fix(core): update tests to match tool changes and resolve timeouts

- ensure sub-agent tools and ActivateSkillTool respect the coreTools configuration
- prevents unauthorized tool calls when using restricted toolsets, particularly with Gemini 3
- added missing web_fetch tool to CodebaseInvestigatorAgent toolConfig to match its system prompt

These changes resolve several behavioral evaluation failures (e.g., save_memory) that occurred when the agent attempted to use incorrectly registered sub-agents or skills in restricted environments.

packages/core/src/agents/codebase-investigator.test.ts

@@ -11,6 +11,7 @@ import {
   GREP_TOOL_NAME,
   LS_TOOL_NAME,
   READ_FILE_TOOL_NAME,
+  WEB_FETCH_TOOL_NAME,
 } from '../tools/tool-names.js';
 import { DEFAULT_GEMINI_MODEL } from '../config/models.js';
 import { makeFakeConfig } from '../test-utils/config.js';
@@ -50,6 +51,7 @@ describe('CodebaseInvestigatorAgent', () => {
       READ_FILE_TOOL_NAME,
       GLOB_TOOL_NAME,
       GREP_TOOL_NAME,
+      WEB_FETCH_TOOL_NAME,
     ]);
   });
 

packages/core/src/agents/codebase-investigator.ts

@@ -10,6 +10,7 @@ import {
   GREP_TOOL_NAME,
   LS_TOOL_NAME,
   READ_FILE_TOOL_NAME,
+  WEB_FETCH_TOOL_NAME,
 } from '../tools/tool-names.js';
 import {
   DEFAULT_THINKING_MODE,
@@ -120,6 +121,7 @@ export const CodebaseInvestigatorAgent = (
         READ_FILE_TOOL_NAME,
         GLOB_TOOL_NAME,
         GREP_TOOL_NAME,
+        WEB_FETCH_TOOL_NAME,
       ],
     },
 

packages/core/src/config/config.ts

@@ -947,7 +947,10 @@ export class Config {
         this.getSkillManager().setDisabledSkills(this.disabledSkills);
 
         // Re-register ActivateSkillTool to update its schema with the discovered enabled skill enums
-        if (this.getSkillManager().getSkills().length > 0) {
+        if (
+          this.getSkillManager().getSkills().length > 0 &&
+          this.isToolEnabledByCore(ActivateSkillTool.Name)
+        ) {
           this.getToolRegistry().unregisterTool(ActivateSkillTool.Name);
           this.getToolRegistry().registerTool(
             new ActivateSkillTool(this, this.messageBus),
@@ -971,6 +974,21 @@ export class Config {
     this.syncPlanModeTools();
   }
 
+  /**
+   * Returns true if the specified tool name is enabled based on the current
+   * coreTools configuration.
+   */
+  private isToolEnabledByCore(toolName: string): boolean {
+    const coreTools = this.getCoreTools();
+    if (!coreTools) {
+      return true;
+    }
+
+    return coreTools.some(
+      (name) => name === toolName || name === toolName.toLowerCase(),
+    );
+  }
+
   getContentGenerator(): ContentGenerator {
     return this.contentGenerator;
   }
@@ -2002,7 +2020,10 @@ export class Config {
       this.getSkillManager().setDisabledSkills(this.disabledSkills);
 
       // Re-register ActivateSkillTool to update its schema with the newly discovered skills
-      if (this.getSkillManager().getSkills().length > 0) {
+      if (
+        this.getSkillManager().getSkills().length > 0 &&
+        this.isToolEnabledByCore(ActivateSkillTool.Name)
+      ) {
         this.getToolRegistry().unregisterTool(ActivateSkillTool.Name);
         this.getToolRegistry().registerTool(
           new ActivateSkillTool(this, this.messageBus),
@@ -2234,8 +2255,10 @@ export class Config {
       const definitions = this.agentRegistry.getAllDefinitions();
 
       for (const definition of definitions) {
-        const isAllowed =
+        const isAllowedByTools =
           !allowedTools || allowedTools.includes(definition.name);
+        const isAllowed =
+          isAllowedByTools && this.isToolEnabledByCore(definition.name);
 
         if (isAllowed) {
           try {

packages/core/src/policy/config.test.ts

@@ -12,6 +12,10 @@ import type { PolicySettings } from './types.js';
 import { ApprovalMode, PolicyDecision, InProcessCheckerType } from './types.js';
 import { isDirectorySecure } from '../utils/security.js';
 
+import { Storage } from '../config/storage.js';
+import * as tomlLoader from './toml-loader.js';
+import { createPolicyEngineConfig } from './config.js';
+
 vi.unmock('../config/storage.js');
 
 vi.mock('../utils/security.js', () => ({
@@ -26,8 +30,6 @@ afterEach(() => {
 
 describe('createPolicyEngineConfig', () => {
   beforeEach(async () => {
-    vi.resetModules();
-    const { Storage } = await import('../config/storage.js');
     // Mock Storage to avoid picking up real user/system policies from the host environment
     vi.spyOn(Storage, 'getUserPoliciesDir').mockReturnValue(
       '/non/existent/user/policies',
@@ -40,7 +42,6 @@ describe('createPolicyEngineConfig', () => {
   });
 
   it('should filter out insecure system policy directories', async () => {
-    const { Storage } = await import('../config/storage.js');
     const systemPolicyDir = '/insecure/system/policies';
     vi.spyOn(Storage, 'getSystemPoliciesDir').mockReturnValue(systemPolicyDir);
 
@@ -51,10 +52,6 @@ describe('createPolicyEngineConfig', () => {
       return { secure: true };
     });
 
-    // We need to spy on loadPoliciesFromToml to verify which directories were passed
-    // But it is not exported from config.js, it is imported.
-    // We can spy on the module it comes from.
-    const tomlLoader = await import('./toml-loader.js');
     const loadPoliciesSpy = vi.spyOn(tomlLoader, 'loadPoliciesFromToml');
     loadPoliciesSpy.mockResolvedValue({
       rules: [],
@@ -62,7 +59,6 @@ describe('createPolicyEngineConfig', () => {
       errors: [],
     });
 
-    const { createPolicyEngineConfig } = await import('./config.js');
     const settings: PolicySettings = {};
 
     await createPolicyEngineConfig(
@@ -80,7 +76,7 @@ describe('createPolicyEngineConfig', () => {
     // But other directories (user, default) should be there
     expect(calledDirs).toContain('/non/existent/user/policies');
     expect(calledDirs).toContain('/tmp/mock/default/policies');
-  });
+  }, 30000);
 
   it('should return ASK_USER for write tools and ALLOW for read-only tools by default', async () => {
     const actualFs =