Merge branch 'main' into test-utils-model-fallback

Support model fallbacks in integration tests by allowing a comma-separated list of models via the `GEMINI_MODEL` environment variable or the `model` option in `TestRig.setup()`. This will automatically configure a fallback chain using `dynamicModelConfiguration` with `silent` actions.

Fixes #23568

docs/cli/plan-mode.md

@@ -39,7 +39,9 @@ To start Plan Mode while using Gemini CLI:
   the rotation when Gemini CLI is actively processing or showing confirmation
   dialogs.
 
-- **Command:** Type `/plan` in the input box.
+- **Command:** Type `/plan [goal]` in the input box. The `[goal]` is optional;
+  for example, `/plan implement authentication` will switch to Plan Mode and
+  immediately submit the prompt to the model.
 
 - **Natural Language:** Ask Gemini CLI to "start a plan for...". Gemini CLI
   calls the

evals/subagents.eval.ts

@@ -13,8 +13,21 @@ import { evalTest, TEST_AGENTS } from './test-helper.js';
 
 const INDEX_TS = 'export const add = (a: number, b: number) => a + b;\n';
 
+// A minimal package.json is used to provide a realistic workspace anchor.
+// This prevents the agent from making incorrect assumptions about the environment
+// and helps it properly navigate or act as if it is in a standard Node.js project.
+const MOCK_PACKAGE_JSON = JSON.stringify(
+  {
+    name: 'subagent-eval-project',
+    version: '1.0.0',
+    type: 'module',
+  },
+  null,
+  2,
+);
+
 function readProjectFile(
-  rig: { testDir?: string },
+  rig: { testDir: string | null },
   relativePath: string,
 ): string {
   return fs.readFileSync(path.join(rig.testDir!, relativePath), 'utf8');
@@ -117,15 +130,7 @@ describe('subagent eval test cases', () => {
     files: {
       ...TEST_AGENTS.TESTING_AGENT.asFile(),
       'index.ts': INDEX_TS,
-      'package.json': JSON.stringify(
-        {
-          name: 'subagent-eval-project',
-          version: '1.0.0',
-          type: 'module',
-        },
-        null,
-        2,
-      ),
+      'package.json': MOCK_PACKAGE_JSON,
     },
     assert: async (rig, _result) => {
       const toolLogs = rig.readToolLogs() as Array<{
@@ -164,15 +169,7 @@ describe('subagent eval test cases', () => {
       ...TEST_AGENTS.TESTING_AGENT.asFile(),
       'index.ts': INDEX_TS,
       'README.md': 'TODO: update the README.\n',
-      'package.json': JSON.stringify(
-        {
-          name: 'subagent-eval-project',
-          version: '1.0.0',
-          type: 'module',
-        },
-        null,
-        2,
-      ),
+      'package.json': MOCK_PACKAGE_JSON,
     },
     assert: async (rig, _result) => {
       const toolLogs = rig.readToolLogs() as Array<{
@@ -190,4 +187,105 @@ describe('subagent eval test cases', () => {
       );
     },
   });
+
+  /**
+   * Checks that the main agent can correctly select the appropriate subagent
+   * from a large pool of available subagents (10 total).
+   */
+  evalTest('USUALLY_PASSES', {
+    name: 'should select the correct subagent from a pool of 10 different agents',
+    prompt: 'Please add a new SQL table migration for a user profile.',
+    files: {
+      ...TEST_AGENTS.DOCS_AGENT.asFile(),
+      ...TEST_AGENTS.TESTING_AGENT.asFile(),
+      ...TEST_AGENTS.DATABASE_AGENT.asFile(),
+      ...TEST_AGENTS.CSS_AGENT.asFile(),
+      ...TEST_AGENTS.I18N_AGENT.asFile(),
+      ...TEST_AGENTS.SECURITY_AGENT.asFile(),
+      ...TEST_AGENTS.DEVOPS_AGENT.asFile(),
+      ...TEST_AGENTS.ANALYTICS_AGENT.asFile(),
+      ...TEST_AGENTS.ACCESSIBILITY_AGENT.asFile(),
+      ...TEST_AGENTS.MOBILE_AGENT.asFile(),
+      'package.json': MOCK_PACKAGE_JSON,
+    },
+    assert: async (rig, _result) => {
+      const toolLogs = rig.readToolLogs() as Array<{
+        toolRequest: { name: string };
+      }>;
+      await rig.expectToolCallSuccess(['database-agent']);
+
+      // Ensure the generalist and other irrelevant specialists were not invoked
+      const uncalledAgents = [
+        'generalist',
+        TEST_AGENTS.DOCS_AGENT.name,
+        TEST_AGENTS.TESTING_AGENT.name,
+        TEST_AGENTS.CSS_AGENT.name,
+        TEST_AGENTS.I18N_AGENT.name,
+        TEST_AGENTS.SECURITY_AGENT.name,
+        TEST_AGENTS.DEVOPS_AGENT.name,
+        TEST_AGENTS.ANALYTICS_AGENT.name,
+        TEST_AGENTS.ACCESSIBILITY_AGENT.name,
+        TEST_AGENTS.MOBILE_AGENT.name,
+      ];
+
+      for (const agentName of uncalledAgents) {
+        expect(toolLogs.some((l) => l.toolRequest.name === agentName)).toBe(
+          false,
+        );
+      }
+    },
+  });
+
+  /**
+   * Checks that the main agent can correctly select the appropriate subagent
+   * from a large pool of available subagents, even when many irrelevant MCP tools are present.
+   *
+   * This test includes stress tests the subagent delegation with ~80 tools.
+   */
+  evalTest('USUALLY_PASSES', {
+    name: 'should select the correct subagent from a pool of 10 different agents with MCP tools present',
+    prompt: 'Please add a new SQL table migration for a user profile.',
+    setup: async (rig) => {
+      rig.addTestMcpServer('workspace-server', 'google-workspace');
+    },
+    files: {
+      ...TEST_AGENTS.DOCS_AGENT.asFile(),
+      ...TEST_AGENTS.TESTING_AGENT.asFile(),
+      ...TEST_AGENTS.DATABASE_AGENT.asFile(),
+      ...TEST_AGENTS.CSS_AGENT.asFile(),
+      ...TEST_AGENTS.I18N_AGENT.asFile(),
+      ...TEST_AGENTS.SECURITY_AGENT.asFile(),
+      ...TEST_AGENTS.DEVOPS_AGENT.asFile(),
+      ...TEST_AGENTS.ANALYTICS_AGENT.asFile(),
+      ...TEST_AGENTS.ACCESSIBILITY_AGENT.asFile(),
+      ...TEST_AGENTS.MOBILE_AGENT.asFile(),
+      'package.json': MOCK_PACKAGE_JSON,
+    },
+    assert: async (rig, _result) => {
+      const toolLogs = rig.readToolLogs() as Array<{
+        toolRequest: { name: string };
+      }>;
+      await rig.expectToolCallSuccess(['database-agent']);
+
+      // Ensure the generalist and other irrelevant specialists were not invoked
+      const uncalledAgents = [
+        'generalist',
+        TEST_AGENTS.DOCS_AGENT.name,
+        TEST_AGENTS.TESTING_AGENT.name,
+        TEST_AGENTS.CSS_AGENT.name,
+        TEST_AGENTS.I18N_AGENT.name,
+        TEST_AGENTS.SECURITY_AGENT.name,
+        TEST_AGENTS.DEVOPS_AGENT.name,
+        TEST_AGENTS.ANALYTICS_AGENT.name,
+        TEST_AGENTS.ACCESSIBILITY_AGENT.name,
+        TEST_AGENTS.MOBILE_AGENT.name,
+      ];
+
+      for (const agentName of uncalledAgents) {
+        expect(toolLogs.some((l) => l.toolRequest.name === agentName)).toBe(
+          false,
+        );
+      }
+    },
+  });
 });

evals/test-helper.ts

@@ -61,6 +61,10 @@ export async function internalEvalTest(evalCase: EvalCase) {
     try {
       rig.setup(evalCase.name, evalCase.params);
 
+      if (evalCase.setup) {
+        await evalCase.setup(rig);
+      }
+
       if (evalCase.files) {
         await setupTestFiles(rig, evalCase.files);
       }
@@ -371,6 +375,7 @@ export interface EvalCase {
   prompt: string;
   timeout?: number;
   files?: Record<string, string>;
+  setup?: (rig: TestRig) => Promise<void> | void;
   /** Conversation history to pre-load via --resume. Each entry is a message object with type, content, etc. */
   messages?: Record<string, unknown>[];
   /** Session ID for the resumed session. Auto-generated if not provided. */

packages/a2a-server/src/config/config.test.ts

@@ -424,7 +424,22 @@ describe('loadConfig', () => {
       });
     });
 
-    describe('authentication fallback', () => {
+    describe('authentication logic', () => {
+      const setupConfigMock = (refreshAuthMock: ReturnType<typeof vi.fn>) => {
+        vi.mocked(Config).mockImplementation(
+          (params: unknown) =>
+            ({
+              ...(params as object),
+              initialize: vi.fn(),
+              waitForMcpInit: vi.fn(),
+              refreshAuth: refreshAuthMock,
+              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
+              getRemoteAdminSettings: vi.fn(),
+              setRemoteAdminSettings: vi.fn(),
+            }) as unknown as Config,
+        );
+      };
+
       beforeEach(() => {
         vi.stubEnv('USE_CCPA', 'true');
         vi.stubEnv('GEMINI_API_KEY', '');
@@ -434,182 +449,77 @@ describe('loadConfig', () => {
         vi.unstubAllEnvs();
       });
 
-      it('should fall back to COMPUTE_ADC in Cloud Shell if LOGIN_WITH_GOOGLE fails', async () => {
-        vi.stubEnv('CLOUD_SHELL', 'true');
-        vi.mocked(isHeadlessMode).mockReturnValue(false);
-        const refreshAuthMock = vi.fn().mockImplementation((authType) => {
-          if (authType === AuthType.LOGIN_WITH_GOOGLE) {
-            throw new FatalAuthenticationError('Non-interactive session');
-          }
-          return Promise.resolve();
-        });
-
-        // Update the mock implementation for this test
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
-
-        await loadConfig(mockSettings, mockExtensionLoader, taskId);
-
-        expect(refreshAuthMock).toHaveBeenCalledWith(
-          AuthType.LOGIN_WITH_GOOGLE,
-        );
-        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
-      });
-
-      it('should not fall back to COMPUTE_ADC if not in cloud environment', async () => {
-        vi.mocked(isHeadlessMode).mockReturnValue(false);
-        const refreshAuthMock = vi.fn().mockImplementation((authType) => {
-          if (authType === AuthType.LOGIN_WITH_GOOGLE) {
-            throw new FatalAuthenticationError('Non-interactive session');
-          }
-          return Promise.resolve();
-        });
-
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
-
-        await expect(
-          loadConfig(mockSettings, mockExtensionLoader, taskId),
-        ).rejects.toThrow('Non-interactive session');
-
-        expect(refreshAuthMock).toHaveBeenCalledWith(
-          AuthType.LOGIN_WITH_GOOGLE,
-        );
-        expect(refreshAuthMock).not.toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
-      });
-
-      it('should skip LOGIN_WITH_GOOGLE and use COMPUTE_ADC directly in headless Cloud Shell', async () => {
-        vi.stubEnv('CLOUD_SHELL', 'true');
-        vi.mocked(isHeadlessMode).mockReturnValue(true);
-
+      it('should attempt COMPUTE_ADC by default and bypass LOGIN_WITH_GOOGLE if successful', async () => {
         const refreshAuthMock = vi.fn().mockResolvedValue(undefined);
-
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
+        setupConfigMock(refreshAuthMock);
 
         await loadConfig(mockSettings, mockExtensionLoader, taskId);
 
+        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
         expect(refreshAuthMock).not.toHaveBeenCalledWith(
           AuthType.LOGIN_WITH_GOOGLE,
         );
-        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
       });
 
-      it('should skip LOGIN_WITH_GOOGLE and use COMPUTE_ADC directly if GEMINI_CLI_USE_COMPUTE_ADC is true', async () => {
-        vi.stubEnv('GEMINI_CLI_USE_COMPUTE_ADC', 'true');
-        vi.mocked(isHeadlessMode).mockReturnValue(false); // Even if not headless
-
-        const refreshAuthMock = vi.fn().mockResolvedValue(undefined);
-
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
+      it('should fallback to LOGIN_WITH_GOOGLE if COMPUTE_ADC fails and interactive mode is available', async () => {
+        vi.mocked(isHeadlessMode).mockReturnValue(false);
+        const refreshAuthMock = vi.fn().mockImplementation((authType) => {
+          if (authType === AuthType.COMPUTE_ADC) {
+            return Promise.reject(new Error('ADC failed'));
+          }
+          return Promise.resolve();
+        });
+        setupConfigMock(refreshAuthMock);
 
         await loadConfig(mockSettings, mockExtensionLoader, taskId);
 
-        expect(refreshAuthMock).not.toHaveBeenCalledWith(
+        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
+        expect(refreshAuthMock).toHaveBeenCalledWith(
           AuthType.LOGIN_WITH_GOOGLE,
         );
-        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
       });
 
-      it('should throw FatalAuthenticationError in headless mode if no ADC fallback available', async () => {
+      it('should throw FatalAuthenticationError in headless mode if COMPUTE_ADC fails', async () => {
         vi.mocked(isHeadlessMode).mockReturnValue(true);
 
-        const refreshAuthMock = vi.fn().mockResolvedValue(undefined);
-
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
+        const refreshAuthMock = vi.fn().mockImplementation((authType) => {
+          if (authType === AuthType.COMPUTE_ADC) {
+            return Promise.reject(new Error('ADC not found'));
+          }
+          return Promise.resolve();
+        });
+        setupConfigMock(refreshAuthMock);
 
         await expect(
           loadConfig(mockSettings, mockExtensionLoader, taskId),
         ).rejects.toThrow(
-          'Interactive terminal required for LOGIN_WITH_GOOGLE. Run in an interactive terminal or set GEMINI_CLI_USE_COMPUTE_ADC=true to use Application Default Credentials.',
+          'COMPUTE_ADC failed: ADC not found. (LOGIN_WITH_GOOGLE fallback skipped due to headless mode. Run in an interactive terminal to use OAuth.)',
         );
 
-        expect(refreshAuthMock).not.toHaveBeenCalled();
+        expect(refreshAuthMock).toHaveBeenCalledWith(AuthType.COMPUTE_ADC);
+        expect(refreshAuthMock).not.toHaveBeenCalledWith(
+          AuthType.LOGIN_WITH_GOOGLE,
+        );
       });
 
-      it('should include both original and fallback error when COMPUTE_ADC fallback fails', async () => {
-        vi.stubEnv('CLOUD_SHELL', 'true');
+      it('should include both original and fallback error when LOGIN_WITH_GOOGLE fallback fails', async () => {
         vi.mocked(isHeadlessMode).mockReturnValue(false);
 
         const refreshAuthMock = vi.fn().mockImplementation((authType) => {
-          if (authType === AuthType.LOGIN_WITH_GOOGLE) {
-            throw new FatalAuthenticationError('OAuth failed');
-          }
           if (authType === AuthType.COMPUTE_ADC) {
             throw new Error('ADC failed');
           }
+          if (authType === AuthType.LOGIN_WITH_GOOGLE) {
+            throw new FatalAuthenticationError('OAuth failed');
+          }
           return Promise.resolve();
         });
-
-        vi.mocked(Config).mockImplementation(
-          (params: unknown) =>
-            ({
-              ...(params as object),
-              initialize: vi.fn(),
-              waitForMcpInit: vi.fn(),
-              refreshAuth: refreshAuthMock,
-              getExperiments: vi.fn().mockReturnValue({ flags: {} }),
-              getRemoteAdminSettings: vi.fn(),
-              setRemoteAdminSettings: vi.fn(),
-            }) as unknown as Config,
-        );
+        setupConfigMock(refreshAuthMock);
 
         await expect(
           loadConfig(mockSettings, mockExtensionLoader, taskId),
         ).rejects.toThrow(
-          'OAuth failed. Fallback to COMPUTE_ADC also failed: ADC failed',
+          'OAuth failed. The initial COMPUTE_ADC attempt also failed: ADC failed',
         );
       });
     });

packages/a2a-server/src/config/config.ts

@@ -25,7 +25,6 @@ import {
   ExperimentFlags,
   isHeadlessMode,
   FatalAuthenticationError,
-  isCloudShell,
   PolicyDecision,
   PRIORITY_YOLO_ALLOW_ALL,
   type TelemetryTarget,
@@ -43,7 +42,6 @@ export async function loadConfig(
   taskId: string,
 ): Promise<Config> {
   const workspaceDir = process.cwd();
-  const adcFilePath = process.env['GOOGLE_APPLICATION_CREDENTIALS'];
 
   const folderTrust =
     settings.folderTrust === true ||
@@ -192,7 +190,7 @@ export async function loadConfig(
   await config.waitForMcpInit();
   startupProfiler.flush(config);
 
-  await refreshAuthentication(config, adcFilePath, 'Config');
+  await refreshAuthentication(config, 'Config');
 
   return config;
 }
@@ -263,75 +261,51 @@ function findEnvFile(startDir: string): string | null {
 
 async function refreshAuthentication(
   config: Config,
-  adcFilePath: string | undefined,
   logPrefix: string,
 ): Promise<void> {
   if (process.env['USE_CCPA']) {
     logger.info(`[${logPrefix}] Using CCPA Auth:`);
+
+    logger.info(`[${logPrefix}] Attempting COMPUTE_ADC first.`);
     try {
-      if (adcFilePath) {
-        path.resolve(adcFilePath);
-      }
-    } catch (e) {
-      logger.error(
-        `[${logPrefix}] USE_CCPA env var is true but unable to resolve GOOGLE_APPLICATION_CREDENTIALS file path ${adcFilePath}. Error ${e}`,
+      await config.refreshAuth(AuthType.COMPUTE_ADC);
+      logger.info(`[${logPrefix}] COMPUTE_ADC successful.`);
+    } catch (adcError) {
+      const adcMessage =
+        adcError instanceof Error ? adcError.message : String(adcError);
+      logger.info(
+        `[${logPrefix}] COMPUTE_ADC failed or not available: ${adcMessage}`,
       );
-    }
 
-    const useComputeAdc = process.env['GEMINI_CLI_USE_COMPUTE_ADC'] === 'true';
-    const isHeadless = isHeadlessMode();
-    const shouldSkipOauth = isHeadless || useComputeAdc;
+      const useComputeAdc =
+        process.env['GEMINI_CLI_USE_COMPUTE_ADC'] === 'true';
+      const isHeadless = isHeadlessMode();
 
-    if (shouldSkipOauth) {
-      if (isCloudShell() || useComputeAdc) {
-        logger.info(
-          `[${logPrefix}] Skipping LOGIN_WITH_GOOGLE due to ${isHeadless ? 'headless mode' : 'GEMINI_CLI_USE_COMPUTE_ADC'}. Attempting COMPUTE_ADC.`,
-        );
-        try {
-          await config.refreshAuth(AuthType.COMPUTE_ADC);
-          logger.info(`[${logPrefix}] COMPUTE_ADC successful.`);
-        } catch (adcError) {
-          const adcMessage =
-            adcError instanceof Error ? adcError.message : String(adcError);
-          throw new FatalAuthenticationError(
-            `COMPUTE_ADC failed: ${adcMessage}. (Skipped LOGIN_WITH_GOOGLE due to ${isHeadless ? 'headless mode' : 'GEMINI_CLI_USE_COMPUTE_ADC'})`,
-          );
-        }
-      } else {
+      if (isHeadless || useComputeAdc) {
+        const reason = isHeadless
+          ? 'headless mode'
+          : 'GEMINI_CLI_USE_COMPUTE_ADC=true';
         throw new FatalAuthenticationError(
-          `Interactive terminal required for LOGIN_WITH_GOOGLE. Run in an interactive terminal or set GEMINI_CLI_USE_COMPUTE_ADC=true to use Application Default Credentials.`,
+          `COMPUTE_ADC failed: ${adcMessage}. (LOGIN_WITH_GOOGLE fallback skipped due to ${reason}. Run in an interactive terminal to use OAuth.)`,
         );
       }
-    } else {
+
+      logger.info(
+        `[${logPrefix}] COMPUTE_ADC failed, falling back to LOGIN_WITH_GOOGLE.`,
+      );
       try {
         await config.refreshAuth(AuthType.LOGIN_WITH_GOOGLE);
       } catch (e) {
-        if (
-          e instanceof FatalAuthenticationError &&
-          (isCloudShell() || useComputeAdc)
-        ) {
-          logger.warn(
-            `[${logPrefix}] LOGIN_WITH_GOOGLE failed. Attempting COMPUTE_ADC fallback.`,
+        if (e instanceof FatalAuthenticationError) {
+          const originalMessage = e instanceof Error ? e.message : String(e);
+          throw new FatalAuthenticationError(
+            `${originalMessage}. The initial COMPUTE_ADC attempt also failed: ${adcMessage}`,
           );
-          try {
-            await config.refreshAuth(AuthType.COMPUTE_ADC);
-            logger.info(`[${logPrefix}] COMPUTE_ADC fallback successful.`);
-          } catch (adcError) {
-            logger.error(
-              `[${logPrefix}] COMPUTE_ADC fallback failed: ${adcError}`,
-            );
-            const originalMessage = e instanceof Error ? e.message : String(e);
-            const adcMessage =
-              adcError instanceof Error ? adcError.message : String(adcError);
-            throw new FatalAuthenticationError(
-              `${originalMessage}. Fallback to COMPUTE_ADC also failed: ${adcMessage}`,
-            );
-          }
-        } else {
-          throw e;
         }
+        throw e;
       }
     }
+
     logger.info(
       `[${logPrefix}] GOOGLE_CLOUD_PROJECT: ${process.env['GOOGLE_CLOUD_PROJECT']}`,
     );

packages/cli/src/ui/commands/chatCommand.ts

@@ -75,6 +75,7 @@ const listCommand: SlashCommand = {
   description: 'List saved manual conversation checkpoints',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
+  takesArgs: false,
   action: async (context): Promise<void> => {
     const chatDetails = await getSavedChatTags(context, false);
 
@@ -406,14 +407,24 @@ export const chatResumeSubCommands: SlashCommand[] = [
   checkpointCompatibilityCommand,
 ];
 
+import { parseSlashCommand } from '../../utils/commands.js';
+
 export const chatCommand: SlashCommand = {
   name: 'chat',
   description: 'Browse auto-saved conversations and manage chat checkpoints',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
-  action: async () => ({
-    type: 'dialog',
-    dialog: 'sessionBrowser',
-  }),
+  action: async (context, args) => {
+    if (args) {
+      const parsed = parseSlashCommand(`/${args}`, chatResumeSubCommands);
+      if (parsed.commandToExecute?.action) {
+        return parsed.commandToExecute.action(context, parsed.args);
+      }
+    }
+    return {
+      type: 'dialog',
+      dialog: 'sessionBrowser',
+    };
+  },
   subCommands: chatResumeSubCommands,
 };

packages/cli/src/ui/commands/extensionsCommand.ts

@@ -789,6 +789,7 @@ const listExtensionsCommand: SlashCommand = {
   description: 'List active extensions',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
+  takesArgs: false,
   action: listAction,
 };
 
@@ -849,6 +850,7 @@ const exploreExtensionsCommand: SlashCommand = {
   description: 'Open extensions page in your browser',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
+  takesArgs: false,
   action: exploreAction,
 };
 
@@ -870,6 +872,8 @@ const configCommand: SlashCommand = {
   action: configAction,
 };
 
+import { parseSlashCommand } from '../../utils/commands.js';
+
 export function extensionsCommand(
   enableExtensionReloading?: boolean,
 ): SlashCommand {
@@ -883,20 +887,29 @@ export function extensionsCommand(
         configCommand,
       ]
     : [];
+  const subCommands = [
+    listExtensionsCommand,
+    updateExtensionsCommand,
+    exploreExtensionsCommand,
+    reloadCommand,
+    ...conditionalCommands,
+  ];
+
   return {
     name: 'extensions',
     description: 'Manage extensions',
     kind: CommandKind.BUILT_IN,
     autoExecute: false,
-    subCommands: [
-      listExtensionsCommand,
-      updateExtensionsCommand,
-      exploreExtensionsCommand,
-      reloadCommand,
-      ...conditionalCommands,
-    ],
-    action: (context, args) =>
+    subCommands,
+    action: async (context, args) => {
+      if (args) {
+        const parsed = parseSlashCommand(`/${args}`, subCommands);
+        if (parsed.commandToExecute?.action) {
+          return parsed.commandToExecute.action(context, parsed.args);
+        }
+      }
       // Default to list if no subcommand is provided
-      listExtensionsCommand.action!(context, args),
+      return listExtensionsCommand.action!(context, args);
+    },
   };
 }

packages/cli/src/ui/commands/mcpCommand.test.ts

@@ -17,7 +17,7 @@ import {
 } from '@google/gemini-cli-core';
 
 import type { CallableTool } from '@google/genai';
-import { MessageType } from '../types.js';
+import { MessageType, type HistoryItemMcpStatus } from '../types.js';
 
 vi.mock('@google/gemini-cli-core', async (importOriginal) => {
   const actual =
@@ -280,5 +280,41 @@ describe('mcpCommand', () => {
         }),
       );
     });
+
+    it('should filter servers by name when an argument is provided to list', async () => {
+      await mcpCommand.action!(mockContext, 'list server1');
+
+      expect(mockContext.ui.addItem).toHaveBeenCalledWith(
+        expect.objectContaining({
+          type: MessageType.MCP_STATUS,
+          servers: expect.objectContaining({
+            server1: expect.any(Object),
+          }),
+        }),
+      );
+
+      // Should NOT contain server2 or server3
+      const call = vi.mocked(mockContext.ui.addItem).mock
+        .calls[0][0] as HistoryItemMcpStatus;
+      expect(Object.keys(call.servers)).toEqual(['server1']);
+    });
+
+    it('should filter servers by name and show descriptions when an argument is provided to desc', async () => {
+      await mcpCommand.action!(mockContext, 'desc server2');
+
+      expect(mockContext.ui.addItem).toHaveBeenCalledWith(
+        expect.objectContaining({
+          type: MessageType.MCP_STATUS,
+          showDescriptions: true,
+          servers: expect.objectContaining({
+            server2: expect.any(Object),
+          }),
+        }),
+      );
+
+      const call = vi.mocked(mockContext.ui.addItem).mock
+        .calls[0][0] as HistoryItemMcpStatus;
+      expect(Object.keys(call.servers)).toEqual(['server2']);
+    });
   });
 });

packages/cli/src/ui/commands/mcpCommand.ts

@@ -31,6 +31,7 @@ import {
   canLoadServer,
 } from '../../config/mcp/mcpServerEnablement.js';
 import { loadSettings } from '../../config/settings.js';
+import { parseSlashCommand } from '../../utils/commands.js';
 
 const authCommand: SlashCommand = {
   name: 'auth',
@@ -177,6 +178,7 @@ const listAction = async (
   context: CommandContext,
   showDescriptions = false,
   showSchema = false,
+  serverNameFilter?: string,
 ): Promise<void | MessageActionReturn> => {
   const agentContext = context.services.agentContext;
   const config = agentContext?.config;
@@ -199,11 +201,25 @@ const listAction = async (
     };
   }
 
-  const mcpServers = config.getMcpClientManager()?.getMcpServers() || {};
-  const serverNames = Object.keys(mcpServers);
+  let mcpServers = config.getMcpClientManager()?.getMcpServers() || {};
   const blockedMcpServers =
     config.getMcpClientManager()?.getBlockedMcpServers() || [];
 
+  if (serverNameFilter) {
+    const filter = serverNameFilter.trim().toLowerCase();
+    if (filter) {
+      mcpServers = Object.fromEntries(
+        Object.entries(mcpServers).filter(
+          ([name]) =>
+            name.toLowerCase().includes(filter) ||
+            normalizeServerId(name).includes(filter),
+        ),
+      );
+    }
+  }
+
+  const serverNames = Object.keys(mcpServers);
+
   const connectingServers = serverNames.filter(
     (name) => getMCPServerStatus(name) === MCPServerStatus.CONNECTING,
   );
@@ -306,7 +322,7 @@ const listCommand: SlashCommand = {
   description: 'List configured MCP servers and tools',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
-  action: (context) => listAction(context),
+  action: (context, args) => listAction(context, false, false, args),
 };
 
 const descCommand: SlashCommand = {
@@ -315,7 +331,7 @@ const descCommand: SlashCommand = {
   description: 'List configured MCP servers and tools with descriptions',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
-  action: (context) => listAction(context, true),
+  action: (context, args) => listAction(context, true, false, args),
 };
 
 const schemaCommand: SlashCommand = {
@@ -324,7 +340,7 @@ const schemaCommand: SlashCommand = {
     'List configured MCP servers and tools with descriptions and schemas',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
-  action: (context) => listAction(context, true, true),
+  action: (context, args) => listAction(context, true, true, args),
 };
 
 const reloadCommand: SlashCommand = {
@@ -333,6 +349,7 @@ const reloadCommand: SlashCommand = {
   description: 'Reloads MCP servers',
   kind: CommandKind.BUILT_IN,
   autoExecute: true,
+  takesArgs: false,
   action: async (
     context: CommandContext,
   ): Promise<void | SlashCommandActionReturn> => {
@@ -530,5 +547,18 @@ export const mcpCommand: SlashCommand = {
     enableCommand,
     disableCommand,
   ],
-  action: async (context: CommandContext) => listAction(context),
+  action: async (
+    context: CommandContext,
+    args: string,
+  ): Promise<void | SlashCommandActionReturn> => {
+    if (args) {
+      const parsed = parseSlashCommand(`/${args}`, mcpCommand.subCommands!);
+      if (parsed.commandToExecute?.action) {
+        return parsed.commandToExecute.action(context, parsed.args);
+      }
+      // If no subcommand matches, treat the whole args as a filter for list
+      return listAction(context, false, false, args);
+    }
+    return listAction(context);
+  },
 };

packages/cli/src/ui/commands/planCommand.test.ts

@@ -104,6 +104,47 @@ describe('planCommand', () => {
     );
   });
 
+  it('should not return a submit_prompt action if arguments are empty', async () => {
+    vi.mocked(
+      mockContext.services.agentContext!.config.isPlanEnabled,
+    ).mockReturnValue(true);
+    mockContext.invocation = {
+      raw: '/plan',
+      name: 'plan',
+      args: '',
+    };
+
+    if (!planCommand.action) throw new Error('Action missing');
+    const result = await planCommand.action(mockContext, '');
+
+    expect(result).toBeUndefined();
+    expect(
+      mockContext.services.agentContext!.config.setApprovalMode,
+    ).toHaveBeenCalledWith(ApprovalMode.PLAN);
+  });
+
+  it('should return a submit_prompt action if arguments are provided', async () => {
+    vi.mocked(
+      mockContext.services.agentContext!.config.isPlanEnabled,
+    ).mockReturnValue(true);
+    mockContext.invocation = {
+      raw: '/plan implement auth',
+      name: 'plan',
+      args: 'implement auth',
+    };
+
+    if (!planCommand.action) throw new Error('Action missing');
+    const result = await planCommand.action(mockContext, 'implement auth');
+
+    expect(result).toEqual({
+      type: 'submit_prompt',
+      content: 'implement auth',
+    });
+    expect(
+      mockContext.services.agentContext!.config.setApprovalMode,
+    ).toHaveBeenCalledWith(ApprovalMode.PLAN);
+  });
+
   it('should display the approved plan from config', async () => {
     const mockPlanPath = '/mock/plans/dir/approved-plan.md';
     vi.mocked(

packages/cli/src/ui/commands/planCommand.ts

@@ -66,6 +66,13 @@ export const planCommand: SlashCommand = {
       coreEvents.emitFeedback('info', 'Switched to Plan Mode.');
     }
 
+    if (context.invocation?.args) {
+      return {
+        type: 'submit_prompt',
+        content: context.invocation.args,
+      };
+    }
+
     const approvedPlanPath = config.getApprovedPlanPath();
 
     if (!approvedPlanPath) {
@@ -86,12 +93,14 @@ export const planCommand: SlashCommand = {
         type: MessageType.GEMINI,
         text: partToString(content.llmContent),
       });
+      return;
     } catch (error) {
       coreEvents.emitFeedback(
         'error',
         `Failed to read approved plan at ${approvedPlanPath}: ${error}`,
         error,
       );
+      return;
     }
   },
   subCommands: [
@@ -100,6 +109,7 @@ export const planCommand: SlashCommand = {
       description: 'Copy the currently approved plan to your clipboard',
       kind: CommandKind.BUILT_IN,
       autoExecute: true,
+      takesArgs: false,
       action: copyAction,
     },
   ],

packages/cli/src/ui/commands/skillsCommand.ts

@@ -357,6 +357,8 @@ function enableCompletion(
     .map((s) => s.name);
 }
 
+import { parseSlashCommand } from '../../utils/commands.js';
+
 export const skillsCommand: SlashCommand = {
   name: 'skills',
   description:
@@ -402,5 +404,13 @@ export const skillsCommand: SlashCommand = {
       action: reloadAction,
     },
   ],
-  action: listAction,
+  action: async (context, args) => {
+    if (args) {
+      const parsed = parseSlashCommand(`/${args}`, skillsCommand.subCommands!);
+      if (parsed.commandToExecute?.action) {
+        return parsed.commandToExecute.action(context, parsed.args);
+      }
+    }
+    return listAction(context, args);
+  },
 };

packages/cli/src/ui/commands/types.ts

@@ -240,5 +240,14 @@ export interface SlashCommand {
    */
   showCompletionLoading?: boolean;
 
+  /**
+   * Whether the command expects arguments.
+   * If false, and the command is a subcommand, the command parser may treat
+   * any following text as arguments for the parent command instead of this subcommand,
+   * provided the parent command has an action.
+   * Defaults to true.
+   */
+  takesArgs?: boolean;
+
   subCommands?: SlashCommand[];
 }

packages/cli/src/utils/commands.test.ts

@@ -137,4 +137,105 @@ describe('parseSlashCommand', () => {
     expect(result.args).toBe('');
     expect(result.canonicalPath).toEqual([]);
   });
+
+  describe('backtracking', () => {
+    const backtrackingCommands: readonly SlashCommand[] = [
+      {
+        name: 'parent',
+        description: 'Parent command',
+        kind: CommandKind.BUILT_IN,
+        action: async () => {},
+        subCommands: [
+          {
+            name: 'notakes',
+            description: 'Subcommand that does not take arguments',
+            kind: CommandKind.BUILT_IN,
+            takesArgs: false,
+            action: async () => {},
+          },
+          {
+            name: 'takes',
+            description: 'Subcommand that takes arguments',
+            kind: CommandKind.BUILT_IN,
+            takesArgs: true,
+            action: async () => {},
+          },
+        ],
+      },
+    ];
+
+    it('should backtrack to parent if subcommand has takesArgs: false and args are provided', () => {
+      const result = parseSlashCommand(
+        '/parent notakes some prompt',
+        backtrackingCommands,
+      );
+      expect(result.commandToExecute?.name).toBe('parent');
+      expect(result.args).toBe('notakes some prompt');
+      expect(result.canonicalPath).toEqual(['parent']);
+    });
+
+    it('should NOT backtrack if subcommand has takesArgs: false but NO args are provided', () => {
+      const result = parseSlashCommand('/parent notakes', backtrackingCommands);
+      expect(result.commandToExecute?.name).toBe('notakes');
+      expect(result.args).toBe('');
+      expect(result.canonicalPath).toEqual(['parent', 'notakes']);
+    });
+
+    it('should NOT backtrack if subcommand has takesArgs: true and args are provided', () => {
+      const result = parseSlashCommand(
+        '/parent takes some args',
+        backtrackingCommands,
+      );
+      expect(result.commandToExecute?.name).toBe('takes');
+      expect(result.args).toBe('some args');
+      expect(result.canonicalPath).toEqual(['parent', 'takes']);
+    });
+
+    it('should NOT backtrack if parent has NO action', () => {
+      const noActionCommands: readonly SlashCommand[] = [
+        {
+          name: 'parent',
+          description: 'Parent without action',
+          kind: CommandKind.BUILT_IN,
+          subCommands: [
+            {
+              name: 'notakes',
+              description: 'Subcommand without args',
+              kind: CommandKind.BUILT_IN,
+              takesArgs: false,
+              action: async () => {},
+            },
+          ],
+        },
+      ];
+      const result = parseSlashCommand(
+        '/parent notakes some args',
+        noActionCommands,
+      );
+      // It stays with the subcommand because parent can't handle it
+      expect(result.commandToExecute?.name).toBe('notakes');
+      expect(result.args).toBe('some args');
+      expect(result.canonicalPath).toEqual(['parent', 'notakes']);
+    });
+
+    it('should NOT backtrack if subcommand is NOT marked with takesArgs: false', () => {
+      const result = parseSlashCommand(
+        '/parent takes some args',
+        backtrackingCommands,
+      );
+      expect(result.commandToExecute?.name).toBe('takes');
+      expect(result.args).toBe('some args');
+      expect(result.canonicalPath).toEqual(['parent', 'takes']);
+    });
+
+    it('should backtrack if subcommand has takesArgs: false and args are provided (like /plan copy foo)', () => {
+      const result = parseSlashCommand(
+        '/parent notakes some prompt',
+        backtrackingCommands,
+      );
+      expect(result.commandToExecute?.name).toBe('parent');
+      expect(result.args).toBe('notakes some prompt');
+      expect(result.canonicalPath).toEqual(['parent']);
+    });
+  });
 });

packages/cli/src/utils/commands.ts

@@ -33,6 +33,7 @@ export const parseSlashCommand = (
   let commandToExecute: SlashCommand | undefined;
   let pathIndex = 0;
   const canonicalPath: string[] = [];
+  let parentCommand: SlashCommand | undefined;
 
   for (const part of commandPath) {
     // TODO: For better performance and architectural clarity, this two-pass
@@ -52,6 +53,7 @@ export const parseSlashCommand = (
     }
 
     if (foundCommand) {
+      parentCommand = commandToExecute;
       commandToExecute = foundCommand;
       canonicalPath.push(foundCommand.name);
       pathIndex++;
@@ -67,5 +69,21 @@ export const parseSlashCommand = (
 
   const args = parts.slice(pathIndex).join(' ');
 
+  // Backtrack if the matched (sub)command doesn't take arguments but some were provided,
+  // AND the parent command is capable of handling them.
+  if (
+    commandToExecute &&
+    commandToExecute.takesArgs === false &&
+    args.length > 0 &&
+    parentCommand &&
+    parentCommand.action
+  ) {
+    return {
+      commandToExecute: parentCommand,
+      args: parts.slice(pathIndex - 1).join(' '),
+      canonicalPath: canonicalPath.slice(0, -1),
+    };
+  }
+
   return { commandToExecute, args, canonicalPath };
 };

packages/core/src/code_assist/oauth2.ts

@@ -119,7 +119,8 @@ async function initOauthClient(
     credentials &&
     typeof credentials === 'object' &&
     'type' in credentials &&
-    credentials.type === 'external_account_authorized_user'
+    (credentials.type === 'external_account_authorized_user' ||
+      credentials.type === 'service_account')
   ) {
     const auth = new GoogleAuth({
       scopes: OAUTH_SCOPE,
@@ -130,7 +131,7 @@ async function initOauthClient(
     });
     const token = await byoidClient.getAccessToken();
     if (token) {
-      debugLogger.debug('Created BYOID auth client.');
+      debugLogger.debug(`Created ${credentials.type} auth client.`);
       return byoidClient;
     }
   }

packages/test-utils/src/fixtures/agents.ts

@@ -69,4 +69,84 @@ export const TEST_AGENTS = {
     tools: ['read_file', 'write_file'],
     body: 'You are the test agent. Add or update tests.',
   }),
+  /**
+   * An agent with expertise in database schemas, SQL, and creating database migrations.
+   */
+  DATABASE_AGENT: createAgent({
+    name: 'database-agent',
+    description:
+      'An expert in database schemas, SQL, and creating database migrations.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the database agent. Create and update SQL migrations.',
+  }),
+
+  /**
+   * An agent with expertise in CSS, styling, and UI design.
+   */
+  CSS_AGENT: createAgent({
+    name: 'css-agent',
+    description: 'An expert in CSS, styling, and UI design.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the CSS agent.',
+  }),
+
+  /**
+   * An agent with expertise in internationalization and translations.
+   */
+  I18N_AGENT: createAgent({
+    name: 'i18n-agent',
+    description: 'An expert in internationalization and translations.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the i18n agent.',
+  }),
+
+  /**
+   * An agent with expertise in security audits and vulnerability patches.
+   */
+  SECURITY_AGENT: createAgent({
+    name: 'security-agent',
+    description: 'An expert in security audits and vulnerability patches.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the security agent.',
+  }),
+
+  /**
+   * An agent with expertise in CI/CD, Docker, and deployment scripts.
+   */
+  DEVOPS_AGENT: createAgent({
+    name: 'devops-agent',
+    description: 'An expert in CI/CD, Docker, and deployment scripts.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the devops agent.',
+  }),
+
+  /**
+   * An agent with expertise in tracking, analytics, and metrics.
+   */
+  ANALYTICS_AGENT: createAgent({
+    name: 'analytics-agent',
+    description: 'An expert in tracking, analytics, and metrics.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the analytics agent.',
+  }),
+
+  /**
+   * An agent with expertise in web accessibility and ARIA roles.
+   */
+  ACCESSIBILITY_AGENT: createAgent({
+    name: 'accessibility-agent',
+    description: 'An expert in web accessibility and ARIA roles.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the accessibility agent.',
+  }),
+
+  /**
+   * An agent with expertise in React Native and mobile app development.
+   */
+  MOBILE_AGENT: createAgent({
+    name: 'mobile-agent',
+    description: 'An expert in React Native and mobile app development.',
+    tools: ['read_file', 'write_file'],
+    body: 'You are the mobile agent.',
+  }),
 } as const;

packages/test-utils/src/test-rig.ts

@@ -349,6 +349,7 @@ export class TestRig {
   private _interactiveRuns: InteractiveRun[] = [];
   private _spawnedProcesses: ChildProcess[] = [];
   private _initialized = false;
+  private _modelOverride?: string;
 
   setup(
     testName: string,
@@ -356,9 +357,11 @@ export class TestRig {
       settings?: Record<string, unknown>;
       state?: Record<string, unknown>;
       fakeResponsesPath?: string;
+      model?: string;
     } = {},
   ) {
     this.testName = testName;
+    this._modelOverride = options.model;
     const sanitizedName = sanitizeTestName(testName);
     const testFileDir =
       env['INTEGRATION_TEST_FILE_DIR'] || join(os.tmpdir(), 'gemini-cli-tests');
@@ -430,6 +433,53 @@ export class TestRig {
     // The container mounts the test directory at the same path as the host
     const telemetryPath = join(this.homeDir!, 'telemetry.log'); // Always use home directory for telemetry
 
+    const modelEnv = this._modelOverride || env['GEMINI_MODEL'];
+    const models = modelEnv
+      ? modelEnv.split(',').map((m) => m.trim())
+      : [DEFAULT_GEMINI_MODEL];
+
+    const modelSettings: Record<string, any> = {};
+    if (models.length > 1) {
+      // Setup custom fallback chain for integration tests
+      const chainName = 'integration-test-model';
+      modelSettings['model'] = { name: chainName };
+      modelSettings['experimental'] = { dynamicModelConfiguration: true };
+      modelSettings['modelConfigs'] = {
+        modelDefinitions: {
+          [chainName]: {
+            tier: 'auto',
+            isPreview: true,
+            isVisible: false,
+          },
+        },
+        modelIdResolutions: {
+          [chainName]: {
+            default: models[0],
+          },
+        },
+        modelChains: {
+          [chainName]: models.map((m, i) => ({
+            model: m,
+            isLastResort: i === models.length - 1,
+            actions: {
+              terminal: 'silent',
+              transient: 'silent',
+              not_found: 'silent',
+              unknown: 'silent',
+            },
+            stateTransitions: {
+              terminal: 'terminal',
+              transient: 'terminal',
+              not_found: 'terminal',
+              unknown: 'terminal',
+            },
+          })),
+        },
+      };
+    } else {
+      modelSettings['model'] = { name: models[0] };
+    }
+
     const settings = deepMerge(
       {
         general: {
@@ -454,13 +504,7 @@ export class TestRig {
         ui: {
           useAlternateBuffer: true,
         },
-        ...(env['GEMINI_TEST_TYPE'] === 'integration'
-          ? {
-              model: {
-                name: DEFAULT_GEMINI_MODEL,
-              },
-            }
-          : {}),
+        ...(env['GEMINI_TEST_TYPE'] === 'integration' ? modelSettings : {}),
         sandbox:
           env['GEMINI_SANDBOX'] !== 'false' ? env['GEMINI_SANDBOX'] : false,
         // Don't show the IDE connection dialog when running from VsCode