@vellumai/assistant 0.8.5 → 0.8.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@vellumai/assistant",
-  "version": "0.8.5",
+  "version": "0.8.6",
   "license": "MIT",
   "type": "module",
   "exports": {

package/src/__tests__/actor-token-service.test.ts

@@ -29,7 +29,7 @@ mock.module("../config/env.js", () => ({
   checkUnrecognizedEnvVars: () => {},
 }));
 
-import { getDb, resetDb } from "../memory/db-connection.js";
+import { getDb } from "../memory/db-connection.js";
 import { initializeDb } from "../memory/db-init.js";
 import { resetExternalAssistantIdCache } from "../runtime/auth/external-assistant-id.js";
 import { initAuthSigningKey } from "../runtime/auth/token-service.js";
@@ -37,6 +37,7 @@ import {
   resolveLocalAuthContext,
   resolveLocalTrustContext,
 } from "../runtime/local-actor-identity.js";
+import { resetDbForTesting } from "./db-test-helpers.js";
 
 // ---------------------------------------------------------------------------
 // Test signing key
@@ -50,7 +51,7 @@ initializeDb();
 beforeEach(() => {
   initAuthSigningKey(TEST_KEY);
   resetExternalAssistantIdCache();
-  resetDb();
+  resetDbForTesting();
   initializeDb();
 });
 

package/src/__tests__/agent-loop-exit-reason.test.ts

@@ -21,7 +21,7 @@ import type {
   CheckpointDecision,
   CheckpointInfo,
 } from "../agent/loop.js";
-import { AgentLoop } from "../agent/loop.js";
+import { AgentLoop, isMaxTokensStopReason } from "../agent/loop.js";
 import type {
   Message,
   Provider,
@@ -63,6 +63,18 @@ function textResponse(text: string): ProviderResponse {
   };
 }
 
+function maxTokensResponse(
+  text: string,
+  stopReason: string = "max_tokens",
+): ProviderResponse {
+  return {
+    content: [{ type: "text", text }],
+    model: "mock-model",
+    usage: { inputTokens: 10, outputTokens: 5 },
+    stopReason,
+  };
+}
+
 function toolUseResponse(
   id: string,
   name: string,
@@ -107,6 +119,15 @@ function countExitEvents(events: AgentEvent[]): number {
 // ---------------------------------------------------------------------------
 
 describe("AgentLoop exit-reason instrumentation", () => {
+  test("recognizes provider output-token stop reasons", () => {
+    expect(isMaxTokensStopReason("max_tokens")).toBe(true);
+    expect(isMaxTokensStopReason("MAX_TOKENS")).toBe(true);
+    expect(isMaxTokensStopReason("length")).toBe(true);
+    expect(isMaxTokensStopReason("max_output_tokens")).toBe(true);
+    expect(isMaxTokensStopReason("end_turn")).toBe(false);
+    expect(isMaxTokensStopReason(undefined)).toBe(false);
+  });
+
   test("emits exit event exactly once with 'no_tool_calls' on plain text response", async () => {
     const { provider } = createMockProvider([textResponse("Hi there!")]);
     const loop = new AgentLoop(provider, "system prompt");
@@ -134,6 +155,59 @@ describe("AgentLoop exit-reason instrumentation", () => {
     expect(events[events.length - 1].type).toBe("agent_loop_exit");
   });
 
+  test("emits continuation surface event and exits on max_tokens", async () => {
+    const { provider } = createMockProvider([
+      maxTokensResponse("Partial answer"),
+    ]);
+    const loop = new AgentLoop(provider, "system prompt");
+
+    const events: AgentEvent[] = [];
+    await loop.run([userMessage], (e) => {
+      events.push(e);
+    });
+
+    expect(events.map((e) => e.type)).toEqual([
+      "llm_call_started",
+      "usage",
+      "max_tokens_reached",
+      "message_complete",
+      "agent_loop_exit",
+    ]);
+    expect(countExitEvents(events)).toBe(1);
+    expect(lastExitEvent(events)?.reason).toBe("max_tokens_reached");
+  });
+
+  test("does not persist unexecuted tool_use blocks when max_tokens stops output", async () => {
+    const { provider } = createMockProvider([
+      {
+        content: [
+          { type: "text", text: "I need to check that." },
+          {
+            type: "tool_use",
+            id: "tool-1",
+            name: "read_file",
+            input: { path: "/tmp/example.txt" },
+          },
+        ],
+        model: "mock-model",
+        usage: { inputTokens: 10, outputTokens: 5 },
+        stopReason: "max_tokens",
+      },
+    ]);
+    const loop = new AgentLoop(provider, "system prompt", {}, dummyTools);
+
+    const events: AgentEvent[] = [];
+    const result = await loop.run([userMessage], (e) => {
+      events.push(e);
+    });
+
+    expect(events.some((e) => e.type === "tool_use")).toBe(false);
+    expect(lastExitEvent(events)?.reason).toBe("max_tokens_reached");
+    expect(result[result.length - 1]!.content).toEqual([
+      { type: "text", text: "I need to check that." },
+    ]);
+  });
+
   test("emits 'aborted_pre_call' when signal is already aborted at run start", async () => {
     const { provider } = createMockProvider([textResponse("never sent")]);
     const loop = new AgentLoop(provider, "system prompt");
@@ -142,7 +216,13 @@ describe("AgentLoop exit-reason instrumentation", () => {
     controller.abort();
 
     const events: AgentEvent[] = [];
-    await loop.run([userMessage], (e) => { events.push(e); }, controller.signal);
+    await loop.run(
+      [userMessage],
+      (e) => {
+        events.push(e);
+      },
+      controller.signal,
+    );
 
     expect(countExitEvents(events)).toBe(1);
     expect(lastExitEvent(events)?.reason).toBe("aborted_pre_call");
@@ -166,7 +246,9 @@ describe("AgentLoop exit-reason instrumentation", () => {
     );
 
     const events: AgentEvent[] = [];
-    await loop.run([userMessage], (e) => { events.push(e); });
+    await loop.run([userMessage], (e) => {
+      events.push(e);
+    });
 
     expect(countExitEvents(events)).toBe(1);
     expect(lastExitEvent(events)?.reason).toBe("yield_to_user");
@@ -186,13 +268,14 @@ describe("AgentLoop exit-reason instrumentation", () => {
       toolExecutor,
     );
 
-    const onCheckpoint = (_info: CheckpointInfo): CheckpointDecision =>
-      "yield";
+    const onCheckpoint = (_info: CheckpointInfo): CheckpointDecision => "yield";
 
     const events: AgentEvent[] = [];
     await loop.run(
       [userMessage],
-      (e) => { events.push(e); },
+      (e) => {
+        events.push(e);
+      },
       undefined,
       undefined,
       onCheckpoint,
@@ -211,7 +294,9 @@ describe("AgentLoop exit-reason instrumentation", () => {
     const loop = new AgentLoop(provider, "system prompt");
 
     const events: AgentEvent[] = [];
-    await loop.run([userMessage], (e) => { events.push(e); });
+    await loop.run([userMessage], (e) => {
+      events.push(e);
+    });
 
     expect(countExitEvents(events)).toBe(1);
     expect(lastExitEvent(events)?.reason).toBe("error");
@@ -238,7 +323,9 @@ describe("AgentLoop exit-reason instrumentation", () => {
     );
 
     const events: AgentEvent[] = [];
-    await loop.run([userMessage], (e) => { events.push(e); });
+    await loop.run([userMessage], (e) => {
+      events.push(e);
+    });
 
     expect(countExitEvents(events)).toBe(1);
   });
@@ -263,7 +350,13 @@ describe("AgentLoop exit-reason instrumentation", () => {
     );
 
     const events: AgentEvent[] = [];
-    await loop.run([userMessage], (e) => { events.push(e); }, controller.signal);
+    await loop.run(
+      [userMessage],
+      (e) => {
+        events.push(e);
+      },
+      controller.signal,
+    );
 
     expect(countExitEvents(events)).toBe(1);
     expect(lastExitEvent(events)?.reason).toBe("aborted_during_tools");

package/src/__tests__/agent-loop-override-profile.test.ts

@@ -223,7 +223,7 @@ class FakeConversation {
     return Promise.resolve();
   }
   persistUserMessage() {
-    return Promise.resolve("msg-id");
+    return Promise.resolve({ id: "msg-id", deduplicated: false });
   }
   runAgentLoop(
     _content: string,
@@ -353,6 +353,7 @@ mock.module("../memory/conversation-crud.js", () => ({
   // Always return undefined for the row read so the test fails fast unless
   // executeSubagentSpawn reads from context.overrideProfile first.
   getConversationOverrideProfile: () => undefined,
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 import { getSubagentManager } from "../subagent/index.js";

package/src/__tests__/agent-wake-disk-pressure-callsite.test.ts

@@ -22,6 +22,7 @@ import type { Message } from "../providers/types.js";
 
 mock.module("../memory/conversation-crud.js", () => ({
   getConversationOverrideProfile: () => undefined,
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 mock.module("../config/loader.js", () => ({

package/src/__tests__/agent-wake-override-profile.test.ts

@@ -32,6 +32,7 @@ let mockOverrideProfile: string | undefined = undefined;
 
 mock.module("../memory/conversation-crud.js", () => ({
   getConversationOverrideProfile: (_id: string) => mockOverrideProfile,
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 // Mutable stub for `getConfig().llm` consumed by `RetryProvider`'s

package/src/__tests__/always-loaded-tools-guard.test.ts

@@ -14,12 +14,12 @@
 import { afterAll, describe, expect, test } from "bun:test";
 
 import {
-  buildToolDefinitions,
   isToolActiveForContext,
   type SkillProjectionContext,
 } from "../daemon/conversation-tool-setup.js";
 import {
   __resetRegistryForTesting,
+  getAllToolDefinitions,
   initializeTools,
 } from "../tools/registry.js";
 
@@ -30,7 +30,7 @@ afterAll(() => {
 describe("always-loaded tool count", () => {
   test("should be exactly 11 with recall occupying the existing slot", async () => {
     await initializeTools();
-    const allDefs = buildToolDefinitions();
+    const allDefs = getAllToolDefinitions();
 
     // Minimal context: no client, no capabilities
     const minimalContext: SkillProjectionContext = {

package/src/__tests__/annotate-risk-options.test.ts

@@ -57,6 +57,7 @@ mock.module("../memory/conversation-crud.js", () => ({
     updates.push({ id, content });
   },
   provenanceFromTrustContext: () => ({}),
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 mock.module("../memory/llm-request-log-store.js", () => ({

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

@@ -151,6 +151,7 @@ mock.module("../memory/conversation-crud.js", () => ({
   addMessage: () => ({ id: `msg-${Date.now()}` }),
   updateConversationUsage: () => {},
   updateConversationTitle: () => {},
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 mock.module("../memory/conversation-queries.js", () => ({

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

@@ -154,13 +154,9 @@ function makeIdleSession(opts?: {
   let processing = false;
   return {
     isProcessing: () => processing,
-    persistUserMessage: (
-      _content: string,
-      _attachments: unknown[],
-      requestId?: string,
-    ) => {
+    persistUserMessage: (options: { requestId?: string }) => {
       processing = true;
-      return requestId ?? "msg-1";
+      return { id: options.requestId ?? "msg-1", deduplicated: false };
     },
     memoryPolicy: {
       scopeId: "default",
@@ -223,13 +219,9 @@ function makeConfirmationEmittingSession(opts?: {
   const tool = opts?.toolName ?? "shell_command";
   return {
     isProcessing: () => processing,
-    persistUserMessage: (
-      _content: string,
-      _attachments: unknown[],
-      requestId?: string,
-    ) => {
+    persistUserMessage: (options: { requestId?: string }) => {
       processing = true;
-      return requestId ?? "msg-1";
+      return { id: options.requestId ?? "msg-1", deduplicated: false };
     },
     memoryPolicy: {
       scopeId: "default",
@@ -266,7 +258,11 @@ function makeConfirmationEmittingSession(opts?: {
           input: { command: "ls" },
           riskLevel: "medium",
           allowlistOptions: [
-            { label: "Allow ls", description: "Allow ls command", pattern: "ls" },
+            {
+              label: "Allow ls",
+              description: "Allow ls command",
+              pattern: "ls",
+            },
           ],
           scopeOptions: [{ label: "This conversation", scope: "session" }],
           persistentDecisionsAllowed: true,

package/src/__tests__/assert-not-live-db.ts

@@ -0,0 +1,79 @@
+/**
+ * Test-only utilities for safely interacting with on-disk DB files.
+ *
+ * Two exports:
+ *
+ * - `assertNotLiveDb(dbPath)` — per-callsite guard. Throws unless the
+ *   path resolves under `os.tmpdir()`. Use it before any custom
+ *   destructive operation not covered by `removeTestDbFiles`.
+ * - `removeTestDbFiles(dbPath)` — guarded `rmSync` of the DB file and
+ *   its sibling `-shm` / `-wal` files. The standard way to clear a
+ *   test DB between migrations.
+ *
+ * Defence-in-depth complement to the test-preload-verifier
+ * (`os.tmpdir()` containment check during preload). If a future bug
+ * lets a test see a non-tmpdir path despite the preload guard, every
+ * destructive call still has its own line of defence.
+ *
+ * Positive assertion: the path MUST resolve under `os.tmpdir()`. Same
+ * shape as the preload verifier — no enumeration of "real" paths,
+ * works across every deployment.
+ *
+ * No source-module imports — only node stdlib. Helpers in `__tests__/`
+ * must not import from `src/` (see assistant/AGENTS.md). Callers pass
+ * the resolved `dbPath` in so this file stays decoupled from
+ * `src/util/platform.ts`.
+ */
+
+import { realpathSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { dirname, sep } from "node:path";
+
+function canonicalize(p: string): string {
+  try {
+    return realpathSync(p);
+  } catch {
+    // realpathSync throws if the path doesn't exist. Fall back to the
+    // parent directory — which must exist for any rmSync target that's
+    // about to be touched. Returning the parent path here lets the
+    // containment check succeed in the common "file already deleted"
+    // case (rmSync with `force: true`) without falsely failing.
+    try {
+      return realpathSync(dirname(p));
+    } catch {
+      return p;
+    }
+  }
+}
+
+export function assertNotLiveDb(dbPath: string): void {
+  const resolved = canonicalize(dbPath);
+  const tmpRoot = canonicalize(tmpdir());
+  if (resolved !== tmpRoot && !resolved.startsWith(tmpRoot + sep)) {
+    throw new Error(
+      [
+        `assertNotLiveDb refused: ${dbPath} resolves to ${resolved}, which is not under ${tmpRoot}.`,
+        "",
+        "Test code must never destructively touch a DB path outside the",
+        "process temp directory. If you're seeing this in a passing test,",
+        "the test-preload likely didn't run (the verifier should have",
+        "caught that already — investigate why it didn't).",
+      ].join("\n"),
+    );
+  }
+}
+
+/**
+ * Remove a SQLite DB file and its sibling `-shm` / `-wal` files. Always
+ * `force: true` (no error if the file doesn't exist yet). Guarded by
+ * `assertNotLiveDb` so a test can never destroy a non-tmpdir path.
+ *
+ * Caller passes `dbPath` (typically `getDbPath()` from the test file)
+ * to keep this helper free of source-module imports.
+ */
+export function removeTestDbFiles(dbPath: string): void {
+  assertNotLiveDb(dbPath);
+  rmSync(dbPath, { force: true });
+  rmSync(`${dbPath}-shm`, { force: true });
+  rmSync(`${dbPath}-wal`, { force: true });
+}

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

@@ -14,10 +14,11 @@ import { afterEach, beforeEach, describe, expect, test } from "bun:test";
 
 const DECLARED_FLAG_ID = "email-channel";
 const DECLARED_FLAG_KEY = DECLARED_FLAG_ID;
-const SAFE_STORAGE_LIMITS_FLAG = "safe-storage-limits";
 
-const { isAssistantFeatureFlagEnabled, _setOverridesForTesting } =
+const { isAssistantFeatureFlagEnabled } =
   await import("../config/assistant-feature-flags.js");
+const { setOverridesForTesting } =
+  await import("./feature-flag-test-helpers.js");
 const { skillFlagKey } = await import("../config/skill-state.js");
 
 // ---------------------------------------------------------------------------
@@ -25,11 +26,11 @@ const { skillFlagKey } = await import("../config/skill-state.js");
 // ---------------------------------------------------------------------------
 
 beforeEach(() => {
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 afterEach(() => {
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 // ---------------------------------------------------------------------------
@@ -38,14 +39,14 @@ afterEach(() => {
 
 describe("isAssistantFeatureFlagEnabled", () => {
   test("reads from file-based overrides", () => {
-    _setOverridesForTesting({ [DECLARED_FLAG_KEY]: true });
+    setOverridesForTesting({ [DECLARED_FLAG_KEY]: true });
     const config = {} as any;
 
     expect(isAssistantFeatureFlagEnabled(DECLARED_FLAG_KEY, config)).toBe(true);
   });
 
   test("explicit false override in file-based overrides", () => {
-    _setOverridesForTesting({ [DECLARED_FLAG_KEY]: false });
+    setOverridesForTesting({ [DECLARED_FLAG_KEY]: false });
     const config = {} as any;
 
     expect(isAssistantFeatureFlagEnabled(DECLARED_FLAG_KEY, config)).toBe(
@@ -63,23 +64,6 @@ describe("isAssistantFeatureFlagEnabled", () => {
     );
   });
 
-  test("safe-storage-limits defaults to disabled", () => {
-    const config = {} as any;
-
-    expect(
-      isAssistantFeatureFlagEnabled(SAFE_STORAGE_LIMITS_FLAG, config),
-    ).toBe(false);
-  });
-
-  test("safe-storage-limits respects explicit override", () => {
-    _setOverridesForTesting({ [SAFE_STORAGE_LIMITS_FLAG]: true });
-    const config = {} as any;
-
-    expect(
-      isAssistantFeatureFlagEnabled(SAFE_STORAGE_LIMITS_FLAG, config),
-    ).toBe(true);
-  });
-
   test("unknown flag defaults to false when no persisted override", () => {
     const config = {} as any;
 
@@ -87,7 +71,7 @@ describe("isAssistantFeatureFlagEnabled", () => {
   });
 
   test("undeclared flag respects persisted override", () => {
-    _setOverridesForTesting({ "some-undeclared-flag": false });
+    setOverridesForTesting({ "some-undeclared-flag": false });
     const config = {} as any;
 
     expect(isAssistantFeatureFlagEnabled("some-undeclared-flag", config)).toBe(
@@ -98,7 +82,7 @@ describe("isAssistantFeatureFlagEnabled", () => {
 
 describe("isAssistantFeatureFlagEnabled with skillFlagKey", () => {
   test("resolves skill flag via canonical path", () => {
-    _setOverridesForTesting({ [DECLARED_FLAG_KEY]: false });
+    setOverridesForTesting({ [DECLARED_FLAG_KEY]: false });
     const config = {} as any;
 
     expect(

package/src/__tests__/audit-log-rotation.test.ts

@@ -12,13 +12,13 @@ mock.module("../util/logger.js", () => ({
     }),
 }));
 
-import { resetDb } from "../memory/db-connection.js";
 import { getSqlite } from "../memory/db-connection.js";
 import { initializeDb } from "../memory/db-init.js";
 import {
   getRecentInvocations,
   rotateToolInvocations,
 } from "../memory/tool-usage-store.js";
+import { resetDbForTesting } from "./db-test-helpers.js";
 
 // ---------------------------------------------------------------------------
 // Helpers
@@ -57,7 +57,7 @@ const ONE_DAY_MS = 24 * 60 * 60 * 1000;
 
 describe("audit log rotation", () => {
   beforeAll(() => {
-    resetDb();
+    resetDbForTesting();
     initializeDb();
     // Insert a conversations row so FK-enforced ORM inserts succeed
     getSqlite().run(

package/src/__tests__/auto-analysis-end-to-end.test.ts

@@ -130,7 +130,6 @@ mock.module("../runtime/services/analyze-conversation.js", () => ({
 
 // ── Real imports ──────────────────────────────────────────────────
 
-import { _setOverridesForTesting } from "../config/assistant-feature-flags.js";
 import { conversationAnalyzeJob } from "../memory/conversation-analyze-job.js";
 import { createConversation } from "../memory/conversation-crud.js";
 import { getDb } from "../memory/db-connection.js";
@@ -138,6 +137,7 @@ import { initializeDb } from "../memory/db-init.js";
 import { indexMessageNow } from "../memory/indexer.js";
 import type { MemoryJob } from "../memory/jobs-store.js";
 import { conversations, memoryJobs, messages } from "../memory/schema.js";
+import { setOverridesForTesting } from "./feature-flag-test-helpers.js";
 
 // ── Helpers ───────────────────────────────────────────────────────
 
@@ -279,14 +279,14 @@ beforeEach(() => {
   resetTables();
   analyzeCalls.length = 0;
   // Clear any stale feature-flag overrides between tests.
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 // ─────────────────────────────────────────────────────────────────
 
 describe("auto-analysis end-to-end trigger path", () => {
   test("flag on, batch threshold reached → conversation_analyze enqueued, handler runs, analysis conversation created", async () => {
-    _setOverridesForTesting({ "auto-analyze": true });
+    setOverridesForTesting({ "auto-analyze": true });
 
     const source = createConversation("source-conv");
 
@@ -329,7 +329,7 @@ describe("auto-analysis end-to-end trigger path", () => {
   });
 
   test("flag off → no conversation_analyze job is ever enqueued", async () => {
-    _setOverridesForTesting({ "auto-analyze": false });
+    setOverridesForTesting({ "auto-analyze": false });
 
     const source = createConversation("source-conv-flag-off");
     await indexMessages(source.id, 3);
@@ -343,7 +343,7 @@ describe("auto-analysis end-to-end trigger path", () => {
   });
 
   test("recursion guard: indexing into an auto-analysis conversation does not enqueue conversation_analyze or graph_extract", async () => {
-    _setOverridesForTesting({ "auto-analyze": true });
+    setOverridesForTesting({ "auto-analyze": true });
 
     // Set up an auto-analysis conversation directly. This is the
     // scenario we need to protect against — the analysis agent writes
@@ -392,7 +392,7 @@ describe("auto-analysis batch trigger uses analysis.batchSize cadence", () => {
   const originalV2Enabled = TEST_CONFIG.memory.v2.enabled;
 
   beforeEach(() => {
-    _setOverridesForTesting({ "auto-analyze": true });
+    setOverridesForTesting({ "auto-analyze": true });
     // memory.v2.enabled gates v1 graph_extract enqueue; force off so
     // these cadence tests can observe the v1 path.
     TEST_CONFIG.memory.v2.enabled = false;

package/src/__tests__/background-workers-disk-pressure.test.ts

@@ -1,5 +1,7 @@
 import { beforeEach, describe, expect, mock, test } from "bun:test";
 
+import { createMockLoggerModule } from "./helpers/mock-logger.js";
+
 // Default the warm-pool gate to OPEN — these tests probe background-job
 // disk-pressure behavior, not the pre-first-message guard.
 mock.module("../runtime/pre-first-message-gate.js", () => ({
@@ -7,14 +9,7 @@ mock.module("../runtime/pre-first-message-gate.js", () => ({
   _resetPreFirstMessageGateCacheForTests: () => {},
 }));
 
-mock.module("../util/logger.js", () => ({
-  getLogger: () => ({
-    info: () => {},
-    debug: () => {},
-    warn: () => {},
-    error: () => {},
-  }),
-}));
+mock.module("../util/logger.js", () => createMockLoggerModule());
 
 mock.module("../config/loader.js", () => ({
   getConfig: () => ({
@@ -157,6 +152,8 @@ mock.module("../memory/conversation-crud.js", () => ({
   setLastNotifiedInferenceProfile: mock(() => {}),
   setConversationHistoryStrippedAt: mock(() => {}),
   wipeConversation: mock(() => ({ memoryIds: [] })),
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
+  extractImageSourcePaths: () => undefined,
 }));
 
 mock.module("../memory/conversation-title-service.js", () => ({

package/src/__tests__/browser-skill-endstate.test.ts

@@ -12,7 +12,6 @@ mock.module("../config/loader.js", () => ({
 
 import { BROWSER_OPERATION_META } from "../browser/operations.js";
 import { BROWSER_OPERATIONS } from "../browser/types.js";
-import { _setOverridesForTesting } from "../config/assistant-feature-flags.js";
 import {
   projectSkillTools,
   resetSkillToolProjection,
@@ -23,6 +22,7 @@ import {
   getAllTools,
   initializeTools,
 } from "../tools/registry.js";
+import { setOverridesForTesting } from "./feature-flag-test-helpers.js";
 import {
   BROWSER_SKILL_ID,
   buildSkillLoadHistory,
@@ -30,13 +30,13 @@ import {
 
 afterAll(() => {
   __resetRegistryForTesting();
-  _setOverridesForTesting({});
+  setOverridesForTesting({});
 });
 
 describe("browser CLI-only architecture end-state", () => {
   beforeAll(async () => {
     __resetRegistryForTesting();
-    _setOverridesForTesting({
+    setOverridesForTesting({
       browser: true,
     });
     await initializeTools();

package/src/__tests__/btw-routes.test.ts

@@ -38,6 +38,7 @@ const mockAddMessage = mock(() => {});
 
 mock.module("../memory/conversation-crud.js", () => ({
   addMessage: mockAddMessage,
+  reserveMessage: mock(async () => ({ id: "msg-reserve" })),
 }));
 
 const MOCK_TOOLS = [
@@ -48,8 +49,8 @@ const MOCK_TOOLS = [
   },
 ];
 
-mock.module("../daemon/conversation-tool-setup.js", () => ({
-  buildToolDefinitions: () => MOCK_TOOLS,
+mock.module("../tools/registry.js", () => ({
+  getAllToolDefinitions: () => MOCK_TOOLS,
 }));
 
 const MOCK_SYSTEM_PROMPT = "You are a helpful assistant.";