@vellumai/assistant 0.5.15 → 0.6.0

package/src/__tests__/assistant-feature-flags-integration.test.ts

@@ -1,85 +1,20 @@
 /**
- * Integration tests for assistant feature flag enforcement at system prompt,
- * skill_load, and session-skill-tools projection layers.
+ * Integration tests for assistant feature flag resolver.
  *
  * Covers:
- *   - Flag OFF blocks all exposure paths
  *   - Missing persisted value falls back to code default
  *   - Protected feature-flags.json is the sole override mechanism
  *   - Undeclared keys default to enabled
  */
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
-import { join } from "node:path";
-import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
 // ---------------------------------------------------------------------------
-// Test-scoped temp directory and config state
+// Test-scoped config state
 // ---------------------------------------------------------------------------
 
-const TEST_DIR = process.env.VELLUM_WORKSPACE_DIR!;
-
-let currentConfig: Record<string, unknown> = {
-  services: {
-    inference: {
-      mode: "your-own",
-      provider: "anthropic",
-      model: "claude-opus-4-6",
-    },
-    "image-generation": {
-      mode: "your-own",
-      provider: "gemini",
-      model: "gemini-3.1-flash-image-preview",
-    },
-    "web-search": { mode: "your-own", provider: "inference-provider-native" },
-  },
-};
-
-const DECLARED_FLAG_ID = "contacts";
+const DECLARED_FLAG_ID = "sounds";
 const DECLARED_FLAG_KEY = DECLARED_FLAG_ID;
-const DECLARED_SKILL_ID = "contacts";
-
-const noopLogger = new Proxy({} as Record<string, unknown>, {
-  get: (_target, prop) => (prop === "child" ? () => noopLogger : () => {}),
-});
-
-// eslint-disable-next-line @typescript-eslint/no-require-imports
-const realLogger = require("../util/logger.js");
-mock.module("../util/logger.js", () => ({
-  ...realLogger,
-  getLogger: () => noopLogger,
-  getCliLogger: () => noopLogger,
-  truncateForLog: (v: string) => v,
-  initLogger: () => {},
-  pruneOldLogFiles: () => 0,
-}));
 
-mock.module("../config/loader.js", () => ({
-  getConfig: () => currentConfig,
-  loadConfig: () => currentConfig,
-  loadRawConfig: () => ({}),
-  saveConfig: () => {},
-  saveRawConfig: () => {},
-  invalidateConfigCache: () => {},
-  getNestedValue: () => undefined,
-  setNestedValue: () => {},
-}));
-
-// eslint-disable-next-line @typescript-eslint/no-require-imports
-const realUserReference = require("../prompts/user-reference.js");
-mock.module("../prompts/user-reference.js", () => ({
-  ...realUserReference,
-  resolveUserReference: () => "TestUser",
-  resolveUserPronouns: () => null,
-}));
-
-// eslint-disable-next-line @typescript-eslint/no-require-imports
-const realCredentialMetadataStore = require("../tools/credentials/metadata-store.js");
-mock.module("../tools/credentials/metadata-store.js", () => ({
-  ...realCredentialMetadataStore,
-  listCredentialMetadata: () => [],
-}));
-
-const { buildSystemPrompt } = await import("../prompts/system-prompt.js");
 const { isAssistantFeatureFlagEnabled, _setOverridesForTesting } =
   await import("../config/assistant-feature-flags.js");
 const { skillFlagKey } = await import("../config/skill-state.js");
@@ -90,319 +25,12 @@ const { skillFlagKey } = await import("../config/skill-state.js");
 
 beforeEach(() => {
   _setOverridesForTesting({});
-  currentConfig = {
-    services: {
-      inference: {
-        mode: "your-own",
-        provider: "anthropic",
-        model: "claude-opus-4-6",
-      },
-      "image-generation": {
-        mode: "your-own",
-        provider: "gemini",
-        model: "gemini-3.1-flash-image-preview",
-      },
-      "web-search": { mode: "your-own", provider: "inference-provider-native" },
-    },
-  };
 });
 
 afterEach(() => {
   _setOverridesForTesting({});
 });
 
-// ---------------------------------------------------------------------------
-// Helpers
-// ---------------------------------------------------------------------------
-
-function createSkillOnDisk(
-  id: string,
-  name: string,
-  description: string,
-  featureFlag?: string,
-): void {
-  const skillsDir = join(TEST_DIR, "skills");
-  mkdirSync(join(skillsDir, id), { recursive: true });
-  const ffBlock = featureFlag
-    ? `\nmetadata: {"vellum":{"feature-flag":"${featureFlag}"}}`
-    : "";
-  writeFileSync(
-    join(skillsDir, id, "SKILL.md"),
-    `---\nname: "${name}"\ndescription: "${description}"${ffBlock}\n---\n\nInstructions for ${id}.\n`,
-  );
-  const indexPath = join(skillsDir, "SKILLS.md");
-  const existing = existsSync(indexPath)
-    ? readFileSync(indexPath, "utf-8")
-    : "";
-  writeFileSync(indexPath, existing + `- ${id}\n`);
-}
-
-// ---------------------------------------------------------------------------
-// System prompt — assistant feature flag filtering
-// ---------------------------------------------------------------------------
-
-describe("buildSystemPrompt assistant feature flag filtering", () => {
-  test("flag OFF skill does not appear in skills catalog", () => {
-    createSkillOnDisk(
-      DECLARED_SKILL_ID,
-      "Contacts",
-      "Toggle contacts behavior",
-      DECLARED_FLAG_ID,
-    );
-    createSkillOnDisk(
-      "browser",
-      "Browser",
-      "Web browsing automation",
-      "browser",
-    );
-
-    _setOverridesForTesting({
-      [DECLARED_FLAG_KEY]: false,
-      browser: true,
-    });
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    // browser is explicitly enabled, declared flagged skill is explicitly off
-    expect(result).toContain("**browser**");
-    expect(result).not.toContain(`**${DECLARED_SKILL_ID}**`);
-  });
-
-  test("contacts visible but email-channel hidden when no flag overrides set (contacts defaults true, email-channel defaults false)", () => {
-    createSkillOnDisk(
-      DECLARED_SKILL_ID,
-      "Contacts",
-      "Toggle contacts behavior",
-      DECLARED_FLAG_ID,
-    );
-    createSkillOnDisk(
-      "email-channel",
-      "Email Channel",
-      "Email channel setup",
-      "email-channel",
-    );
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    // contacts defaults to true, email-channel defaults to false
-    expect(result).toContain(`**${DECLARED_SKILL_ID}**`);
-    expect(result).not.toContain("**email-channel**");
-  });
-
-  test("flagged-off skills hidden when all flags are OFF", () => {
-    createSkillOnDisk(
-      DECLARED_SKILL_ID,
-      "Contacts",
-      "Toggle contacts behavior",
-      DECLARED_FLAG_ID,
-    );
-    createSkillOnDisk(
-      "email-channel",
-      "Email Channel",
-      "Email channel setup",
-      "email-channel",
-    );
-
-    _setOverridesForTesting({
-      [DECLARED_FLAG_KEY]: false,
-      "email-channel": false,
-    });
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    expect(result).not.toContain(`**${DECLARED_SKILL_ID}**`);
-    expect(result).not.toContain("**email-channel**");
-  });
-
-  test("file-based overrides control visibility", () => {
-    createSkillOnDisk(
-      DECLARED_SKILL_ID,
-      "Contacts",
-      "Toggle contacts behavior",
-      DECLARED_FLAG_ID,
-    );
-
-    _setOverridesForTesting({ [DECLARED_FLAG_KEY]: true });
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    expect(result).toContain(`**${DECLARED_SKILL_ID}**`);
-  });
-
-  test("persisted overrides for undeclared flags are respected", () => {
-    createSkillOnDisk(
-      "browser",
-      "Browser",
-      "Web browsing automation",
-      "browser",
-    );
-
-    _setOverridesForTesting({ browser: false });
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    // browser declares featureFlag: "browser" and the user
-    // explicitly disabled it — that override must be honored.
-    expect(result).not.toContain("**browser**");
-  });
-
-  test("declared flags with no persisted override use registry default", () => {
-    createSkillOnDisk(
-      "browser",
-      "Browser",
-      "Web browsing automation",
-      "browser",
-    );
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    // browser is declared in the registry with defaultEnabled: true
-    expect(result).toContain("**browser**");
-  });
-
-  test("skill without featureFlag is never flag-gated", () => {
-    createSkillOnDisk("my-skill", "My Skill", "A skill without feature flag");
-
-    currentConfig = {
-      services: {
-        inference: {
-          mode: "your-own",
-          provider: "anthropic",
-          model: "claude-opus-4-6",
-        },
-        "image-generation": {
-          mode: "your-own",
-          provider: "gemini",
-          model: "gemini-3.1-flash-image-preview",
-        },
-        "web-search": {
-          mode: "your-own",
-          provider: "inference-provider-native",
-        },
-      },
-    };
-
-    const result = buildSystemPrompt();
-
-    // Skills without featureFlag declared are never gated — always pass through
-    expect(result).toContain("**my-skill**");
-  });
-});
-
 // ---------------------------------------------------------------------------
 // Resolver unit tests (within integration context)
 // ---------------------------------------------------------------------------
@@ -426,7 +54,7 @@ describe("isAssistantFeatureFlagEnabled", () => {
 
   test("missing persisted value falls back to defaults registry defaultEnabled", () => {
     // No explicit config at all — should fall back to defaults registry
-    // which has defaultEnabled: true for contacts
+    // which has defaultEnabled: true for sounds
     const config = {} as any;
 
     expect(isAssistantFeatureFlagEnabled(DECLARED_FLAG_KEY, config)).toBe(true);

package/src/__tests__/call-conversation-messages.test.ts

@@ -1,4 +1,4 @@
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -17,7 +17,7 @@ import {
   updateCallSession,
 } from "../calls/call-store.js";
 import { getMessages } from "../memory/conversation-crud.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 
 initializeDb();
@@ -65,10 +65,6 @@ describe("call-conversation-messages", () => {
     resetTables();
   });
 
-  afterAll(() => {
-    resetDb();
-  });
-
   test("buildCallCompletionMessage labels failed calls correctly", () => {
     const conversationId = "conv-call-msg-failed";
     ensureConversation(conversationId);

package/src/__tests__/call-domain.test.ts

@@ -7,7 +7,7 @@
  * - Voice-ingress preflight blocks doomed outbound calls before Twilio dialing.
  * - Pointer messages are written on successful call start and on failure.
  */
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -120,7 +120,7 @@ import {
 } from "../calls/call-domain.js";
 import type { AssistantConfig } from "../config/types.js";
 import { getMessages } from "../memory/conversation-crud.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 
 initializeDb();
@@ -185,10 +185,6 @@ function getLatestAssistantText(conversationId: string): string | null {
   return latest.content;
 }
 
-afterAll(() => {
-  resetDb();
-});
-
 function makeConfig(
   overrides: {
     allowPerCallOverride?: boolean;

package/src/__tests__/call-pointer-messages.test.ts

@@ -1,12 +1,4 @@
-import {
-  afterAll,
-  afterEach,
-  beforeEach,
-  describe,
-  expect,
-  mock,
-  test,
-} from "bun:test";
+import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -22,7 +14,7 @@ import {
   setPointerMessageProcessor,
 } from "../calls/call-pointer-messages.js";
 import { addMessage, getMessages } from "../memory/conversation-crud.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 
 initializeDb();
@@ -110,10 +102,6 @@ describe("addPointerMessage", () => {
     resetPointerMessageProcessor();
   });
 
-  afterAll(() => {
-    resetDb();
-  });
-
   test("adds a started pointer message", () => {
     const convId = "conv-ptr-started";
     ensureConversation(convId);

package/src/__tests__/call-recovery.test.ts

@@ -1,4 +1,4 @@
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -26,15 +26,11 @@ import {
   updateCallSession,
 } from "../calls/call-store.js";
 import type { VoiceProvider } from "../calls/voice-provider.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 
 initializeDb();
 
-afterAll(() => {
-  resetDb();
-});
-
 /** Ensure a conversation row exists for the given ID so FK constraints pass. */
 let ensuredConvIds = new Set<string>();
 function ensureConversation(id: string): void {

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

@@ -5,7 +5,7 @@
  * POST /v1/calls/:id/cancel, and POST /v1/calls/:id/answer
  * through RuntimeHttpServer.
  */
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../config/env.js", () => ({ isHttpAuthDisabled: () => true }));
 
@@ -105,7 +105,7 @@ import {
   createPendingQuestion,
   updateCallSession,
 } from "../calls/call-store.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 import { RuntimeHttpServer } from "../runtime/http-server.js";
 
@@ -162,10 +162,6 @@ describe("runtime call routes — HTTP layer", () => {
     resetTables();
   });
 
-  afterAll(() => {
-    resetDb();
-  });
-
   async function startServer(): Promise<void> {
     server = new RuntimeHttpServer({ port: 0, bearerToken: TEST_TOKEN });
     await server.start();

package/src/__tests__/call-store.test.ts

@@ -1,4 +1,4 @@
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -23,15 +23,11 @@ import {
   releaseCallbackClaim,
   updateCallSession,
 } from "../calls/call-store.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { conversations } from "../memory/schema.js";
 
 initializeDb();
 
-afterAll(() => {
-  resetDb();
-});
-
 /** Ensure a conversation row exists for the given ID so FK constraints pass. */
 let ensuredConvIds = new Set<string>();
 function ensureConversation(id: string): void {

package/src/__tests__/cancel-resolves-conversation-key.test.ts

@@ -6,7 +6,7 @@
  * from the daemon's internal ID), fails to find the conversation, and
  * silently ignores the cancel — leaving the stream running.
  */
-import { afterAll, describe, expect, mock, test } from "bun:test";
+import { describe, expect, mock, test } from "bun:test";
 
 mock.module("../config/env.js", () => ({ isHttpAuthDisabled: () => true }));
 
@@ -43,15 +43,11 @@ mock.module("../config/loader.js", () => ({
 }));
 
 import { getOrCreateConversation } from "../memory/conversation-key-store.js";
-import { initializeDb, resetDb } from "../memory/db.js";
+import { initializeDb } from "../memory/db.js";
 import { conversationManagementRouteDefinitions } from "../runtime/routes/conversation-management-routes.js";
 
 initializeDb();
 
-afterAll(() => {
-  resetDb();
-});
-
 describe("POST /v1/conversations/:id/cancel", () => {
   test("resolves conversation key to internal ID before cancelling", () => {
     // Create a conversation via key — this assigns an internal ID that

package/src/__tests__/canonical-guardian-store.test.ts

@@ -1,4 +1,4 @@
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -21,7 +21,7 @@ import {
   updateCanonicalGuardianDelivery,
   updateCanonicalGuardianRequest,
 } from "../memory/canonical-guardian-store.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 
 initializeDb();
 
@@ -40,10 +40,6 @@ describe("canonical-guardian-store", () => {
     resetTables();
   });
 
-  afterAll(() => {
-    resetDb();
-  });
-
   // ── createCanonicalGuardianRequest ────────────────────────────────
 
   test("creates a request with all fields populated", () => {

package/src/__tests__/ces-rpc-credential-backend.test.ts

@@ -185,7 +185,10 @@ describe("CesRpcCredentialBackend", () => {
       const result = await backend.list();
 
       expect(callFn).toHaveBeenCalledWith(CesRpcMethod.ListCredentials, {});
-      expect(result).toEqual({ accounts: ["account-a", "account-b"], unreachable: false });
+      expect(result).toEqual({
+        accounts: ["account-a", "account-b"],
+        unreachable: false,
+      });
     });
 
     test("returns unreachable when RPC call throws", async () => {

package/src/__tests__/channel-delivery-store.test.ts

@@ -1,4 +1,4 @@
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -13,7 +13,7 @@ import {
   getConversationByKey,
   setConversationKey,
 } from "../memory/conversation-key-store.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import {
   clearPayload,
   findMessageBySourceId,
@@ -40,10 +40,6 @@ import { handleDeleteConversation } from "../runtime/routes/channel-routes.js";
 
 initializeDb();
 
-afterAll(() => {
-  resetDb();
-});
-
 function resetTables() {
   const db = getDb();
   db.run("DELETE FROM channel_inbound_events");