",
+      provider: "anthropic",
+      model: "claude-haiku-3",
+      inputTokens: 500,
+      outputTokens: 80,
+      callSite: "conversationSummarization",
+      inferenceProfile: "summaryProfile",
+      inferenceProfileSource: "call_site",
+    });
+  });
+
+  test("persists a pre-resolved attribution snapshot", () => {
+    const usageStats = {
+      inputTokens: 0,
+      outputTokens: 0,
+      estimatedCost: 0,
+    };
+
+    recordUsage(
+      {
+        conversationId: "conv-attrib-3",
+        providerName: "anthropic",
+        usageStats,
+      },
+      25,
+      5,
+      "claude-opus-4-6",
+      () => {},
+      "main_agent",
+      "req-attrib-3",
+      0,
+      0,
+      undefined,
+      1,
+      undefined,
+      {
+        callSite: "mainAgent",
+        activeProfile: null,
+        overrideProfile: null,
+        callSiteProfile: null,
+        appliedProfile: null,
+        profileSource: "default",
+        resolvedProvider: "anthropic",
+        resolvedModel: "claude-opus-4-6",
+      },
+    );
+
+    const events = listUsageEvents();
+    expect(events).toHaveLength(1);
+    expect(events[0]).toMatchObject({
+      callSite: "mainAgent",
+      inferenceProfile: null,
+      inferenceProfileSource: "default",
+    });
+  });
 });

package/src/__tests__/credential-execution-feature-gates.test.ts

@@ -17,12 +17,10 @@ import {
 import type { AssistantConfig } from "../config/schema.js";
 import {
   CES_GRANT_AUDIT_FLAG_KEY,
-  CES_MANAGED_SIDECAR_FLAG_KEY,
   CES_SECURE_INSTALL_FLAG_KEY,
   CES_SHELL_LOCKDOWN_FLAG_KEY,
   CES_TOOLS_FLAG_KEY,
   isCesGrantAuditEnabled,
-  isCesManagedSidecarEnabled,
   isCesSecureInstallEnabled,
   isCesShellLockdownEnabled,
   isCesToolsEnabled,
@@ -51,7 +49,6 @@ const ALL_CES_FLAG_KEYS = [
   CES_SHELL_LOCKDOWN_FLAG_KEY,
   CES_SECURE_INSTALL_FLAG_KEY,
   CES_GRANT_AUDIT_FLAG_KEY,
-  CES_MANAGED_SIDECAR_FLAG_KEY,
 ] as const;
 
 /** All CES predicate functions paired with their flag keys and expected defaults. */
@@ -80,12 +77,6 @@ const ALL_CES_PREDICATES = [
     key: CES_GRANT_AUDIT_FLAG_KEY,
     defaultEnabled: false,
   },
-  {
-    name: "isCesManagedSidecarEnabled",
-    fn: isCesManagedSidecarEnabled,
-    key: CES_MANAGED_SIDECAR_FLAG_KEY,
-    defaultEnabled: true,
-  },
 ] as const;
 
 // ---------------------------------------------------------------------------
@@ -180,7 +171,7 @@ describe("CES flags do not affect unrelated flags", () => {
     expect(isAssistantFeatureFlagEnabled("browser", config)).toBe(true);
   });
 
-  test("enabling all CES flags does not change sounds flag (defaultEnabled: true)", () => {
+  test("enabling all CES flags does not change unrelated default-open flags", () => {
     const overrides: Record<string, boolean> = {};
     for (const key of ALL_CES_FLAG_KEYS) {
       overrides[key] = true;
@@ -188,7 +179,9 @@ describe("CES flags do not affect unrelated flags", () => {
     _setOverridesForTesting(overrides);
     const config = makeConfig();
 
-    // sounds defaults to true in the registry and should stay true
-    expect(isAssistantFeatureFlagEnabled("sounds", config)).toBe(true);
+    // Unknown flags default open unless explicitly overridden.
+    expect(
+      isAssistantFeatureFlagEnabled("unrelated-default-open", config),
+    ).toBe(true);
   });
 });