@vellumai/assistant 0.4.56 → 0.5.0

package/src/__tests__/intent-routing.test.ts

@@ -54,7 +54,6 @@ mock.module("../config/loader.js", () => ({
   getConfig: () => ({
     ui: {},
 
-    sandbox: { enabled: false, backend: "local" },
     assistantFeatureFlagValues: {},
     services: {
       inference: {
@@ -65,9 +64,9 @@ mock.module("../config/loader.js", () => ({
       "image-generation": {
         mode: "your-own",
         provider: "gemini",
-        model: "gemini-2.5-flash-image",
+        model: "gemini-3.1-flash-image-preview",
       },
-      "web-search": { mode: "your-own", provider: "anthropic-native" },
+      "web-search": { mode: "your-own", provider: "inference-provider-native" },
     },
   }),
   loadConfig: () => ({}),
@@ -105,11 +104,25 @@ const scheduleCreateDef = scheduleToolsJson.tools.find(
   (t: { name: string }) => t.name === "schedule_create",
 );
 
+// Load send_notification description from the bundled skill TOOLS.json
+const notifToolsJson = JSON.parse(
+  readFileSync(
+    join(
+      import.meta.dirname,
+      "../config/bundled-skills/notifications/TOOLS.json",
+    ),
+    "utf-8",
+  ),
+);
+const sendNotificationDef = notifToolsJson.tools.find(
+  (t: { name: string }) => t.name === "send_notification",
+);
+
 // =====================================================================
-// 1. System prompt: buildTaskScheduleReminderRoutingSection
+// 1. Routing section removed from system prompt — guidance in tool descriptions
 // =====================================================================
 
-describe("Task/Schedule routing section in system prompt", () => {
+describe("Task/Schedule routing NOT in system prompt (moved to tool descriptions)", () => {
   beforeEach(() => {
     mkdirSync(TEST_DIR, { recursive: true });
   });
@@ -120,55 +133,12 @@ describe("Task/Schedule routing section in system prompt", () => {
     }
   });
 
-  test("system prompt includes the routing section heading", () => {
+  test("system prompt does not contain the old routing section", () => {
     const prompt = buildSystemPrompt();
-    expect(prompt).toContain(
+    expect(prompt).not.toContain(
       "## Tool Routing: Tasks vs Schedules vs Notifications",
     );
   });
-
-  test("routing section lists all three tools in the summary table", () => {
-    const prompt = buildSystemPrompt();
-    expect(prompt).toContain("`task_list_add`");
-    expect(prompt).toContain("`schedule_create`");
-    expect(prompt).toContain("`send_notification`");
-  });
-
-  test("routing section warns that send_notification is immediate-only", () => {
-    const prompt = buildSystemPrompt();
-    expect(prompt).toContain("send_notification` is immediate-only");
-    expect(prompt).toContain("fires NOW");
-  });
-
-  test("routing section includes quick routing rules", () => {
-    const prompt = buildSystemPrompt();
-    expect(prompt).toContain("Quick routing rules");
-    expect(prompt).toContain("Future time, one-shot");
-    expect(prompt).toContain("Recurring pattern");
-    expect(prompt).toContain("No time, track as work");
-  });
-
-  test("routing section documents entity type routing", () => {
-    const prompt = buildSystemPrompt();
-    expect(prompt).toContain(
-      "Entity type routing: work items vs task templates",
-    );
-    expect(prompt).toContain("**Work items**");
-    expect(prompt).toContain("**Task templates**");
-  });
-
-  test("routing section references the Time-Based Actions skill", () => {
-    const prompt = buildSystemPrompt();
-    expect(prompt).toContain("Time-Based Actions");
-  });
-
-  test("routing section is present in the system prompt", () => {
-    const prompt = buildSystemPrompt();
-    const taskRoutingIdx = prompt.indexOf(
-      "## Tool Routing: Tasks vs Schedules vs Notifications",
-    );
-    expect(taskRoutingIdx).toBeGreaterThanOrEqual(0);
-  });
 });
 
 // =====================================================================
@@ -228,6 +198,18 @@ describe("schedule_create tool description", () => {
   });
 });
 
+describe("send_notification tool description", () => {
+  test("states it fires immediately with no delay", () => {
+    expect(sendNotificationDef).toBeDefined();
+    expect(sendNotificationDef.description).toContain("immediate");
+    expect(sendNotificationDef.description).toContain("no delay");
+  });
+
+  test("redirects to schedule_create for future alerts", () => {
+    expect(sendNotificationDef.description).toContain("schedule_create");
+  });
+});
+
 // =====================================================================
 // 3. Cross-tool consistency: schedule and task tools agree on routing boundaries
 // =====================================================================
@@ -244,10 +226,10 @@ describe("cross-tool routing consistency", () => {
 });
 
 // =====================================================================
-// 4. Activation hints in <available_skills> XML (replaces domain routing sections)
+// 4. Activation hints in skills catalog (replaces domain routing sections)
 // =====================================================================
 
-describe("Activation hints in available_skills XML", () => {
+describe("Activation hints in skills catalog", () => {
   beforeEach(() => {
     mkdirSync(TEST_DIR, { recursive: true });
   });
@@ -258,61 +240,31 @@ describe("Activation hints in available_skills XML", () => {
     }
   });
 
-  test("available_skills XML does NOT contain location attribute", () => {
+  test("phone-calls skill includes hints and avoid-when in catalog line", () => {
     const prompt = buildSystemPrompt();
-    const start = prompt.indexOf("<available_skills>");
-    const end = prompt.indexOf("</available_skills>");
-    expect(start).not.toBe(-1);
-    expect(end).not.toBe(-1);
-    const skillsXml = prompt.substring(start, end);
-    expect(skillsXml).not.toContain("location=");
+    const line = prompt.split("\n").find((l) => l.includes("**phone-calls**"));
+    expect(line).toBeDefined();
+    // Activation hints and avoid-when are folded into the description
+    expect(line).toContain("Twilio");
+    expect(line).toContain("Avoid: ");
+    expect(line).toContain("voice-setup");
   });
 
-  test("phone-calls bundled skill has hints and avoid-when attributes in XML", () => {
+  test("orchestration skill includes hints and avoid-when in catalog line", () => {
     const prompt = buildSystemPrompt();
-    const start = prompt.indexOf("<available_skills>");
-    const end = prompt.indexOf("</available_skills>");
-    expect(start).not.toBe(-1);
-    expect(end).not.toBe(-1);
-    const skillsXml = prompt.substring(start, end);
-    expect(skillsXml).toContain('id="phone-calls"');
-    const skillLine = skillsXml
+    const line = prompt
       .split("\n")
-      .find((l) => l.includes('id="phone-calls"'));
-    expect(skillLine).toBeDefined();
-    expect(skillLine).toContain("hints=");
-    expect(skillLine).toContain("avoid-when=");
+      .find((l) => l.includes("**orchestration**"));
+    expect(line).toBeDefined();
+    expect(line).toContain("parallel");
+    expect(line).toContain("Single-focus");
   });
 
-  test("orchestration bundled skill has hints and avoid-when attributes in XML", () => {
+  test("browser skill includes hints in catalog line", () => {
     const prompt = buildSystemPrompt();
-    const start = prompt.indexOf("<available_skills>");
-    const end = prompt.indexOf("</available_skills>");
-    expect(start).not.toBe(-1);
-    expect(end).not.toBe(-1);
-    const skillsXml = prompt.substring(start, end);
-    expect(skillsXml).toContain('id="orchestration"');
-    const skillLine = skillsXml
-      .split("\n")
-      .find((l) => l.includes('id="orchestration"'));
-    expect(skillLine).toBeDefined();
-    expect(skillLine).toContain("hints=");
-    expect(skillLine).toContain("avoid-when=");
-  });
-
-  test("browser bundled skill has hints attribute in XML", () => {
-    const prompt = buildSystemPrompt();
-    const start = prompt.indexOf("<available_skills>");
-    const end = prompt.indexOf("</available_skills>");
-    expect(start).not.toBe(-1);
-    expect(end).not.toBe(-1);
-    const skillsXml = prompt.substring(start, end);
-    expect(skillsXml).toContain('id="browser"');
-    const skillLine = skillsXml
-      .split("\n")
-      .find((l) => l.includes('id="browser"'));
-    expect(skillLine).toBeDefined();
-    expect(skillLine).toContain("hints=");
+    const line = prompt.split("\n").find((l) => l.includes("**browser**"));
+    expect(line).toBeDefined();
+    expect(line).toContain("browser_*");
   });
 
   test("domain routing sections are no longer in system prompt", () => {

package/src/__tests__/invite-routes-http.test.ts

@@ -38,6 +38,11 @@ mock.module("../telegram/bot-username.js", () => ({
   getTelegramBotUsername: () => mockTelegramBotUsername,
 }));
 
+// Mock invite instruction generator — skip LLM calls in tests.
+mock.module("../runtime/invite-instruction-generator.js", () => ({
+  generateInviteInstruction: async () => "Mock invite instruction for testing.",
+}));
+
 // Mock startInviteCall from call-domain — test env lacks Twilio credentials.
 let mockStartInviteCallResult:
   | { ok: true; callSid: string }

package/src/__tests__/list-messages-attachments.test.ts

@@ -39,7 +39,7 @@ mock.module("../config/loader.js", () => ({
     model: "test",
     provider: "test",
     memory: { enabled: false },
-    rateLimit: { maxRequestsPerMinute: 0, maxTokensPerSession: 0 },
+    rateLimit: { maxRequestsPerMinute: 0 },
   }),
 }));
 

package/src/__tests__/managed-proxy-context.test.ts

@@ -110,10 +110,7 @@ describe("buildManagedBaseUrl", () => {
       "https://platform.example.com/v1/runtime-proxy/anthropic",
     );
     expect(await buildManagedBaseUrl("gemini")).toBe(
-      "https://platform.example.com/v1/runtime-proxy/vertex",
-    );
-    expect(await buildManagedBaseUrl("vertex")).toBe(
-      "https://platform.example.com/v1/runtime-proxy/vertex",
+      "https://platform.example.com/v1/runtime-proxy/gemini",
     );
   });
 
@@ -132,7 +129,7 @@ describe("buildManagedBaseUrl", () => {
     mockPlatformBaseUrl = "";
     mockAssistantApiKey = null;
     expect(await buildManagedBaseUrl("anthropic")).toBeUndefined();
-    expect(await buildManagedBaseUrl("vertex")).toBeUndefined();
+    expect(await buildManagedBaseUrl("gemini")).toBeUndefined();
   });
 });
 

package/src/__tests__/managed-skill-lifecycle.test.ts

@@ -16,8 +16,7 @@ const mockConfig = {
     shellMaxTimeoutSec: 600,
     permissionTimeoutSec: 300,
   },
-  sandbox: { enabled: false },
-  rateLimit: { maxRequestsPerMinute: 0, maxTokensPerSession: 0 },
+  rateLimit: { maxRequestsPerMinute: 0 },
   secretDetection: {
     enabled: true,
     action: "warn" as const,
@@ -33,9 +32,9 @@ const mockConfig = {
     "image-generation": {
       mode: "your-own",
       provider: "gemini",
-      model: "gemini-2.5-flash-image",
+      model: "gemini-3.1-flash-image-preview",
     },
-    "web-search": { mode: "your-own", provider: "anthropic-native" },
+    "web-search": { mode: "your-own", provider: "inference-provider-native" },
   },
 };
 
@@ -138,11 +137,12 @@ describe("managed skill lifecycle: scaffold → catalog → prompt → delete",
     expect(found!.name).toBe("Lifecycle Test");
     expect(found!.description).toBe("Integration test skill.");
 
-    // Step 4: Verify skill appears in system prompt
+    // Step 4: Verify skill appears in system prompt (markdown bullet: **id**: description)
     const prompt = buildSystemPrompt();
-    expect(prompt).toContain("lifecycle-test");
-    expect(prompt).toContain("Lifecycle Test");
-    expect(prompt).toContain("## Dynamic Skill Authoring Workflow");
+    expect(prompt).toContain("**lifecycle-test**");
+    expect(prompt).toContain("Integration test skill");
+    // Dynamic Skill Authoring section moved to tool descriptions; prompt should not contain it
+    expect(prompt).not.toContain("## Dynamic Skill Authoring Workflow");
 
     // Step 5: Delete the skill
     const deleteResult = await executeDeleteManagedSkill(

package/src/__tests__/media-generate-image.test.ts

@@ -11,10 +11,11 @@ import type { ToolContext } from "../tools/types.js";
 // ---------------------------------------------------------------------------
 
 let mockApiKey: string | undefined = "test-gemini-key";
+let mockImageGenMode: "your-own" | "managed" = "your-own";
 let mockGenerateResult = {
   images: [{ mimeType: "image/png", dataBase64: "generated-data" }],
   text: "A beautiful image",
-  resolvedModel: "gemini-2.5-flash-image",
+  resolvedModel: "gemini-3.1-flash-image-preview",
 };
 let mockGenerateError: Error | null = null;
 let lastGenerateCredentials: unknown = null;
@@ -29,11 +30,11 @@ mock.module("../config/loader.js", () => ({
         model: "claude-opus-4-6",
       },
       "image-generation": {
-        mode: "your-own",
+        mode: mockImageGenMode,
         provider: "gemini",
-        model: "gemini-2.5-flash-image",
+        model: "gemini-3.1-flash-image-preview",
       },
-      "web-search": { mode: "your-own", provider: "anthropic-native" },
+      "web-search": { mode: "your-own", provider: "inference-provider-native" },
     },
   }),
 }));
@@ -43,6 +44,10 @@ mock.module("../security/secure-keys.js", () => ({
     if (account === "gemini") return mockApiKey;
     return undefined;
   },
+  getProviderKeyAsync: async (provider: string) => {
+    if (provider === "gemini") return mockApiKey;
+    return undefined;
+  },
 }));
 
 mock.module("../media/gemini-image-service.js", () => ({
@@ -171,10 +176,11 @@ const CONFIG_DIR = join(
 
 beforeEach(() => {
   mockApiKey = "test-gemini-key";
+  mockImageGenMode = "your-own";
   mockGenerateResult = {
     images: [{ mimeType: "image/png", dataBase64: "generated-data" }],
     text: "A beautiful image",
-    resolvedModel: "gemini-2.5-flash-image",
+    resolvedModel: "gemini-3.1-flash-image-preview",
   };
   mockGenerateError = null;
   mockAttachments = [];
@@ -208,9 +214,9 @@ describe("image-studio skill script wrapper", () => {
     expect(result.content).toContain("No Gemini API key");
   });
 
-  test("falls back to managed proxy when no API key is configured", async () => {
-    mockApiKey = undefined;
-    mockManagedBaseUrl = "https://platform.example.com/v1/runtime-proxy/vertex";
+  test("managed mode uses managed proxy credentials", async () => {
+    mockImageGenMode = "managed";
+    mockManagedBaseUrl = "https://platform.example.com/v1/runtime-proxy/gemini";
     mockManagedProxyContext = {
       enabled: true,
       platformBaseUrl: "https://platform.example.com",
@@ -224,13 +230,25 @@ describe("image-studio skill script wrapper", () => {
     expect(lastGenerateCredentials).toEqual({
       type: "managed-proxy",
       assistantApiKey: "managed-key-123",
-      baseUrl: "https://platform.example.com/v1/runtime-proxy/vertex",
+      baseUrl: "https://platform.example.com/v1/runtime-proxy/gemini",
     });
   });
 
-  test("prefers direct API key over managed proxy", async () => {
+  test("managed mode returns error when managed proxy is unavailable", async () => {
+    mockImageGenMode = "managed";
+    mockApiKey = "direct-key"; // should be ignored in managed mode
+    mockManagedBaseUrl = undefined;
+
+    const result = await run({ prompt: "a cat" }, fakeContext);
+
+    expect(result.isError).toBe(true);
+    expect(result.content).toContain("Managed proxy is not available");
+  });
+
+  test("your-own mode uses direct API key", async () => {
+    mockImageGenMode = "your-own";
     mockApiKey = "direct-key";
-    mockManagedBaseUrl = "https://platform.example.com/v1/runtime-proxy/vertex";
+    mockManagedBaseUrl = "https://platform.example.com/v1/runtime-proxy/gemini";
     mockManagedProxyContext = {
       enabled: true,
       platformBaseUrl: "https://platform.example.com",
@@ -250,7 +268,7 @@ describe("image-studio skill script wrapper", () => {
 
     expect(result.isError).toBe(false);
     expect(result.content).toContain("Generated 1 image");
-    expect(result.content).toContain("gemini-2.5-flash-image");
+    expect(result.content).toContain("gemini-3.1-flash-image-preview");
     expect(result.content).toContain("A beautiful image");
     expect(result.contentBlocks).toHaveLength(1);
     expect(result.contentBlocks![0]).toEqual({
@@ -270,7 +288,7 @@ describe("image-studio skill script wrapper", () => {
         { mimeType: "image/png", dataBase64: "img2" },
       ],
       text: undefined as unknown as string,
-      resolvedModel: "gemini-2.5-flash-image",
+      resolvedModel: "gemini-3.1-flash-image-preview",
     };
 
     const result = await run({ prompt: "test", variants: 2 }, fakeContext);
@@ -357,8 +375,7 @@ describe("image-studio TOOLS.json manifest", () => {
     expect(props.mode.enum).toEqual(["generate", "edit"]);
     expect(props.attachment_ids.type).toBe("array");
     expect(props.model.enum).toEqual([
-      "gemini-2.5-flash-image",
-      "gemini-3-pro-image",
+      "gemini-3.1-flash-image-preview",
       "gemini-3-pro-image-preview",
     ]);
     expect(props.variants.type).toBe("number");

package/src/__tests__/media-reuse-story.e2e.test.ts

@@ -53,7 +53,7 @@ mock.module("../config/loader.js", () => ({
     model: "test",
     provider: "test",
     memory: { enabled: false },
-    rateLimit: { maxRequestsPerMinute: 0, maxTokensPerSession: 0 },
+    rateLimit: { maxRequestsPerMinute: 0 },
     timeouts: { shellDefaultTimeoutSec: 30, shellMaxTimeoutSec: 60 },
     sandbox: { enabled: false, backend: "native" },
   }),

package/src/__tests__/memory-context-benchmark.benchmark.test.ts

@@ -248,7 +248,7 @@ describe("Memory context benchmark", () => {
         ...DEFAULT_CONFIG.memory,
         embeddings: {
           ...DEFAULT_CONFIG.memory.embeddings,
-          provider: "openai" as const,
+          provider: "local" as const,
           required: false,
         },
         retrieval: {

package/src/__tests__/memory-lifecycle-e2e.test.ts

@@ -82,6 +82,7 @@ mock.module("../config/loader.js", () => ({
   invalidateConfigCache: () => {},
 }));
 
+import { stripUserTextBlocksByPrefix } from "../daemon/conversation-runtime-assembly.js";
 import { getDb, initializeDb, resetDb } from "../memory/db.js";
 import {
   resetCleanupScheduleThrottle,
@@ -89,8 +90,7 @@ import {
 } from "../memory/jobs-worker.js";
 import {
   buildMemoryRecall,
-  injectMemoryRecallAsSeparateMessage,
-  stripMemoryRecallMessages,
+  injectMemoryRecallAsUserBlock,
 } from "../memory/retriever.js";
 import {
   conversations,
@@ -98,6 +98,7 @@ import {
   memoryItemSources,
   messages,
 } from "../memory/schema.js";
+import type { Message } from "../providers/types.js";
 
 describe("Memory lifecycle E2E regression", () => {
   beforeAll(() => {
@@ -322,7 +323,7 @@ describe("Memory lifecycle E2E regression", () => {
         totalOutputTokens: 0,
         totalEstimatedCost: 0,
         contextSummary: null,
-        contextCompactedMessageCount: 0,
+        contextCompactedMessageCount: 1,
         contextCompactedAt: null,
       })
       .run();
@@ -365,34 +366,39 @@ describe("Memory lifecycle E2E regression", () => {
     expect(recall.enabled).toBe(true);
     expect(recall.injectedText.length).toBeGreaterThan(0);
     expect(recall.injectedTokens).toBeGreaterThan(0);
-    expect(recall.injectedText).toContain("<memory_context>");
+    expect(recall.injectedText).toContain("<memory_context __injected>");
     expect(recall.injectedText).toContain("</memory_context>");
   });
 
-  test("stripping removes <memory_context> tags from injected recall", () => {
+  test("stripping removes <memory_context> block from injected recall", () => {
     const memoryRecallText =
-      "<memory_context>\n\n<relevant_context>\nuser prefers concise answers\n</relevant_context>\n\n</memory_context>";
-    const originalMessages = [
+      "<memory_context __injected>\n\n<relevant_context>\nuser prefers concise answers\n</relevant_context>\n\n</memory_context>";
+    const originalMessages: Message[] = [
       {
-        role: "user" as const,
-        content: [{ type: "text", text: "Actual user request" }],
+        role: "user",
+        content: [{ type: "text" as const, text: "Actual user request" }],
       },
     ];
-    const injected = injectMemoryRecallAsSeparateMessage(
+    const injected = injectMemoryRecallAsUserBlock(
       originalMessages,
       memoryRecallText,
     );
 
-    expect(injected).toHaveLength(3);
+    // Memory context prepended to the last user message as a content block
+    expect(injected).toHaveLength(1);
     expect(injected[0].role).toBe("user");
-    expect(injected[0].content[0].text).toBe(memoryRecallText);
-    expect(injected[1].role as string).toBe("assistant");
-    expect(injected[2].role).toBe("user");
-    expect(injected[2].content[0].text).toBe("Actual user request");
-
-    const cleaned = stripMemoryRecallMessages(injected, memoryRecallText);
+    expect(injected[0].content).toHaveLength(2);
+    const b0 = injected[0].content[0];
+    const b1 = injected[0].content[1];
+    expect(b0.type === "text" && b0.text).toBe(memoryRecallText);
+    expect(b1.type === "text" && b1.text).toBe("Actual user request");
+
+    // Stripped by prefix-based stripping (same mechanism as workspace/temporal)
+    const cleaned = stripUserTextBlocksByPrefix(injected, ["<memory_context __injected>"]);
     expect(cleaned).toHaveLength(1);
-    expect(cleaned[0].content[0].text).toBe("Actual user request");
+    expect(cleaned[0].content).toHaveLength(1);
+    const cb0 = cleaned[0].content[0];
+    expect(cb0.type === "text" && cb0.text).toBe("Actual user request");
   });
 
   test("empty retrieval returns no injection", async () => {

package/src/__tests__/memory-recall-quality.test.ts

@@ -112,6 +112,7 @@ function insertConversation(
   db: ReturnType<typeof getDb>,
   id: string,
   createdAt: number,
+  contextCompactedMessageCount = 0,
 ) {
   db.insert(conversations)
     .values({
@@ -123,7 +124,7 @@ function insertConversation(
       totalOutputTokens: 0,
       totalEstimatedCost: 0,
       contextSummary: null,
-      contextCompactedMessageCount: 0,
+      contextCompactedMessageCount,
       contextCompactedAt: null,
     })
     .run();
@@ -629,7 +630,7 @@ describe("Memory Recall Quality", () => {
     test("invalidated items are excluded from recall", async () => {
       const db = getDb();
       const now = 1_700_000_275_000;
-      insertConversation(db, "conv-invalid-status", now);
+      insertConversation(db, "conv-invalid-status", now, 1);
       insertMessage(
         db,
         "msg-invalid-status",
@@ -989,7 +990,7 @@ describe("Memory Recall Quality", () => {
     test("precision@k guard verifies pipeline completes with seeded segments", async () => {
       const db = getDb();
       const now = 1_700_000_700_000;
-      insertConversation(db, "conv-pk", now);
+      insertConversation(db, "conv-pk", now, 3);
 
       const prefs = [
         {