@vellumai/assistant 0.5.12 → 0.5.14

package/src/__tests__/anthropic-provider.test.ts

@@ -37,25 +37,37 @@ mock.module("@anthropic-ai/sdk", () => ({
     constructor(args: Record<string, unknown>) {
       lastConstructorArgs = { ...args };
     }
+    #streamImpl = (
+      params: Record<string, unknown>,
+      options?: Record<string, unknown>,
+    ) => {
+      lastStreamParams = JSON.parse(JSON.stringify(params));
+      _lastStreamOptions = options ?? null;
+      const handlers: Record<string, ((...args: unknown[]) => void)[]> = {};
+      return {
+        on(event: string, cb: (...args: unknown[]) => void) {
+          (handlers[event] ??= []).push(cb);
+          return this;
+        },
+        async finalMessage() {
+          // Fire text events
+          for (const cb of handlers["text"] ?? []) cb("Hello");
+          return fakeResponse;
+        },
+      };
+    };
     messages = {
       stream: (
         params: Record<string, unknown>,
         options?: Record<string, unknown>,
-      ) => {
-        lastStreamParams = JSON.parse(JSON.stringify(params));
-        _lastStreamOptions = options ?? null;
-        const handlers: Record<string, ((...args: unknown[]) => void)[]> = {};
-        return {
-          on(event: string, cb: (...args: unknown[]) => void) {
-            (handlers[event] ??= []).push(cb);
-            return this;
-          },
-          async finalMessage() {
-            // Fire text events
-            for (const cb of handlers["text"] ?? []) cb("Hello");
-            return fakeResponse;
-          },
-        };
+      ) => this.#streamImpl(params, options),
+    };
+    beta = {
+      messages: {
+        stream: (
+          params: Record<string, unknown>,
+          options?: Record<string, unknown>,
+        ) => this.#streamImpl(params, options),
       },
     };
   },
@@ -138,16 +150,16 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
   // -----------------------------------------------------------------------
   // System prompt cache control
   // -----------------------------------------------------------------------
-  test("system prompt has cache_control ephemeral", async () => {
+  test("system prompt has cache_control ephemeral with 1h TTL", async () => {
     await provider.sendMessage([userMsg("Hi")], undefined, "You are helpful.");
 
     const system = lastStreamParams!.system as Array<{
       type: string;
       text: string;
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
     expect(system).toHaveLength(1);
-    expect(system[0].cache_control).toEqual({ type: "ephemeral" });
+    expect(system[0].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
   });
 
   test("no system param when system prompt is omitted", async () => {
@@ -166,13 +178,13 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
     const system = lastStreamParams!.system as Array<{
       type: string;
       text: string;
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
     expect(system).toHaveLength(2);
     expect(system[0].text).toBe(staticBlock);
-    expect(system[0].cache_control).toEqual({ type: "ephemeral" });
+    expect(system[0].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
     expect(system[1].text).toBe(dynamicBlock);
-    expect(system[1].cache_control).toEqual({ type: "ephemeral" });
+    expect(system[1].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
   });
 
   // -----------------------------------------------------------------------
@@ -183,7 +195,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
 
     const tools = lastStreamParams!.tools as Array<{
       name: string;
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
     expect(tools).toHaveLength(3);
 
@@ -192,7 +204,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
     expect(tools[1].cache_control).toBeUndefined();
 
     // Last tool: cache_control ephemeral
-    expect(tools[2].cache_control).toEqual({ type: "ephemeral" });
+    expect(tools[2].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
   });
 
   test("single tool gets cache_control", async () => {
@@ -200,10 +212,10 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
 
     const tools = lastStreamParams!.tools as Array<{
       name: string;
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
     expect(tools).toHaveLength(1);
-    expect(tools[0].cache_control).toEqual({ type: "ephemeral" });
+    expect(tools[0].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
   });
 
   test("no tools param when tools are omitted", async () => {
@@ -223,7 +235,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const lastUser = messages[messages.length - 1];
@@ -248,7 +260,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
 
@@ -264,7 +276,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
     // Second user turn (second-to-last): cache_control ephemeral
     const secondUserLastBlock =
       userMessages[1].content[userMessages[1].content.length - 1];
-    expect(secondUserLastBlock.cache_control).toEqual({ type: "ephemeral" });
+    expect(secondUserLastBlock.cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
 
     // Third user turn (last): no cache_control
     const thirdUserLastBlock =
@@ -280,7 +292,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const userMessages = sent.filter((m) => m.role === "user");
@@ -303,12 +315,12 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
 
     const sent = lastStreamParams!.messages as Array<{
       role: string;
-      content: Array<{ type: string; cache_control?: { type: string } }>;
+      content: Array<{ type: string; cache_control?: { type: string; ttl?: string } }>;
     }>;
     const userMsgs = sent.filter((m) => m.role === "user");
     // First user msg (second-to-last) should get cache
     const firstLast = userMsgs[0].content[userMsgs[0].content.length - 1];
-    expect(firstLast.cache_control).toEqual({ type: "ephemeral" });
+    expect(firstLast.cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
     // tool_result msg (last) should NOT get cache
     const secondLast = userMsgs[1].content[userMsgs[1].content.length - 1];
     expect(secondLast.cache_control).toBeUndefined();
@@ -330,7 +342,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const assistantMsgs = sent.filter((m) => m.role === "assistant");
@@ -361,7 +373,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const user = sent[0];
@@ -402,7 +414,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const user = sent[0];
@@ -436,7 +448,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const user = sent[0];
@@ -1309,7 +1321,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
       content: Array<{
         type: string;
         text: string;
-        cache_control?: { type: string };
+        cache_control?: { type: string; ttl?: string };
       }>;
     }>;
     const userMsgs = sent.filter((m) => m.role === "user");
@@ -1321,7 +1333,7 @@ describe("AnthropicProvider — Cache-Control Characterization", () => {
 
     // Turn 2 (second-to-last): cache on last block only
     expect(userMsgs[1].content[0].cache_control).toBeUndefined();
-    expect(userMsgs[1].content[1].cache_control).toEqual({ type: "ephemeral" });
+    expect(userMsgs[1].content[1].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
 
     // Turn 3 (last): no cache
     expect(userMsgs[2].content[0].cache_control).toBeUndefined();
@@ -1548,16 +1560,17 @@ describe("AnthropicProvider — Managed Proxy Fallback", () => {
 
     // System prompt cache control
     const system = lastStreamParams!.system as Array<{
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
-    expect(system[0].cache_control).toEqual({ type: "ephemeral" });
+    expect(system[0].cache_control).toEqual({ type: "ephemeral", ttl: "1h" });
 
     // Last tool cache control
     const tools = lastStreamParams!.tools as Array<{
-      cache_control?: { type: string };
+      cache_control?: { type: string; ttl?: string };
     }>;
     expect(tools[tools.length - 1].cache_control).toEqual({
       type: "ephemeral",
+      ttl: "1h",
     });
   });
 });

package/src/__tests__/app-git-history.test.ts

@@ -1,28 +1,20 @@
-import { existsSync, mkdirSync, readFileSync, rmSync } from "node:fs";
+import { existsSync, readFileSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
 import { _resetGitServiceRegistry } from "../workspace/git-service.js";
 
-// Mock getDataDir to use a temp directory
 let testDataDir: string;
 
-mock.module("../util/platform.js", () => ({
-  getDataDir: () => testDataDir,
-  getProjectDir: () => testDataDir,
-}));
-
-// Re-import after mocking so modules use our temp dir
-const { createApp, updateApp, getAppDirPath } =
-  await import("../memory/app-store.js");
-const {
-  getAppHistory,
+import {
+  commitAppTurnChanges,
   getAppDiff,
   getAppFileAtVersion,
+  getAppHistory,
   restoreAppVersion,
-  commitAppTurnChanges,
-} = await import("../memory/app-git-service.js");
+} from "../memory/app-git-service.js";
+import { createApp, getAppDirPath, updateApp } from "../memory/app-store.js";
 
 describe("App Git History", () => {
   beforeEach(() => {
@@ -32,7 +24,7 @@ describe("App Git History", () => {
         .toString(36)
         .slice(2)}`,
     );
-    mkdirSync(join(testDataDir, "apps"), { recursive: true });
+    process.env.VELLUM_WORKSPACE_DIR = testDataDir;
     _resetGitServiceRegistry();
   });
 
@@ -55,7 +47,7 @@ describe("App Git History", () => {
 
     const history = await getAppHistory(app.id);
     expect(history.length).toBeGreaterThanOrEqual(2);
-    expect(history[0].message).toContain("Turn 2");
+    expect(history[0].message).toContain("update ");
     expect(history[0].commitHash).toMatch(/^[0-9a-f]+$/);
     expect(history[0].timestamp).toBeGreaterThan(0);
   });

package/src/__tests__/app-git-service.test.ts

@@ -1,29 +1,21 @@
 import { execFileSync } from "node:child_process";
-import { existsSync, mkdirSync, readFileSync, rmSync } from "node:fs";
+import { existsSync, readFileSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
 import { commitAppTurnChanges } from "../memory/app-git-service.js";
-import { _resetGitServiceRegistry } from "../workspace/git-service.js";
-
-// Mock getDataDir to use a temp directory
-let testDataDir: string;
-
-mock.module("../util/platform.js", () => ({
-  getDataDir: () => testDataDir,
-  getProjectDir: () => testDataDir,
-}));
-
-// Re-import app-store after mocking so it uses our temp dir
-const {
+import {
   createApp,
-  updateApp,
   deleteApp,
-  writeAppFile,
   editAppFile,
   getAppsDir,
-} = await import("../memory/app-store.js");
+  updateApp,
+  writeAppFile,
+} from "../memory/app-store.js";
+import { _resetGitServiceRegistry } from "../workspace/git-service.js";
+
+let testDataDir: string;
 
 describe("App Git Service", () => {
   beforeEach(() => {
@@ -31,7 +23,7 @@ describe("App Git Service", () => {
       tmpdir(),
       `vellum-app-git-test-${Date.now()}-${Math.random().toString(36).slice(2)}`,
     );
-    mkdirSync(join(testDataDir, "apps"), { recursive: true });
+    process.env.VELLUM_WORKSPACE_DIR = testDataDir;
     _resetGitServiceRegistry();
   });
 
@@ -100,7 +92,9 @@ describe("App Git Service", () => {
     expect(commits.length).toBeLessThanOrEqual(2);
     // The turn commit message should appear (or files are in the initial commit)
     expect(
-      commits.some((c) => c.includes("Turn 1") || c.includes("Initial commit")),
+      commits.some(
+        (c) => c.includes("update ") || c.includes("Initial commit"),
+      ),
     ).toBe(true);
   });
 
@@ -141,6 +135,6 @@ describe("App Git Service", () => {
 
     const appsDir = getAppsDir();
     const commits = getGitLog(appsDir);
-    expect(commits[0]).toContain("Turn 2: app changes");
+    expect(commits[0]).toContain("update ");
   });
 });

package/src/__tests__/app-store-dir-names.test.ts

@@ -7,32 +7,22 @@ import {
 } from "node:fs";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import { afterEach, beforeEach, describe, expect, mock, test } from "bun:test";
-
-// ---------------------------------------------------------------------------
-// Mock getDataDir to use a temp directory
-// ---------------------------------------------------------------------------
+import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
 let testDataDir: string;
 
-mock.module("../util/platform.js", () => ({
-  getDataDir: () => testDataDir,
-  getProjectDir: () => testDataDir,
-}));
-
-// Re-import app-store after mocking so it uses our temp dir
-const {
-  slugify,
+import {
+  createApp,
+  deleteApp,
   generateAppDirName,
-  validateDirName,
-  resolveAppDir,
+  getApp,
   getAppDirPath,
   getAppsDir,
-  createApp,
-  getApp,
+  resolveAppDir,
+  slugify,
   updateApp,
-  deleteApp,
-} = await import("../memory/app-store.js");
+  validateDirName,
+} from "../memory/app-store.js";
 
 // ---------------------------------------------------------------------------
 // Helpers
@@ -59,7 +49,7 @@ function makeAppParams(name: string) {
 
 beforeEach(() => {
   testDataDir = freshTempDir();
-  mkdirSync(join(testDataDir, "apps"), { recursive: true });
+  process.env.VELLUM_WORKSPACE_DIR = testDataDir;
 });
 
 afterEach(() => {

package/src/__tests__/approval-cascade.test.ts

@@ -5,10 +5,7 @@
  * allow_conversation, always_allow, always_deny), other pending confirmations in
  * the same conversation that match are auto-resolved.
  */
-import { mkdtempSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 import { Minimatch } from "minimatch";
 
@@ -22,8 +19,6 @@ import type { ConfirmationStateChanged } from "../daemon/message-types/messages.
 import type { Message, ProviderResponse } from "../providers/types.js";
 import type { ConfirmationDetails } from "../runtime/pending-interactions.js";
 
-const testDir = mkdtempSync(join(tmpdir(), "approval-cascade-test-"));
-
 // ---------------------------------------------------------------------------
 // Mocks — must precede Conversation import
 // ---------------------------------------------------------------------------
@@ -49,10 +44,6 @@ mock.module("../util/logger.js", () => ({
   getLogger: () => makeLoggerStub(),
 }));
 
-mock.module("../util/platform.js", () => ({
-  getDataDir: () => testDir,
-}));
-
 mock.module("../memory/guardian-action-store.js", () => ({
   getGuardianActionRequest: () => null,
   resolveGuardianActionRequest: () => {},
@@ -265,7 +256,7 @@ function makeConversation(
     "system prompt",
     4096,
     sendToClient ?? (() => {}),
-    testDir,
+    process.env.VELLUM_WORKSPACE_DIR!,
   );
 }
 
@@ -325,14 +316,6 @@ function makeConfirmationDetails(patterns: string[]): ConfirmationDetails {
   };
 }
 
-afterAll(() => {
-  try {
-    rmSync(testDir, { recursive: true, force: true });
-  } catch {
-    /* best effort */
-  }
-});
-
 beforeEach(() => {
   pendingInteractions.clear();
 });

package/src/__tests__/approval-primitive.test.ts

@@ -1,20 +1,4 @@
-import { mkdtempSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
-
-const testDir = mkdtempSync(join(tmpdir(), "approval-primitive-test-"));
-
-mock.module("../util/platform.js", () => ({
-  getDataDir: () => testDir,
-  isMacOS: () => process.platform === "darwin",
-  isLinux: () => process.platform === "linux",
-  isWindows: () => process.platform === "win32",
-  getPidPath: () => join(testDir, "test.pid"),
-  getDbPath: () => join(testDir, "test.db"),
-  getLogPath: () => join(testDir, "test.log"),
-  ensureDataDir: () => {},
-}));
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -29,7 +13,7 @@ import {
   mintGrantFromDecision,
   type MintGrantParams,
 } from "../approvals/approval-primitive.js";
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { scopedApprovalGrants } from "../memory/schema.js";
 import { computeToolApprovalDigest } from "../security/tool-approval-digest.js";
 
@@ -40,15 +24,6 @@ function clearTables(): void {
   db.delete(scopedApprovalGrants).run();
 }
 
-afterAll(() => {
-  resetDb();
-  try {
-    rmSync(testDir, { recursive: true });
-  } catch {
-    /* best effort */
-  }
-});
-
 // ---------------------------------------------------------------------------
 // Helper to build mint params with sensible defaults
 // ---------------------------------------------------------------------------

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

@@ -4,30 +4,11 @@
  * Tests POST /v1/confirm, POST /v1/secret, and POST /v1/trust-rules
  * through RuntimeHttpServer with pending-interactions tracking.
  */
-import { mkdtempSync, realpathSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 import type { Conversation } from "../daemon/conversation.js";
 import type { ServerMessage } from "../daemon/message-protocol.js";
 
-const testDir = realpathSync(
-  mkdtempSync(join(tmpdir(), "approval-routes-http-test-")),
-);
-
-mock.module("../util/platform.js", () => ({
-  getRootDir: () => testDir,
-  getDataDir: () => testDir,
-  isMacOS: () => process.platform === "darwin",
-  isLinux: () => process.platform === "linux",
-  isWindows: () => process.platform === "win32",
-  getPidPath: () => join(testDir, "test.pid"),
-  getDbPath: () => join(testDir, "test.db"),
-  getLogPath: () => join(testDir, "test.log"),
-  ensureDataDir: () => {},
-}));
-
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
     new Proxy({} as Record<string, unknown>, {
@@ -86,7 +67,7 @@ mock.module("../permissions/trust-store.js", () => ({
   getRules: () => [],
 }));
 
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { AssistantEventHub } from "../runtime/assistant-event-hub.js";
 import { RuntimeHttpServer } from "../runtime/http-server.js";
 import * as pendingInteractions from "../runtime/pending-interactions.js";
@@ -248,15 +229,6 @@ describe("standalone approval endpoints — HTTP layer", () => {
     eventHub = new AssistantEventHub();
   });
 
-  afterAll(() => {
-    resetDb();
-    try {
-      rmSync(testDir, { recursive: true, force: true });
-    } catch {
-      /* best effort */
-    }
-  });
-
   async function startServer(
     conversationFactory: () => Conversation,
   ): Promise<void> {

package/src/__tests__/assistant-events-sse-hardening.test.ts

@@ -8,24 +8,7 @@
  *   - Subscription cleanup on request abort.
  *   - Subscription cleanup on reader cancel.
  */
-import { mkdtempSync, realpathSync, rmSync } from "node:fs";
-import { tmpdir } from "node:os";
-import { join } from "node:path";
-import { afterAll, beforeEach, describe, expect, mock, test } from "bun:test";
-
-const testDir = realpathSync(mkdtempSync(join(tmpdir(), "sse-hardening-")));
-
-mock.module("../util/platform.js", () => ({
-  getRootDir: () => testDir,
-  getDataDir: () => testDir,
-  isMacOS: () => process.platform === "darwin",
-  isLinux: () => process.platform === "linux",
-  isWindows: () => process.platform === "win32",
-  getPidPath: () => join(testDir, "test.pid"),
-  getDbPath: () => join(testDir, "test.db"),
-  getLogPath: () => join(testDir, "test.log"),
-  ensureDataDir: () => {},
-}));
+import { beforeEach, describe, expect, mock, test } from "bun:test";
 
 mock.module("../util/logger.js", () => ({
   getLogger: () =>
@@ -46,21 +29,12 @@ mock.module("../config/loader.js", () => ({
   }),
 }));
 
-import { getDb, initializeDb, resetDb } from "../memory/db.js";
+import { getDb, initializeDb } from "../memory/db.js";
 import { AssistantEventHub } from "../runtime/assistant-event-hub.js";
 import { handleSubscribeAssistantEvents } from "../runtime/routes/events-routes.js";
 
 initializeDb();
 
-afterAll(() => {
-  resetDb();
-  try {
-    rmSync(testDir, { recursive: true, force: true });
-  } catch {
-    /* best effort */
-  }
-});
-
 // ── Helpers ───────────────────────────────────────────────────────────────────
 
 function clearTables() {