minovative-mind-cli 2.13.4 → 2.14.0

package/dist/services/ai.js

@@ -1,11 +1,11 @@
 import { SchemaType } from '@google/generative-ai';
-import { GEMINI_MODELS, DEFAULT_MODEL, MAX_OUTPUT_TOKENS, isByokEnabled } from '../utils/config.js';
+import { GEMINI_MODELS, DEFAULT_MODEL, MAX_OUTPUT_TOKENS, isByokEnabled, DEFAULT_MODEL_THINKING_LEVELS, } from '../utils/config.js';
 import { getToolDeclarations } from './agent-tools.js';
 import { getMetricCollector } from './metrics.js';
 import { getAuthorizedIdToken, checkByokSubscription } from './auth.js';
 import { debugLog } from '../utils/logger.js';
 import { readCache, writeCache } from '../utils/projectStorage.js';
-import { GENERAL_CHAT_INSTRUCTION, PLAN_EXECUTION_INSTRUCTION, PLAN_MODE_INSTRUCTION, CONTEXT_SYSTEM_INSTRUCTION, INTENT_ROUTER_SYSTEM_INSTRUCTION, WEB_SEARCH_SYSTEM_INSTRUCTION, EXECUTION_COMPLEXITY_SYSTEM_INSTRUCTION, INVESTIGATION_COMPLEXITY_SYSTEM_INSTRUCTION, INVESTIGATION_SEMANTIC_SYSTEM_INSTRUCTION, HISTORY_SUMMARIZER_SYSTEM_INSTRUCTION, USER_PROFILE_EXTRACTOR_SYSTEM_INSTRUCTION, TRIVIAL_MESSAGE_CLASSIFIER_SYSTEM_INSTRUCTION, } from '../utils/systemPrompts.js';
+import { GENERAL_CHAT_INSTRUCTION, PLAN_EXECUTION_INSTRUCTION, CONTEXT_SYSTEM_INSTRUCTION, INTENT_ROUTER_SYSTEM_INSTRUCTION, WEB_SEARCH_SYSTEM_INSTRUCTION, EXECUTION_COMPLEXITY_SYSTEM_INSTRUCTION, INVESTIGATION_COMPLEXITY_SYSTEM_INSTRUCTION, INVESTIGATION_SEMANTIC_SYSTEM_INSTRUCTION, HISTORY_SUMMARIZER_SYSTEM_INSTRUCTION, USER_PROFILE_EXTRACTOR_SYSTEM_INSTRUCTION, TRIVIAL_MESSAGE_CLASSIFIER_SYSTEM_INSTRUCTION, } from '../utils/systemPrompts.js';
 import { workspaceRegistry } from './workspaceRegistry.js';
 import { loadCredentials } from '../utils/credentialStore.js';
 import { ProxyClient } from './proxyClient.js';
@@ -56,12 +56,40 @@ export function clearModelOverrides() {
 }
 // ─── AI Service ──────────────────────────────────────────────────────
 let globalActiveModel = DEFAULT_MODEL;
+const userModelThinkingLevels = new Map();
 export function setGlobalActiveModel(model) {
     globalActiveModel = model;
 }
 export function getGlobalActiveModel() {
     return globalActiveModel;
 }
+/**
+ * Gets the thinking reasoning level configured for a specific model or active model.
+ * Falls back to default model thinking levels or 'MEDIUM' if not explicitly configured.
+ */
+export function getModelThinkingLevel(model) {
+    const targetModel = model && model !== 'auto'
+        ? model
+        : globalActiveModel === 'auto'
+            ? GEMINI_MODELS.FLASH
+            : globalActiveModel;
+    if (userModelThinkingLevels.has(targetModel)) {
+        return userModelThinkingLevels.get(targetModel);
+    }
+    return DEFAULT_MODEL_THINKING_LEVELS[targetModel] || 'MEDIUM';
+}
+/**
+ * Sets the user-configured thinking reasoning level for a specific model.
+ */
+export function setModelThinkingLevel(model, level) {
+    userModelThinkingLevels.set(model, level);
+}
+/**
+ * Resets user-configured thinking levels back to default model presets.
+ */
+export function resetModelThinkingLevels() {
+    userModelThinkingLevels.clear();
+}
 const proxyClient = new ProxyClient();
 // ─── History Limits ──────────────────────────────────────────────────
 /** Maximum number of Content entries to keep in the sliding history window. */
@@ -101,6 +129,7 @@ export class ProxyChatSession {
     tools;
     toolConfig;
     generationConfig;
+    configuredThinkingLevel;
     latestUsageMetadata = undefined;
     constructor(modelName, systemInstruction, tools, generationConfig, toolConfig) {
         this.modelName = modelName;
@@ -108,6 +137,9 @@ export class ProxyChatSession {
         this.tools = tools;
         this.generationConfig = generationConfig;
         this.toolConfig = toolConfig;
+        if (generationConfig?.thinkingConfig?.thinkingLevel) {
+            this.configuredThinkingLevel = generationConfig.thinkingConfig.thinkingLevel;
+        }
     }
     getLatestUsageMetadata() {
         return this.latestUsageMetadata;
@@ -129,6 +161,28 @@ export class ProxyChatSession {
     getModel() {
         return this.modelName;
     }
+    setThinkingLevel(level) {
+        this.configuredThinkingLevel = level;
+        this.generationConfig = {
+            ...(this.generationConfig || {}),
+            thinkingConfig: {
+                ...(this.generationConfig?.thinkingConfig || {}),
+                thinkingLevel: level,
+            },
+        };
+    }
+    getThinkingLevel() {
+        return this.configuredThinkingLevel || this.generationConfig?.thinkingConfig?.thinkingLevel;
+    }
+    setGenerationConfig(generationConfig) {
+        this.generationConfig = generationConfig;
+        if (generationConfig?.thinkingConfig?.thinkingLevel) {
+            this.configuredThinkingLevel = generationConfig.thinkingConfig.thinkingLevel;
+        }
+    }
+    getGenerationConfig() {
+        return this.generationConfig;
+    }
     clearHistory() {
         this.history = [];
         this.fullHistory = [];
@@ -309,6 +363,12 @@ export class ProxyChatSession {
                             if (typeof respObj.result === 'string') {
                                 respObj.result = collapseHistoricalOutput(respObj.result, threshold);
                             }
+                            if (typeof respObj.stdout === 'string') {
+                                respObj.stdout = collapseHistoricalOutput(respObj.stdout, threshold);
+                            }
+                            if (typeof respObj.stderr === 'string') {
+                                respObj.stderr = collapseHistoricalOutput(respObj.stderr, threshold);
+                            }
                         }
                     }
                 }
@@ -461,11 +521,15 @@ export function createSharedChatSession() {
     let model = executionModelOverride || getGlobalActiveModel();
     if (model === 'auto')
         model = GEMINI_MODELS.FLASH; // Will be overridden per-turn in executeSingleTurn
+    const thinkingLevel = getModelThinkingLevel(model);
     return new ProxyChatSession(model, GENERAL_CHAT_INSTRUCTION.replace('{{MULTI_WORKSPACE_BLOCK}}', getMultiWorkspaceBlock()), [], {
         maxOutputTokens: MAX_OUTPUT_TOKENS,
         temperature: executionTempOverride !== null ? executionTempOverride : 1,
         topP: 0.95,
         topK: 40,
+        thinkingConfig: {
+            thinkingLevel,
+        },
     });
 }
 export function getGeneralChatConfig() {
@@ -480,12 +544,6 @@ export function getPlanExecutionConfig() {
         tools: [{ functionDeclarations: getToolDeclarations({ isExecutionAgent: true }) }],
     };
 }
-export function getPlanModeConfig() {
-    return {
-        systemInstruction: PLAN_MODE_INSTRUCTION.replace('{{MULTI_WORKSPACE_BLOCK}}', getMultiWorkspaceBlock()),
-        tools: [], // No tools allowed in plan mode
-    };
-}
 /**
  * Compresses a large string of text using Gemini Flash.
  * Used for shrinking context payloads to prevent OOM/choking.
@@ -515,11 +573,11 @@ export async function compressTextUsingFlashLite(text, instruction = "<directive
             }
             const creds = await loadCredentials();
             result = await proxyClient.generateViaBYOK(creds.geminiApiKey, model, contents, [], // no tools
-            undefined, instruction, { temperature: 0.2 }, undefined, abortSignal);
+            undefined, instruction, { temperature: 0.2, thinkingConfig: { thinkingLevel: 'MINIMAL' } }, undefined, abortSignal);
         }
         else {
             result = await proxyClient.generateFunctionCallViaProxy(idToken, model, contents, [], // no tools
-            undefined, instruction, { temperature: 0.2 }, // low temp for factual summary
+            undefined, instruction, { temperature: 0.2, thinkingConfig: { thinkingLevel: 'MINIMAL' } }, // low temp for factual summary
             undefined, abortSignal);
         }
         let textPart = '';
@@ -704,6 +762,9 @@ export function createContextAgentSession() {
         temperature: contextTempOverride !== null ? contextTempOverride : 1,
         topP: 0.95,
         topK: 40,
+        thinkingConfig: {
+            thinkingLevel: 'MEDIUM',
+        },
     }, {
         functionCallingConfig: {
             mode: 'ANY',
@@ -718,6 +779,9 @@ export function createIntentRouterSession() {
     return new ProxyChatSession(model, INTENT_ROUTER_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -745,6 +809,9 @@ export function createExecutionComplexitySession() {
     return new ProxyChatSession(model, EXECUTION_COMPLEXITY_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -767,6 +834,9 @@ export function createInvestigationComplexitySession() {
     return new ProxyChatSession(model, INVESTIGATION_COMPLEXITY_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MEDIUM',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -830,6 +900,9 @@ export function createInvestigationSemanticSession() {
     return new ProxyChatSession(model, INVESTIGATION_SEMANTIC_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -871,6 +944,9 @@ export function createWebSearchAgentSession() {
         temperature: 1,
         topP: 0.95,
         topK: 40,
+        thinkingConfig: {
+            thinkingLevel: 'MEDIUM',
+        },
     });
 }
 // ─── History Summarizer Agent Service ─────────────────────────────────
@@ -881,6 +957,9 @@ export function createHistorySummarizerSession() {
     return new ProxyChatSession(model, HISTORY_SUMMARIZER_SYSTEM_INSTRUCTION, [], {
         temperature: 0.2,
         maxOutputTokens: MAX_OUTPUT_TOKENS,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
     });
 }
 /**
@@ -980,6 +1059,9 @@ export function createUserProfileExtractorSession() {
     return new ProxyChatSession(model, USER_PROFILE_EXTRACTOR_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -1100,6 +1182,9 @@ export function createTrivialMessageClassifierSession() {
     return new ProxyChatSession(model, TRIVIAL_MESSAGE_CLASSIFIER_SYSTEM_INSTRUCTION, [], // no tools
     {
         temperature: 0,
+        thinkingConfig: {
+            thinkingLevel: 'MINIMAL',
+        },
         responseMimeType: 'application/json',
         responseSchema: {
             type: SchemaType.OBJECT,
@@ -1144,11 +1229,11 @@ export async function generateChatTitle(firstMessage, abortSignal) {
             }
             const creds = await loadCredentials();
             result = await proxyClient.generateViaBYOK(creds.geminiApiKey, model, contents, [], // no tools
-            undefined, instruction, { temperature: 0.2 }, undefined, abortSignal);
+            undefined, instruction, { temperature: 0.2, thinkingConfig: { thinkingLevel: 'MINIMAL' } }, undefined, abortSignal);
         }
         else {
             result = await proxyClient.generateFunctionCallViaProxy(idToken, model, contents, [], // no tools
-            undefined, instruction, { temperature: 0.2 }, undefined, abortSignal);
+            undefined, instruction, { temperature: 0.2, thinkingConfig: { thinkingLevel: 'MINIMAL' } }, undefined, abortSignal);
         }
         let title = '';
         if (result.parts) {

package/dist/services/contextAgent.js

@@ -10,13 +10,13 @@ import { getMetricCollector } from './metrics.js';
 import { estimateTokenCount, pruneTextToTokenBudget } from '../utils/historyPrompt.js';
 const metrics = getMetricCollector();
 /** Token budget constants for context optimization */
-const MAX_TREE_TOKENS = 12000;
-const MAX_CHAT_HISTORY_ROUTER_TOKENS = 3000;
-const MAX_CHAT_HISTORY_INVESTIGATION_TOKENS = 4500;
-const MAX_SINGLE_FILE_TOKENS = 25000;
-const MAX_TOTAL_FILE_TOKENS = 150000;
-const MAX_TOTAL_FILES = 30;
-const MAX_TOOL_OUTPUT_TOKENS = 10000;
+const MAX_TREE_TOKENS = 100000;
+const MAX_CHAT_HISTORY_ROUTER_TOKENS = 10000;
+const MAX_CHAT_HISTORY_INVESTIGATION_TOKENS = 50000;
+const MAX_SINGLE_FILE_TOKENS = 200000;
+const MAX_TOTAL_FILE_TOKENS = 200000;
+const MAX_TOTAL_FILES = 50;
+const MAX_TOOL_OUTPUT_TOKENS = 200000;
 /**
  * Bounds individual file content to ensure it does not exceed the single-file token budget.
  *
@@ -209,6 +209,7 @@ export async function routeIntent(userRequest, chatHistory = '', abortSignal) {
     }
     try {
         const session = createIntentRouterSession();
+        debugLog(`Intent Router evaluating intent [Model: ${session.getModel()}, Thinking: ${session.getThinkingLevel() || 'MINIMAL'}]`);
         let prompt = `User Request: "${userRequest}"`;
         if (chatHistory) {
             const prunedHistory = pruneTextToTokenBudget(chatHistory, MAX_CHAT_HISTORY_ROUTER_TOKENS, { fromStart: true });
@@ -242,6 +243,7 @@ export async function evaluateExecutionComplexity(userRequest, investigationSumm
     }
     try {
         const session = createExecutionComplexitySession();
+        debugLog(`Execution Complexity evaluating [Model: ${session.getModel()}, Thinking: ${session.getThinkingLevel() || 'MINIMAL'}]`);
         let prompt = `User Request: "${userRequest}"
 Investigation Summary: ${investigationSummary || 'None (0 files needed)'}
 Number of Relevant Files: ${numRelevantFiles}`;
@@ -418,6 +420,7 @@ export async function gatherContext(workspaceRoot, userRequest, chatHistory = ''
     }
     // ─── Single Context Agent (Unified Reconnaissance) ───────────
     const session = createContextAgentSession();
+    debugLog(`[Context Agent] Starting reconnaissance [Model: ${session.getModel()}, Thinking: ${session.getThinkingLevel() || 'MEDIUM'}]`);
     const relevantFiles = new Map();
     let summary = 'No relevant context found.';
     let isInvestigationFinished = false;
@@ -430,6 +433,26 @@ export async function gatherContext(workspaceRoot, userRequest, chatHistory = ''
         const prunedHistory = pruneTextToTokenBudget(chatHistory, MAX_CHAT_HISTORY_INVESTIGATION_TOKENS, { fromStart: true });
         currentMessage = `Previous Conversation Context:\n${prunedHistory}\n\n` + currentMessage;
     }
+    // Parse explicit context mentions from user request to prioritize referenced files
+    try {
+        const { mentionEngine } = await import('./mentionEngine.js');
+        const parsedMentions = mentionEngine.parseMentions(userRequest, workspaceRoot);
+        const explicitMentionedFiles = parsedMentions
+            .filter((m) => m.type === 'file' && m.valid)
+            .map((m) => m.target);
+        if (parsedMentions.length > 0) {
+            const validMentions = parsedMentions.filter((m) => m.valid);
+            if (validMentions.length > 0 && onProgress) {
+                const mentionLabels = validMentions.map((m) => m.raw).slice(0, 4).join(', ');
+                const overflow = validMentions.length > 4 ? ` (+${validMentions.length - 4} more)` : '';
+                onProgress(`Found explicit context: ${mentionLabels}${overflow}`);
+            }
+        }
+        if (explicitMentionedFiles.length > 0) {
+            currentMessage += `\n\nExplicit Mentioned Files from User Request:\n${explicitMentionedFiles.map((f) => `- ${f}`).join('\n')}`;
+        }
+    }
+    catch { }
     currentMessage += `\n\nStart investigating to find relevant files.`;
     const MAX_TURNS = Infinity;
     let textRetryCount = 0;
@@ -723,7 +746,9 @@ export async function gatherContext(workspaceRoot, userRequest, chatHistory = ''
                 };
             }
             else if (call.name === 'run_analysis_script') {
-                const analysisResult = await runEphemeralScript(workspaceRoot, args.language, args.code, { abortSignal });
+                const analysisResult = await runEphemeralScript(workspaceRoot, args.language, args.code, {
+                    abortSignal,
+                });
                 const rawOutput = analysisResult.exitCode === 0
                     ? analysisResult.stdout || '(script produced no output)'
                     : `Script failed (exit ${analysisResult.exitCode}):\n${analysisResult.stderr}`;

package/dist/services/investigationComplexity.js

@@ -235,7 +235,7 @@ Approximate File Count: ${approximateFileCount}`;
         if (chatHistory.trim()) {
             prompt += `\n\nRecent Conversation:\n${chatHistory.trim()}`;
         }
-        debugLog(`[InvestigationComplexity] Evaluating prompt complexity (autoFocus: ${activeSubPath || 'none'}, override: ${explicitOverride || 'none'})`);
+        debugLog(`[InvestigationComplexity] Evaluating prompt complexity (autoFocus: ${activeSubPath || 'none'}, override: ${explicitOverride || 'none'}) [Model: ${session.getModel()}, Thinking: ${session.getThinkingLevel() || 'MEDIUM'}]`);
         const response = await session.sendMessage(prompt, undefined, abortSignal);
         const rawText = typeof response?.response?.text === 'function'
             ? response.response.text()