@vellumai/assistant 0.8.5 → 0.8.7

package/src/__tests__/conversation-workspace-injection.test.ts

@@ -1,6 +1,7 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
 
-import type { AgentEvent } from "../agent/loop.js";
+import { CompactionCircuit } from "../agent/compaction-circuit.js";
+import type { AgentEvent, AgentLoopRunResult } from "../agent/loop.js";
 import type { Message, ProviderResponse } from "../providers/types.js";
 
 // ---------------------------------------------------------------------------
@@ -145,6 +146,8 @@ mock.module("../memory/conversation-crud.js", () => ({
   setLastNotifiedInferenceProfile: () => {},
   getConversationOverrideProfileFromRow: () => undefined,
   updateMessageMetadata: () => {},
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
+  updateMessageContent: mock(() => {}),
 }));
 
 mock.module("../memory/conversation-queries.js", () => ({
@@ -230,6 +233,7 @@ mock.module("../workspace/turn-commit.js", () => ({
 
 mock.module("../agent/loop.js", () => ({
   AgentLoop: class {
+    compactionCircuit = new CompactionCircuit("test-conv");
     constructor() {}
     getToolTokenBudget() {
       return 0;
@@ -243,7 +247,10 @@ mock.module("../agent/loop.js", () => ({
     async run(
       messages: Message[],
       onEvent: (event: AgentEvent) => void,
-    ): Promise<Message[]> {
+    ): Promise<AgentLoopRunResult> {
+      // Prime the assistant row anchor — production code emits this from
+      // `AgentLoop.run` just before `provider.sendMessage`.
+      await onEvent({ type: "llm_call_started" });
       runCalls.push(messages);
       agentLoopScript(onEvent);
       onEvent({
@@ -258,7 +265,7 @@ mock.module("../agent/loop.js", () => ({
         content: [{ type: "text", text: "ok" }],
       };
       onEvent({ type: "message_complete", message: assistantMessage });
-      return [...messages, assistantMessage];
+      return { history: [...messages, assistantMessage], exitReason: null };
     }
   },
 }));
@@ -283,6 +290,7 @@ mock.module("../memory/canonical-guardian-store.js", () => ({
 }));
 
 import { Conversation } from "../daemon/conversation.js";
+import { resetPluginRegistryAndRegisterDefaults } from "../plugins/defaults/index.js";
 
 function makeConversation(): Conversation {
   const provider = {
@@ -300,9 +308,9 @@ function makeConversation(): Conversation {
     "conv-1",
     provider,
     "system prompt",
-    4096,
     () => {},
     "/tmp",
+    { maxTokens: 4096 },
   );
 }
 
@@ -322,13 +330,14 @@ describe("Conversation workspace injection", () => {
     runCalls = [];
     agentLoopScript = () => {};
     scanCallCount = 0;
+    resetPluginRegistryAndRegisterDefaults();
   });
 
   test("runtime messages include workspace top-level context", async () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
 
     expect(runCalls).toHaveLength(1);
     const runtimeUser = runCalls[0][runCalls[0].length - 1];
@@ -342,7 +351,7 @@ describe("Conversation workspace injection", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
 
     expect(runCalls).toHaveLength(1);
     const runtimeUser = runCalls[0][runCalls[0].length - 1];
@@ -356,7 +365,7 @@ describe("Conversation workspace injection", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
 
     expect(runCalls).toHaveLength(1);
     const runtimeUser = runCalls[0][runCalls[0].length - 1];
@@ -370,7 +379,7 @@ describe("Conversation workspace injection", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
 
     // Workspace blocks use <workspace> tag which is intentionally NOT stripped.
     const persistedMessages = conversation.getMessages();
@@ -385,14 +394,17 @@ describe("Conversation workspace injection", () => {
     await conversation.loadFromDb();
 
     // First message — workspace is injected
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
     expect(runCalls).toHaveLength(1);
     const firstCallUser = runCalls[0][runCalls[0].length - 1];
     const firstText = messageText(firstCallUser);
     expect(firstText).toContain("<workspace>");
 
     // Second message — workspace is NOT injected (not first message, no compaction)
-    await conversation.processMessage("Follow up", [], () => {});
+    await conversation.processMessage({
+      content: "Follow up",
+      attachments: [],
+    });
     expect(runCalls).toHaveLength(2);
     const secondCallUser = runCalls[1][runCalls[1].length - 1];
     const secondText = messageText(secondCallUser);
@@ -405,13 +417,14 @@ describe("Conversation workspace dirty-refresh E2E", () => {
     runCalls = [];
     agentLoopScript = () => {};
     scanCallCount = 0;
+    resetPluginRegistryAndRegisterDefaults();
   });
 
   test("first turn computes snapshot", async () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
 
     expect(scanCallCount).toBe(1);
     const text = messageText(runCalls[0][runCalls[0].length - 1]);
@@ -422,10 +435,10 @@ describe("Conversation workspace dirty-refresh E2E", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
     const afterFirst = scanCallCount;
 
-    await conversation.processMessage("Again", [], () => {});
+    await conversation.processMessage({ content: "Again", attachments: [] });
 
     // Scanner should NOT have been called again
     expect(scanCallCount).toBe(afterFirst);
@@ -435,7 +448,7 @@ describe("Conversation workspace dirty-refresh E2E", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
     const afterFirst = scanCallCount;
 
     // Simulate a turn where the agent uses file_edit
@@ -448,7 +461,10 @@ describe("Conversation workspace dirty-refresh E2E", () => {
         isError: false,
       });
     };
-    await conversation.processMessage("Edit a file", [], () => {});
+    await conversation.processMessage({
+      content: "Edit a file",
+      attachments: [],
+    });
 
     // No rescan should happen during the mutation turn itself
     const afterMutation = scanCallCount;
@@ -456,7 +472,10 @@ describe("Conversation workspace dirty-refresh E2E", () => {
 
     // Next turn should trigger exactly one fresh scan
     agentLoopScript = () => {};
-    await conversation.processMessage("What happened?", [], () => {});
+    await conversation.processMessage({
+      content: "What happened?",
+      attachments: [],
+    });
 
     expect(scanCallCount).toBe(afterMutation + 1);
   });
@@ -465,7 +484,7 @@ describe("Conversation workspace dirty-refresh E2E", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
     const afterFirst = scanCallCount;
 
     agentLoopScript = (onEvent) => {
@@ -477,7 +496,10 @@ describe("Conversation workspace dirty-refresh E2E", () => {
         isError: false,
       });
     };
-    await conversation.processMessage("Run a command", [], () => {});
+    await conversation.processMessage({
+      content: "Run a command",
+      attachments: [],
+    });
 
     // No rescan should happen during the mutation turn itself
     const afterMutation = scanCallCount;
@@ -485,7 +507,10 @@ describe("Conversation workspace dirty-refresh E2E", () => {
 
     // Next turn should trigger exactly one fresh scan
     agentLoopScript = () => {};
-    await conversation.processMessage("What now?", [], () => {});
+    await conversation.processMessage({
+      content: "What now?",
+      attachments: [],
+    });
 
     expect(scanCallCount).toBe(afterMutation + 1);
   });
@@ -494,7 +519,7 @@ describe("Conversation workspace dirty-refresh E2E", () => {
     const conversation = makeConversation();
     await conversation.loadFromDb();
 
-    await conversation.processMessage("Hello", [], () => {});
+    await conversation.processMessage({ content: "Hello", attachments: [] });
     const afterFirst = scanCallCount;
 
     agentLoopScript = (onEvent) => {
@@ -506,10 +531,16 @@ describe("Conversation workspace dirty-refresh E2E", () => {
         isError: true,
       });
     };
-    await conversation.processMessage("Try editing", [], () => {});
+    await conversation.processMessage({
+      content: "Try editing",
+      attachments: [],
+    });
 
     agentLoopScript = () => {};
-    await conversation.processMessage("What happened?", [], () => {});
+    await conversation.processMessage({
+      content: "What happened?",
+      attachments: [],
+    });
 
     // Scanner should NOT have been re-called since the tool failed
     expect(scanCallCount).toBe(afterFirst);

package/src/__tests__/conversation-workspace-tool-tracking.test.ts

@@ -1,5 +1,6 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
 
+import { CompactionCircuit } from "../agent/compaction-circuit.js";
 import type { AgentEvent } from "../agent/loop.js";
 import type { Message, ProviderResponse } from "../providers/types.js";
 
@@ -142,6 +143,8 @@ mock.module("../memory/conversation-crud.js", () => ({
   setLastNotifiedInferenceProfile: () => {},
   getConversationOverrideProfileFromRow: () => undefined,
   updateMessageMetadata: () => {},
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
+  updateMessageContent: mock(() => {}),
 }));
 
 mock.module("../memory/conversation-queries.js", () => ({
@@ -214,6 +217,7 @@ mock.module("../workspace/turn-commit.js", () => ({
 
 mock.module("../agent/loop.js", () => ({
   AgentLoop: class {
+    compactionCircuit = new CompactionCircuit("test-conv");
     constructor() {}
     getToolTokenBudget() {
       return 0;
@@ -228,6 +232,9 @@ mock.module("../agent/loop.js", () => ({
       messages: Message[],
       onEvent: (event: AgentEvent) => void,
     ): Promise<Message[]> {
+      // Prime the assistant row anchor — production code emits this from
+      // `AgentLoop.run` just before `provider.sendMessage`.
+      await onEvent({ type: "llm_call_started" });
       agentLoopScript(onEvent);
       onEvent({
         type: "usage",
@@ -283,9 +290,9 @@ function makeConversation(): Conversation {
     "conv-1",
     provider,
     "system prompt",
-    4096,
     () => {},
     "/tmp",
+    { maxTokens: 4096 },
   );
 }
 
@@ -321,7 +328,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Write a file", [], () => {});
+    await conversation.processMessage({
+      content: "Write a file",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(true);
   });
 
@@ -347,7 +357,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Edit a file", [], () => {});
+    await conversation.processMessage({
+      content: "Edit a file",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(true);
   });
 
@@ -375,7 +388,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Write a file", [], () => {});
+    await conversation.processMessage({
+      content: "Write a file",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(true);
   });
 
@@ -401,7 +417,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Run a command", [], () => {});
+    await conversation.processMessage({
+      content: "Run a command",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(true);
   });
 
@@ -427,7 +446,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Run a command", [], () => {});
+    await conversation.processMessage({
+      content: "Run a command",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(true);
   });
 
@@ -453,7 +475,10 @@ describe("Conversation workspace dirty on file mutations", () => {
       });
     };
 
-    await conversation.processMessage("Read a file", [], () => {});
+    await conversation.processMessage({
+      content: "Read a file",
+      attachments: [],
+    });
     expect(conversation.isWorkspaceTopLevelDirty()).toBe(false);
   });
 });

package/src/__tests__/credential-broker-browser-fill.test.ts

@@ -28,8 +28,8 @@ mock.module("../util/logger.js", () => ({
 // Use encrypted backend with a temp store path
 // ---------------------------------------------------------------------------
 
-import { _setStorePath } from "../security/encrypted-store.js";
 import { _resetBackend } from "../security/secure-keys.js";
+import { setStorePathForTesting } from "./encrypted-store-test-helpers.js";
 
 const TEST_DIR = join(
   tmpdir(),
@@ -74,7 +74,7 @@ describe("CredentialBroker.browserFill", () => {
     for (const entry of readdirSync(TEST_DIR)) {
       rmSync(join(TEST_DIR, entry), { recursive: true, force: true });
     }
-    _setStorePath(STORE_PATH);
+    setStorePathForTesting(STORE_PATH);
     _resetBackend();
     _setMetadataPath(join(TEST_DIR, "metadata.json"));
     broker = new CredentialBroker();
@@ -82,7 +82,7 @@ describe("CredentialBroker.browserFill", () => {
 
   afterEach(() => {
     _setMetadataPath(null);
-    _setStorePath(null);
+    setStorePathForTesting(null);
     _resetBackend();
   });
 

package/src/__tests__/credential-broker-server-use.test.ts

@@ -27,8 +27,8 @@ mock.module("../util/logger.js", () => ({
 // Use encrypted backend with a temp store path
 // ---------------------------------------------------------------------------
 
-import { _setStorePath } from "../security/encrypted-store.js";
 import { _resetBackend } from "../security/secure-keys.js";
+import { setStorePathForTesting } from "./encrypted-store-test-helpers.js";
 
 const TEST_DIR = join(
   tmpdir(),
@@ -69,7 +69,7 @@ describe("CredentialBroker.serverUse", () => {
 
   beforeEach(() => {
     mkdirSync(TEST_DIR, { recursive: true });
-    _setStorePath(STORE_PATH);
+    setStorePathForTesting(STORE_PATH);
     _resetBackend();
     _setMetadataPath(join(TEST_DIR, "metadata.json"));
     broker = new CredentialBroker();
@@ -77,7 +77,7 @@ describe("CredentialBroker.serverUse", () => {
 
   afterEach(() => {
     _setMetadataPath(null);
-    _setStorePath(null);
+    setStorePathForTesting(null);
     _resetBackend();
     rmSync(TEST_DIR, { recursive: true, force: true });
   });
@@ -464,7 +464,7 @@ describe("CredentialBroker.serverUseById", () => {
 
   beforeEach(() => {
     mkdirSync(TEST_DIR, { recursive: true });
-    _setStorePath(STORE_PATH);
+    setStorePathForTesting(STORE_PATH);
     _resetBackend();
     _setMetadataPath(join(TEST_DIR, "metadata.json"));
     broker = new CredentialBroker();
@@ -472,7 +472,7 @@ describe("CredentialBroker.serverUseById", () => {
 
   afterEach(() => {
     _setMetadataPath(null);
-    _setStorePath(null);
+    setStorePathForTesting(null);
     _resetBackend();
     rmSync(TEST_DIR, { recursive: true, force: true });
   });

package/src/__tests__/credential-execution-client.test.ts

@@ -8,7 +8,15 @@
  * 4. The CES RPC client correctly frames requests and validates responses.
  */
 
-import { readdirSync, readFileSync, statSync } from "node:fs";
+import {
+  mkdtempSync,
+  readdirSync,
+  readFileSync,
+  rmSync,
+  statSync,
+  writeFileSync,
+} from "node:fs";
+import { tmpdir } from "node:os";
 import { join, resolve } from "node:path";
 import { describe, expect, test } from "bun:test";
 
@@ -22,6 +30,7 @@ import {
   createCesClient,
 } from "../credential-execution/client.js";
 import {
+  discoverCesWithRetry,
   discoverLocalCes,
   discoverManagedCes,
 } from "../credential-execution/executable-discovery.js";
@@ -147,6 +156,68 @@ describe("managed CES discovery", () => {
   });
 });
 
+// ---------------------------------------------------------------------------
+// Managed discovery retry — absorbs the sidecar's socket re-bind window
+// ---------------------------------------------------------------------------
+
+describe("discoverCesWithRetry", () => {
+  function withManagedEnv(socketPath: string): () => void {
+    const savedSocket = process.env["CES_BOOTSTRAP_SOCKET"];
+    const savedSocketDir = process.env["CES_BOOTSTRAP_SOCKET_DIR"];
+    const savedContainerized = process.env["IS_CONTAINERIZED"];
+    delete process.env["CES_BOOTSTRAP_SOCKET_DIR"];
+    process.env["CES_BOOTSTRAP_SOCKET"] = socketPath;
+    process.env["IS_CONTAINERIZED"] = "true";
+    return () => {
+      const restore = (key: string, value: string | undefined) => {
+        if (value !== undefined) process.env[key] = value;
+        else delete process.env[key];
+      };
+      restore("CES_BOOTSTRAP_SOCKET", savedSocket);
+      restore("CES_BOOTSTRAP_SOCKET_DIR", savedSocketDir);
+      restore("IS_CONTAINERIZED", savedContainerized);
+    };
+  }
+
+  test("returns unavailable after polling when the socket never appears", async () => {
+    const socketPath = join(tmpdir(), `ces-retry-missing-${Date.now()}.sock`);
+    const restore = withManagedEnv(socketPath);
+    try {
+      const start = Date.now();
+      const result = await discoverCesWithRetry({
+        timeoutMs: 200,
+        intervalMs: 20,
+      });
+      expect(result.mode).toBe("unavailable");
+      // It must actually have waited rather than failing on the first probe.
+      expect(Date.now() - start).toBeGreaterThanOrEqual(150);
+    } finally {
+      restore();
+    }
+  });
+
+  test("resolves to managed once the socket is re-bound mid-poll", async () => {
+    const dir = mkdtempSync(join(tmpdir(), "ces-retry-"));
+    const socketPath = join(dir, "ces.sock");
+    const restore = withManagedEnv(socketPath);
+    try {
+      // Simulate the sidecar re-binding its bootstrap socket shortly after
+      // the assistant begins probing.
+      setTimeout(() => writeFileSync(socketPath, ""), 80);
+
+      const result = await discoverCesWithRetry({
+        timeoutMs: 2_000,
+        intervalMs: 20,
+      });
+      expect(result.mode).toBe("managed");
+      expect((result as { socketPath: string }).socketPath).toBe(socketPath);
+    } finally {
+      restore();
+      rmSync(dir, { recursive: true, force: true });
+    }
+  });
+});
+
 // ---------------------------------------------------------------------------
 // CES client — transport and framing
 // ---------------------------------------------------------------------------

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

@@ -10,10 +10,7 @@
 
 import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
-import {
-  _setOverridesForTesting,
-  isAssistantFeatureFlagEnabled,
-} from "../config/assistant-feature-flags.js";
+import { isAssistantFeatureFlagEnabled } from "../config/assistant-feature-flags.js";
 import type { AssistantConfig } from "../config/schema.js";
 import {
   CES_GRANT_AUDIT_FLAG_KEY,
@@ -25,20 +22,21 @@ import {
   isCesShellLockdownEnabled,
   isCesToolsEnabled,
 } from "../credential-execution/feature-gates.js";
+import { setOverridesForTesting } from "./feature-flag-test-helpers.js";
 
 beforeEach(() => {
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 afterEach(() => {
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 // ---------------------------------------------------------------------------
 // Helpers
 // ---------------------------------------------------------------------------
 
-/** Create a minimal AssistantConfig (flag overrides are now set via _setOverridesForTesting). */
+/** Create a minimal AssistantConfig (flag overrides are now set via setOverridesForTesting). */
 function makeConfig(): AssistantConfig {
   return {} as AssistantConfig;
 }
@@ -120,13 +118,13 @@ describe("CES flags match registry defaults", () => {
 describe("CES flags can be toggled independently", () => {
   for (const { name, fn, key } of ALL_CES_PREDICATES) {
     test(`enabling ${key} makes ${name} return true`, () => {
-      _setOverridesForTesting({ [key]: true });
+      setOverridesForTesting({ [key]: true });
       const config = makeConfig();
       expect(fn(config)).toBe(true);
     });
 
     test(`enabling ${key} does not change other CES flags from their defaults`, () => {
-      _setOverridesForTesting({ [key]: true });
+      setOverridesForTesting({ [key]: true });
       const config = makeConfig();
       for (const {
         fn: otherFn,
@@ -147,7 +145,7 @@ describe("CES flags can be toggled independently", () => {
 describe("CES flags respect explicit false overrides", () => {
   for (const { name, fn, key } of ALL_CES_PREDICATES) {
     test(`${name} returns false when explicitly set to false`, () => {
-      _setOverridesForTesting({ [key]: false });
+      setOverridesForTesting({ [key]: false });
       const config = makeConfig();
       expect(fn(config)).toBe(false);
     });
@@ -164,7 +162,7 @@ describe("CES flags do not affect unrelated flags", () => {
     for (const key of ALL_CES_FLAG_KEYS) {
       overrides[key] = true;
     }
-    _setOverridesForTesting(overrides);
+    setOverridesForTesting(overrides);
     const config = makeConfig();
 
     // account-deletion defaults to true in the registry and should stay true.
@@ -178,7 +176,7 @@ describe("CES flags do not affect unrelated flags", () => {
     for (const key of ALL_CES_FLAG_KEYS) {
       overrides[key] = true;
     }
-    _setOverridesForTesting(overrides);
+    setOverridesForTesting(overrides);
     const config = makeConfig();
 
     // Unknown flags fail closed unless explicitly overridden.

package/src/__tests__/credential-execution-tools.test.ts

@@ -5,7 +5,6 @@ import { makeAuthenticatedRequestTool } from "../tools/credential-execution/make
 import { manageSecureCommandTool } from "../tools/credential-execution/manage-secure-command-tool.js";
 import { runAuthenticatedCommandTool } from "../tools/credential-execution/run-authenticated-command.js";
 import { cesTools, getCesToolsIfEnabled } from "../tools/tool-manifest.js";
-import type { Tool } from "../tools/types.js";
 
 // ---------------------------------------------------------------------------
 // Schema shape tests
@@ -79,7 +78,7 @@ describe("CES tool schema shapes", () => {
 describe("CES tool manifest registration", () => {
   test("cesTools contains exactly three CES tools", () => {
     expect(cesTools).toHaveLength(3);
-    const names = cesTools.map((t: Tool) => t.name);
+    const names = cesTools.map((t) => t.name);
     expect(names).toContain("make_authenticated_request");
     expect(names).toContain("run_authenticated_command");
     expect(names).toContain("manage_secure_command_tool");