minovative-mind-cli 2.2.5 → 2.3.0

package/dist/services/agent.js

@@ -27,7 +27,8 @@ const execAsync = promisify(exec);
 import { debugLog, isDebugOn } from '../utils/logger.js';
 import { ensureProjectStorage, ensureIgnored, readCache, writeCache, invalidateCacheForDependents, } from '../utils/projectStorage.js';
 import { GEMINI_MODELS } from '../utils/config.js';
-import { createSharedChatSession, getGeneralChatConfig, getPlanExecutionConfig, compressTextUsingFlashLite, generateChatTitle, getGlobalActiveModel, setGlobalActiveModel, getGlobalLatestUsageMetadata, } from './ai.js';
+import { createSharedChatSession, getGeneralChatConfig, getPlanExecutionConfig, compressTextUsingFlashLite, generateChatTitle, getGlobalActiveModel, setGlobalActiveModel, } from './ai.js';
+import { getAndResetTurnUsage } from './proxyClient.js';
 import { changeLogger } from './changeLogger.js';
 import { chatHistoryService } from './chatHistoryService.js';
 import { gatherContext, routeIntent, evaluateExecutionComplexity } from './contextAgent.js';
@@ -85,7 +86,7 @@ export async function startAgentLoop(workspaceRoot, version) {
     chatHistoryService.init(workspaceRoot);
     const chat = createSharedChatSession();
     const inputHandler = new AsyncInputHandler();
-    const chatSessionState = { id: crypto.randomUUID(), title: '', totalTokens: 0 };
+    const chatSessionState = { id: crypto.randomUUID(), title: '', totalTokens: 0, totalInputTokens: 0, totalOutputTokens: 0 };
     chat.setSessionInfo(chatSessionState.id, workspaceRoot);
     const sessionInputHistory = [];
     let isRawPasteMode = false;
@@ -112,6 +113,8 @@ export async function startAgentLoop(workspaceRoot, version) {
             chatSessionState.id = crypto.randomUUID();
             chatSessionState.title = '';
             chatSessionState.totalTokens = 0;
+            chatSessionState.totalInputTokens = 0;
+            chatSessionState.totalOutputTokens = 0;
             chat.setSessionInfo(chatSessionState.id, workspaceRoot);
         }
         inputHandler.stop();
@@ -131,15 +134,15 @@ export async function startAgentLoop(workspaceRoot, version) {
                 message: 'Command Menu',
                 options: [
                     { value: '/models', label: '/models', hint: 'Change the active AI model' },
-                    { value: '/plan', label: '/plan', hint: 'Toggle plan mode (build a plan without executing)' },
+                    { value: '/plan', label: '/plan', hint: `Toggle plan mode (build a plan without executing) ${isPlanMode ? pc.green('(ON)') : pc.red('(OFF)')}` },
                     { value: '/paste', label: '/paste', hint: 'Paste large text directly into the CLI (Press Ctrl+D to submit)' },
                     { value: '/clear', label: '/clear', hint: 'Clear chat session history' },
-                    { value: '/debug', label: '/debug', hint: 'Toggle internal debug logs' },
-                    { value: '/auto-approve', label: '/auto-approve', hint: 'Approve all future terminal commands' },
+                    { value: '/debug', label: '/debug', hint: `Toggle internal debug logs ${isDebugOn() ? pc.green('(ON)') : pc.red('(OFF)')}` },
+                    { value: '/auto-approve', label: '/auto-approve', hint: `Approve all future terminal commands ${getApprovalMode() === 'skip-all' ? pc.green('(ON)') : pc.red('(OFF)')}` },
                     {
                         value: '/sub-agents',
                         label: '/sub-agents',
-                        hint: 'Toggle the MMAAK Engine for parallel investigation and execution',
+                        hint: `Toggle the MMAAK Engine for parallel investigation and execution ${isSubAgentsEnabled() ? pc.green('(ON)') : pc.red('(OFF)')}`,
                     },
                     {
                         value: '/workspaces',
@@ -311,9 +314,9 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
             // Apply the dynamic prompt updates and tool registrations to the active chat session
             chat.setAgentConfig(dynamicSystemInstruction, config.tools);
             if (getGlobalActiveModel() === GEMINI_MODELS.AUTO) {
-                let selectedModel = GEMINI_MODELS.FLASH_3_5;
+                let selectedModel = GEMINI_MODELS.FLASH;
                 if (effectiveTargetAgent === 'CHAT') {
-                    selectedModel = GEMINI_MODELS.FLASH_LITE_3_1;
+                    selectedModel = GEMINI_MODELS.FLASH_LITE;
                 }
                 else {
                     spinner.stop(); // MUST clear the investigation spinner first to prevent leaking the setInterval
@@ -321,7 +324,7 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                     const complexity = await evaluateExecutionComplexity(finalInput, gatherRes.contextResult?.summary, gatherRes.contextResult?.relevantFiles?.size || 0);
                     spinner.stop();
                     process.stdout.write('\x1b[2K\r');
-                    selectedModel = complexity === 'EASY' ? GEMINI_MODELS.FLASH_LITE_3_1 : GEMINI_MODELS.FLASH_3_5;
+                    selectedModel = complexity === 'EASY' ? GEMINI_MODELS.FLASH_LITE : GEMINI_MODELS.FLASH;
                 }
                 chat.setModel(selectedModel);
                 debugLog(`Auto-routing to model ${selectedModel} based on intent ${effectiveTargetAgent}`);
@@ -330,21 +333,12 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                 chat.setModel(getGlobalActiveModel());
             }
             if (gatherRes.contextResult) {
-                if (!gatherRes.contextResult.isParallel) {
-                    if (!inputHandler.isCurrentlyPrompting()) {
-                        spinner.stop(pc.green('🔍 Investigation complete.'));
-                        spinner.start('Thinking...');
-                    }
-                    else {
-                        p.log.success(pc.green('🔍 Investigation complete.'));
-                    }
+                if (!inputHandler.isCurrentlyPrompting()) {
+                    spinner.stop(pc.green('🔍 Investigation complete.'));
+                    spinner.start('Thinking...');
                 }
                 else {
-                    // Parallel investigation already printed its own detailed summary log
-                    if (!inputHandler.isCurrentlyPrompting()) {
-                        spinner.stop(); // MUST clear the investigation spinner first to prevent leaking the setInterval
-                        spinner.start('Thinking...');
-                    }
+                    p.log.success(pc.green('🔍 Investigation complete.'));
                 }
             }
             else {
@@ -352,6 +346,12 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                     spinner.message('Thinking...');
                 }
             }
+            let gitBranch = '';
+            try {
+                const { stdout } = await execAsync('git branch --show-current', { cwd: workspaceRoot });
+                gitBranch = stdout.trim();
+            }
+            catch { }
             if (!isPlanMode && effectiveTargetAgent === 'EXECUTE' && isSubAgentsEnabled()) {
                 spinner.stop(); // Clear the spinner before delegating
                 process.stdout.write('\x1b[2K\r');
@@ -376,16 +376,19 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                         console.log(marked.parse(cleanText));
                     }
                     // Print Usage Stats
-                    const usage = getGlobalLatestUsageMetadata();
-                    if (usage) {
-                        if (usage.cachedTokens > 0) {
-                            const totalInputTokens = (usage.promptTokens || 0) + usage.cachedTokens;
-                            const percentSaved = totalInputTokens > 0 ? Math.round((usage.cachedTokens / totalInputTokens) * 100) : 0;
-                            p.log.info(`${pc.green('⚡')} ${pc.green('Context Cache Hit:')} ${pc.bold(usage.cachedTokens.toLocaleString())} tokens cached ${pc.dim(`(Saved ~${percentSaved}% of input cost)`)}`);
-                        }
-                        if (usage.remainingBalance !== undefined) {
-                            p.log.info(`${pc.dim('Credits Remaining:')} ${pc.cyan(usage.remainingBalance.toLocaleString())}`);
-                        }
+                    const usage = getAndResetTurnUsage();
+                    chatSessionState.previousUsageMetadata = chatSessionState.latestUsageMetadata;
+                    chatSessionState.latestUsageMetadata = usage;
+                    chatSessionState.totalTokens += usage.totalTokenCount;
+                    chatSessionState.totalInputTokens += (usage.promptTokens || 0) + (usage.cachedTokens || 0);
+                    chatSessionState.totalOutputTokens += usage.candidatesTokens || 0;
+                    if (usage.cachedTokens && usage.cachedTokens > 0) {
+                        const totalInputTokens = (usage.promptTokens || 0) + usage.cachedTokens;
+                        const percentSaved = totalInputTokens > 0 ? Math.round((usage.cachedTokens / totalInputTokens) * 100) : 0;
+                        p.log.info(`${pc.green('⚡')} ${pc.green('Context Cache Hit:')} ${pc.bold(usage.cachedTokens.toLocaleString())} tokens cached ${pc.dim(`(Saved ~${percentSaved}% of input cost)`)}`);
+                    }
+                    if (usage.remainingBalance !== undefined) {
+                        p.log.info(`${pc.dim('Credits Remaining:')} ${pc.cyan(usage.remainingBalance.toLocaleString())}`);
                     }
                     // Post-execution steps for the UI (saving history, invalidate cache)
                     const postExecutionChanges = changeLogger.getCurrentChangeSet()?.changes || [];
@@ -419,6 +422,17 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                             title: chatSessionState.title,
                             timestamp: Date.now(),
                             history,
+                            creditsRemaining: usage.remainingBalance,
+                            lastTurnDuration: turnDuration,
+                            modelName: chat.getModel(),
+                            totalTokens: chatSessionState.totalTokens,
+                            totalInputTokens: chatSessionState.totalInputTokens,
+                            totalOutputTokens: chatSessionState.totalOutputTokens,
+                            gitBranch: gitBranch || undefined,
+                            autoApprove: getApprovalMode() === 'skip-all',
+                            subAgents: isSubAgentsEnabled(),
+                            latestUsageMetadata: chatSessionState.latestUsageMetadata,
+                            previousUsageMetadata: chatSessionState.previousUsageMetadata,
                         })
                             .catch((e) => debugLog('Failed to auto-save session: ' + e));
                     }
@@ -458,7 +472,7 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
             const correctionRes = await executeSelfCorrectionLoop(chat, result, workspaceRoot, inputHandler, effectiveTargetAgent, ac.signal, spinner, userInput, isPlanMode);
             const finalText = correctionRes.finalText;
             // Update latest usage metadata to reflect all completed turns
-            latestUsage = chat.getLatestUsageMetadata() || latestUsage;
+            const usage = getAndResetTurnUsage();
             const postExecutionChanges = changeLogger.getCurrentChangeSet()?.changes || [];
             const modifiedFiles = postExecutionChanges
                 .filter((c) => c.action === 'modify' || c.action === 'create' || c.action === 'delete')
@@ -475,18 +489,18 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                     console.log(marked.parse(cleanText));
                 }
             }
-            if (latestUsage) {
-                if (latestUsage.cachedTokens && latestUsage.cachedTokens > 0) {
-                    const totalInputTokens = (latestUsage.promptTokens || 0) + latestUsage.cachedTokens;
-                    const percentSaved = totalInputTokens > 0 ? Math.round((latestUsage.cachedTokens / totalInputTokens) * 100) : 0;
-                    p.log.info(`${pc.green('⚡')} ${pc.green('Context Cache Hit:')} ${pc.bold(latestUsage.cachedTokens.toLocaleString())} tokens cached ${pc.dim(`(Saved ~${percentSaved}% of input cost)`)}`);
+            if (usage) {
+                if (usage.cachedTokens && usage.cachedTokens > 0) {
+                    const totalInputTokens = (usage.promptTokens || 0) + usage.cachedTokens;
+                    const percentSaved = totalInputTokens > 0 ? Math.round((usage.cachedTokens / totalInputTokens) * 100) : 0;
+                    p.log.info(`${pc.green('⚡')} ${pc.green('Context Cache Hit:')} ${pc.bold(usage.cachedTokens.toLocaleString())} tokens cached ${pc.dim(`(Saved ~${percentSaved}% of input cost)`)}`);
                 }
-                const inputTokens = (latestUsage.promptTokens || 0) + (latestUsage.cachedTokens || 0);
-                const outputTokens = latestUsage.candidatesTokens || 0;
-                const totalTokens = latestUsage.totalTokenCount || inputTokens + outputTokens;
+                const inputTokens = (usage.promptTokens || 0) + (usage.cachedTokens || 0);
+                const outputTokens = usage.candidatesTokens || 0;
+                const totalTokens = inputTokens + outputTokens;
                 p.log.info(`${pc.dim('Tokens Used:')} ${pc.cyan(totalTokens.toLocaleString())} ${pc.dim(`(Input: ${inputTokens.toLocaleString()}, Output: ${outputTokens.toLocaleString()})`)}`);
-                if (latestUsage.remainingBalance !== undefined) {
-                    p.log.info(`${pc.dim('Credits Remaining:')} ${pc.cyan(latestUsage.remainingBalance.toLocaleString())}`);
+                if (usage.remainingBalance !== undefined) {
+                    p.log.info(`${pc.dim('Credits Remaining:')} ${pc.cyan(usage.remainingBalance.toLocaleString())}`);
                 }
             }
             const turnEndTime = Date.now();
@@ -496,18 +510,13 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
             if (finalText !== '[Generation stopped.]') {
                 changeLogger.markComplete();
             }
-            let turnTokens = 0;
-            if (latestUsage) {
-                turnTokens =
-                    (latestUsage.promptTokens || 0) + (latestUsage.cachedTokens || 0) + (latestUsage.candidatesTokens || 0);
+            if (usage) {
+                chatSessionState.previousUsageMetadata = chatSessionState.latestUsageMetadata;
+                chatSessionState.latestUsageMetadata = usage;
+                chatSessionState.totalTokens += usage.totalTokenCount || 0;
+                chatSessionState.totalInputTokens += (usage.promptTokens || 0) + (usage.cachedTokens || 0);
+                chatSessionState.totalOutputTokens += usage.candidatesTokens || 0;
             }
-            chatSessionState.totalTokens += turnTokens;
-            let gitBranch = '';
-            try {
-                const { stdout } = await execAsync('git branch --show-current', { cwd: workspaceRoot });
-                gitBranch = stdout.trim();
-            }
-            catch { }
             // Auto-save chat history
             const history = chat.getRawHistory();
             if (history.length > 0) {
@@ -525,9 +534,13 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                             lastTurnDuration: turnDuration,
                             modelName: chat.getModel(),
                             totalTokens: chatSessionState.totalTokens,
+                            totalInputTokens: chatSessionState.totalInputTokens,
+                            totalOutputTokens: chatSessionState.totalOutputTokens,
                             gitBranch: gitBranch || undefined,
                             autoApprove: getApprovalMode() === 'skip-all',
                             subAgents: isSubAgentsEnabled(),
+                            latestUsageMetadata: chatSessionState.latestUsageMetadata,
+                            previousUsageMetadata: chatSessionState.previousUsageMetadata,
                         });
                     }
                     catch (e) {
@@ -544,9 +557,13 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                     lastTurnDuration: turnDuration,
                     modelName: chat.getModel(),
                     totalTokens: chatSessionState.totalTokens,
+                    totalInputTokens: chatSessionState.totalInputTokens,
+                    totalOutputTokens: chatSessionState.totalOutputTokens,
                     gitBranch: gitBranch || undefined,
                     autoApprove: getApprovalMode() === 'skip-all',
                     subAgents: isSubAgentsEnabled(),
+                    latestUsageMetadata: chatSessionState.latestUsageMetadata,
+                    previousUsageMetadata: chatSessionState.previousUsageMetadata,
                 })
                     .catch((e) => debugLog('Failed to auto-save session: ' + e));
             }
@@ -584,19 +601,19 @@ export async function executeSingleTurn(workspaceRoot, userInput, chat, inputHan
                                     hint: 'The Pro model for complex logic',
                                 },
                                 {
-                                    value: 'gemini-3.5-flash',
-                                    label: 'Gemini 3.5 Flash',
+                                    value: 'gemini-3.6-flash',
+                                    label: 'Gemini 3.6 Flash',
                                     hint: 'Balanced performance',
                                 },
                                 {
-                                    value: 'gemini-3.1-flash-lite',
-                                    label: 'Gemini 3.1 Flash-Lite',
-                                    hint: 'Ultra-fast and cost-effective',
+                                    value: 'gemini-3.5-flash-lite',
+                                    label: 'Gemini 3.5 Flash-Lite',
+                                    hint: 'Best for speed and cost efficiency',
                                 },
                                 {
                                     value: 'auto',
                                     label: 'Auto (Flash-Lite / Flash)',
-                                    hint: 'Dynamically routes between Gemini 3.1 Flash-Lite and Gemini 3.5 based on prompt complexity',
+                                    hint: 'Dynamically routes between Gemini 3.5 Flash-Lite and Gemini 3.6 based on prompt complexity',
                                 },
                             ],
                         });

package/dist/services/ai.d.ts

@@ -3,7 +3,6 @@ export declare function setModelOverride(agent: 'context' | 'execution', overrid
 export declare function clearModelOverrides(): void;
 export declare function setGlobalActiveModel(model: string): void;
 export declare function getGlobalActiveModel(): string;
-export declare function getGlobalLatestUsageMetadata(): any;
 export declare class ProxyChatSession {
     private history;
     private modelName;
@@ -74,7 +73,7 @@ export declare function getPlanModeConfig(): {
     tools: never[];
 };
 /**
- * Compresses a large string of text using gemini-3.1-flash-lite.
+ * Compresses a large string of text using gemini-3.5-flash-lite.
  * Used for shrinking context payloads to prevent OOM/choking.
  */
 export declare function compressTextUsingFlashLite(text: string, instruction?: string, inlineData?: any): Promise<string>;

package/dist/services/ai.js

@@ -52,10 +52,6 @@ export function setGlobalActiveModel(model) {
 export function getGlobalActiveModel() {
     return globalActiveModel;
 }
-let globalLatestUsageMetadata = undefined;
-export function getGlobalLatestUsageMetadata() {
-    return globalLatestUsageMetadata;
-}
 const proxyClient = new ProxyClient();
 // ─── History Limits ──────────────────────────────────────────────────
 /** Maximum number of Content entries to keep in the sliding history window. */
@@ -214,9 +210,6 @@ export class ProxyChatSession {
         this.systemInstruction, effectiveGenerationConfig, onChunk ? { onChunk } : undefined, // streamCallbacks
         abortSignal);
         this.latestUsageMetadata = result.usageMetadata;
-        if (result.usageMetadata) {
-            globalLatestUsageMetadata = result.usageMetadata;
-        }
         // Track token usage metrics
         if (result.usageMetadata) {
             const collector = getMetricCollector();
@@ -283,7 +276,7 @@ export class ProxyChatSession {
 export function createSharedChatSession() {
     let model = executionModelOverride || getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_3_5; // Will be overridden per-turn in executeSingleTurn
+        model = GEMINI_MODELS.FLASH; // Will be overridden per-turn in executeSingleTurn
     return new ProxyChatSession(model, GENERAL_CHAT_INSTRUCTION, [], {
         maxOutputTokens: MAX_OUTPUT_TOKENS,
         temperature: executionTempOverride !== null ? executionTempOverride : 1,
@@ -310,10 +303,10 @@ export function getPlanModeConfig() {
     };
 }
 /**
- * Compresses a large string of text using gemini-3.1-flash-lite.
+ * Compresses a large string of text using gemini-3.5-flash-lite.
  * Used for shrinking context payloads to prevent OOM/choking.
  */
-export async function compressTextUsingFlashLite(text, instruction = 'Summarize the following text concisely. Preserve the most critical technical details, function names, and architecture logic. Keep it under 1500 characters.', inlineData) {
+export async function compressTextUsingFlashLite(text, instruction = "<directives>\nSummarize the following text concisely. Preserve the most critical technical details, function names, and architecture logic. Make sure it's understandable without the fluff.\n</directives>", inlineData) {
     if (!text || (text.length < 1000 && !inlineData))
         return text; // Don't compress tiny texts
     try {
@@ -322,7 +315,7 @@ export async function compressTextUsingFlashLite(text, instruction = 'Summarize
             return text;
         let model = getGlobalActiveModel();
         if (model === 'auto')
-            model = GEMINI_MODELS.FLASH_LITE_3_1;
+            model = GEMINI_MODELS.FLASH_LITE;
         const parts = [{ text }];
         if (inlineData) {
             parts.push({ inlineData });
@@ -461,8 +454,17 @@ export function getContextToolDeclarations() {
                     description: 'Write and execute a disposable analysis script to structurally map code in the workspace. ' +
                         'Use this to get exact line ranges for functions, classes, and variables by leveraging the ' +
                         "language's native AST parser (e.g., TypeScript compiler API, Python ast module, go/parser). " +
+                        'You can also use this to probe the development environment — detecting available runtimes ' +
+                        '(e.g., node --version, python3 --version), checking if ports are in use, identifying project ' +
+                        'type (monorepo, package manager), or diagnosing system-level issues (disk space, memory) that ' +
+                        'may affect execution. ' +
+                        'For complex investigation tasks, you can write lightweight ML scripts (e.g., TF-IDF cosine similarity ' +
+                        'to rank file relevance, Z-score outlier detection for anomalous log lines, K-Means clustering, or Naive ' +
+                        'Bayes classification). Default to "node" for generic math/analysis as a safe baseline, but act like ' +
+                        'a native inhabitant of the host environment — if Python, Go, Rust, or specialized libraries are available ' +
+                        'in the project context, leverage the host\'s native runtimes and standard libraries for maximum efficiency. ' +
                         'The script is executed from a temporary directory and automatically cleaned up after execution. ' +
-                        'Output should be structured JSON to stdout with name, type, startLine, endLine for each code element. ' +
+                        'Output should be structured JSON to stdout. ' +
                         'Use the results to make precise read_file calls with exact startLine/endLine instead of guessing. ' +
                         'CRITICAL: Do not use this tool on binary, document, or non-code files (e.g. PDF, image, audio, docx).',
                     parameters: {
@@ -519,7 +521,7 @@ export function createContextAgentSession() {
     const contextTools = getContextToolDeclarations();
     let model = contextModelOverride || getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_3_5;
+        model = GEMINI_MODELS.FLASH;
     return new ProxyChatSession(model, CONTEXT_SYSTEM_INSTRUCTION.replace('{{MULTI_WORKSPACE_BLOCK}}', getMultiWorkspaceBlock()), contextTools, {
         maxOutputTokens: MAX_OUTPUT_TOKENS,
         temperature: contextTempOverride !== null ? contextTempOverride : 1,
@@ -531,7 +533,7 @@ export function createContextAgentSession() {
 export function createIntentRouterSession() {
     let model = getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_LITE_3_1;
+        model = GEMINI_MODELS.FLASH_LITE;
     return new ProxyChatSession(model, INTENT_ROUTER_SYSTEM_INSTRUCTION, [], // no tools
     { temperature: 0, responseMimeType: 'application/json' });
 }
@@ -539,7 +541,7 @@ export function createIntentRouterSession() {
 export function createExecutionComplexitySession() {
     let model = getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_LITE_3_1;
+        model = GEMINI_MODELS.FLASH_LITE;
     return new ProxyChatSession(model, EXECUTION_COMPLEXITY_SYSTEM_INSTRUCTION, [], // no tools
     { temperature: 0, responseMimeType: 'application/json' });
 }
@@ -547,7 +549,7 @@ export function createExecutionComplexitySession() {
 export function createInvestigationComplexitySession() {
     let model = getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_LITE_3_1;
+        model = GEMINI_MODELS.FLASH_LITE;
     return new ProxyChatSession(model, INVESTIGATION_COMPLEXITY_SYSTEM_INSTRUCTION, [], // no tools
     { temperature: 0, responseMimeType: 'application/json' });
 }
@@ -555,7 +557,7 @@ export function createInvestigationComplexitySession() {
 export function createWebSearchAgentSession() {
     let model = getGlobalActiveModel();
     if (model === 'auto')
-        model = GEMINI_MODELS.FLASH_3_5;
+        model = GEMINI_MODELS.FLASH;
     return new ProxyChatSession(model, WEB_SEARCH_SYSTEM_INSTRUCTION, [{ googleSearch: {} }], {
         maxOutputTokens: MAX_OUTPUT_TOKENS,
         temperature: 1,
@@ -574,11 +576,11 @@ export async function generateChatTitle(firstMessage) {
         const idToken = await getAuthorizedIdToken();
         if (!idToken)
             return firstMessage.substring(0, maxLength);
-        const instruction = "You are a helpful assistant that generates extremely concise chat titles based on a user's first message. Output ONLY the title, no quotes, no markdown, no punctuation.";
+        const instruction = "<identity>\nYou are a helpful assistant that generates extremely concise chat titles based on a user's first message.\n</identity>\n\n<directives>\nOutput ONLY the title, no quotes, no markdown, no punctuation.\n</directives>";
         const contents = [{ role: 'user', parts: [{ text: firstMessage.substring(0, 500) }] }];
         let model = getGlobalActiveModel();
         if (model === 'auto')
-            model = GEMINI_MODELS.FLASH_LITE_3_1;
+            model = GEMINI_MODELS.FLASH_LITE;
         const result = await proxyClient.generateFunctionCallViaProxy(idToken, model, contents, [], // no tools
         undefined, instruction, { temperature: 0.2 });
         let title = '';

package/dist/services/auth.d.ts

@@ -1,3 +1,3 @@
 export declare function login(): Promise<boolean>;
-export declare function logout(): void;
+export declare function logout(): Promise<void>;
 export declare function getAuthorizedIdToken(): Promise<string | undefined>;

package/dist/services/auth.js

@@ -1,23 +1,5 @@
-import * as fs from 'fs';
-import * as path from 'path';
-import * as os from 'os';
 import { FIREBASE_API_KEY, GITHUB_CLIENT_ID } from '../utils/config.js';
-const CONFIG_FILE = path.join(os.homedir(), '.minovative-mind-cli.json');
-function getAuthData() {
-    try {
-        if (fs.existsSync(CONFIG_FILE)) {
-            return JSON.parse(fs.readFileSync(CONFIG_FILE, 'utf-8'));
-        }
-    }
-    catch (err) {
-        // ignore
-    }
-    return {};
-}
-function saveAuthData(data) {
-    const current = getAuthData();
-    fs.writeFileSync(CONFIG_FILE, JSON.stringify({ ...current, ...data }, null, 2));
-}
+import { saveCredentials, loadCredentials, clearCredentials } from '../utils/credentialStore.js';
 export async function login() {
     try {
         // 1. Start GitHub Device Flow
@@ -98,8 +80,8 @@ export async function login() {
         if (!firebaseTokenResponse.idToken) {
             throw new Error('Failed to get Firebase ID token.');
         }
-        // 5. Store the Firebase token securely
-        saveAuthData({
+        // 5. Store the Firebase token in the OS-native secure credential store
+        await saveCredentials({
             idToken: firebaseTokenResponse.idToken,
             refreshToken: firebaseTokenResponse.refreshToken,
             idTokenExpiry: Date.now() + 50 * 60 * 1000, // 50 mins
@@ -139,14 +121,11 @@ async function exchangeGithubTokenForFirebase(githubAccessToken) {
     }
     return await signInResponse.json();
 }
-export function logout() {
-    if (fs.existsSync(CONFIG_FILE)) {
-        fs.unlinkSync(CONFIG_FILE);
-    }
-    console.log('Successfully signed out.');
+export async function logout() {
+    await clearCredentials();
 }
 export async function getAuthorizedIdToken() {
-    const data = getAuthData();
+    const data = await loadCredentials();
     if (!data.idToken)
         return undefined;
     if (data.idTokenExpiry && Date.now() < data.idTokenExpiry) {
@@ -168,7 +147,7 @@ export async function getAuthorizedIdToken() {
         const refreshData = (await response.json());
         const newIdToken = refreshData.id_token;
         if (newIdToken) {
-            saveAuthData({
+            await saveCredentials({
                 idToken: newIdToken,
                 refreshToken: refreshData.refresh_token || data.refreshToken,
                 idTokenExpiry: Date.now() + 50 * 60 * 1000,

package/dist/services/chatHistoryService.d.ts

@@ -21,12 +21,20 @@ export interface ChatSessionData {
     modelName?: string;
     /** Optional total cumulative tokens consumed during the session. */
     totalTokens?: number;
+    /** Optional total cumulative input tokens (prompt + cached) consumed during the session. */
+    totalInputTokens?: number;
+    /** Optional total cumulative output tokens consumed during the session. */
+    totalOutputTokens?: number;
     /** Optional name of the Git branch active when this session was last updated. */
     gitBranch?: string;
     /** Optional flag indicating whether auto-approval of commands was enabled in this session. */
     autoApprove?: boolean;
     /** Optional flag indicating whether sub-agents were enabled or active during this session. */
     subAgents?: boolean;
+    /** Optional metadata about the token usage from the very last turn. */
+    latestUsageMetadata?: any;
+    /** Optional metadata about the token usage from the turn prior to the last turn. */
+    previousUsageMetadata?: any;
 }
 /**
  * Service responsible for managing the persistence, retrieval, and deletion of chat session histories.

package/dist/services/contextAgent.js

@@ -232,6 +232,8 @@ export async function gatherContext(workspaceRoot, userRequest, chatHistory = ''
     let summary = 'No relevant context found.';
     let isInvestigationFinished = false;
     const visitedToolCalls = new Set();
+    let consecutiveDuplicates = 0;
+    const MAX_CONSECUTIVE_DUPLICATES = 3;
     let chainedMessages = [];
     let webSearchSummary = '';
     // Initial prompt
@@ -286,14 +288,29 @@ export async function gatherContext(workspaceRoot, userRequest, chatHistory = ''
             // Deduplicate identical tool calls
             const callSignature = `${call.name}:${JSON.stringify(args)}`;
             if (call.name !== 'finish_investigation' && visitedToolCalls.has(callSignature)) {
+                consecutiveDuplicates++;
+                if (consecutiveDuplicates >= MAX_CONSECUTIVE_DUPLICATES) {
+                    // Force-finish: the model is stuck in a loop, finalize with whatever context we have
+                    const forceMsg = `Investigation auto-completed: the model repeated the same tool call ${MAX_CONSECUTIVE_DUPLICATES} times consecutively.`;
+                    debugLog(`[Context Agent] ${forceMsg}`);
+                    if (onProgress)
+                        onProgress(forceMsg);
+                    if (!summary || summary === 'No relevant context found.') {
+                        summary = 'Investigation was auto-completed due to repeated duplicate tool calls. Review the gathered files for context.';
+                    }
+                    isFinished = true;
+                    isInvestigationFinished = relevantFiles.size > 0;
+                    break;
+                }
                 functionResponses.push({
                     functionResponse: {
                         name: call.name,
-                        response: { error: 'You have already made this exact tool call previously. Please review your context history or try a different action.' }
+                        response: { error: `DUPLICATE CALL BLOCKED (attempt ${consecutiveDuplicates}/${MAX_CONSECUTIVE_DUPLICATES}): You already executed this exact tool call. Do NOT retry it. Use the results you already have and call finish_investigation now, or try a DIFFERENT tool call with different parameters.` }
                     }
                 });
                 continue;
             }
+            consecutiveDuplicates = 0;
             visitedToolCalls.add(callSignature);
             let logMsg = `  [Context Agent] Executing ${call.name}`;
             if (call.name === 'finish_investigation') {

package/dist/services/investigationComplexity.d.ts

@@ -31,7 +31,7 @@ export interface InvestigationComplexityResult {
 /**
  * Evaluates whether the investigation phase should be parallelized.
  *
- * Uses `gemini-3.1-flash-lite` (under auto mode) at temperature 0 to classify the prompt's
+ * Uses `gemini-3.5-flash-lite` (under auto mode) at temperature 0 to classify the prompt's
  * investigation complexity. The PM dynamically identifies domains and groups
  * them into agent assignments. Agent count = `agentAssignments.length`, which
  * may be fewer than `domains.length` when related domains are batched together.

package/dist/services/investigationComplexity.js

@@ -16,7 +16,7 @@ import { debugLog } from '../utils/logger.js';
 /**
  * Evaluates whether the investigation phase should be parallelized.
  *
- * Uses `gemini-3.1-flash-lite` (under auto mode) at temperature 0 to classify the prompt's
+ * Uses `gemini-3.5-flash-lite` (under auto mode) at temperature 0 to classify the prompt's
  * investigation complexity. The PM dynamically identifies domains and groups
  * them into agent assignments. Agent count = `agentAssignments.length`, which
  * may be fewer than `domains.length` when related domains are batched together.

package/dist/services/orchestration/investigationAgent.d.ts

@@ -59,8 +59,8 @@ export declare class InvestigationAgentRunner {
     private creditsUsed;
     private inputTokens;
     private outputTokens;
-    /** Shorter stall timeout for investigation agents (120s to allow for heavy vision/web tasks) */
-    static readonly STALL_TIMEOUT_MS = 120000;
+    /** Stall timeout for investigation agents (300s to allow for heavy vision/web tasks) */
+    static readonly STALL_TIMEOUT_MS = 300000;
     constructor(agentLabel: string, domains: string[], workspaceRoot: string, readCache: ReadCache, projectTree: string, projectType: string);
     /**
      * Builds a domain-scoped system instruction. Extends the base Context Agent