yuangs 6.1.0 → 6.2.0

package/dist/agent/AgentRuntime.js

@@ -56,458 +56,521 @@ const renderer_1 = require("../utils/renderer");
 class AgentRuntime {
     context;
     executionId;
+    maxTurns = 10;
+    readOnlyTools = [
+        'read_file', 'list_files', 'read_file_lines', 'read_file_lines_from_end',
+        'file_info', 'git_status', 'git_log', 'git_diff', 'list_directory_tree',
+        'search_in_files', 'search_symbol', 'continue_reading', 'analyze_dependencies'
+    ];
     constructor(initialContext) {
         this.context = new smartContextManager_1.SmartContextManager(initialContext);
         this.executionId = (0, crypto_1.randomUUID)();
     }
     async run(userInput, mode = "chat", onChunk, model, renderer) {
         let turnCount = 0;
-        const maxTurns = 10;
         let lastError;
-        let shouldComplete = false; // 标记是否应该完成
-        let lastToolCall = null; // 跟踪上次工具调用及重复次数
-        let writeModeFileRead = null; // 跟踪 Write Mode 下已读取的文件
-        let writeModeFileContent = null; // 缓存 Write Mode 下已读取的文件内容
-        // 构建初始动态上下文
-        const initialDynamicContext = await (0, dynamicPrompt_1.buildDynamicContext)();
+        let lastToolCall = null;
+        let writeModeState = null;
         if (userInput) {
             this.context.addMessage("user", userInput);
         }
-        while (turnCount < maxTurns) {
-            // 检查是否应该完成（上一轮只读工具成功）
-            if (shouldComplete) {
-                console.log(chalk_1.default.gray('[Task Complete] 任务已完成'));
-                break;
-            }
+        while (turnCount < this.maxTurns) {
             const currentTurn = ++turnCount;
             if (currentTurn > 1) {
                 console.log(chalk_1.default.blue(`\n--- Turn ${currentTurn} ---`));
             }
-            // Use smart context manager to get relevance-ranked context
-            const enhancedContext = await this.context.getEnhancedContext({
-                query: userInput,
-                minRelevance: 0.3,
-                maxTokens: 8000,
-                enableSmartSummary: true
-            });
-            const messages = [];
-            // Add context overview as system message
-            if (enhancedContext.summary) {
-                messages.push({
-                    role: 'system',
-                    content: enhancedContext.summary
-                });
+            const { agentRenderer, agentOnChunk } = this.prepareRenderer(renderer, onChunk);
+            const { prompt: enhancedPrompt, userInput: query } = await this.buildPrompt(userInput, lastError);
+            // === Step 1: LLM Thinking ===
+            const thought = await this.callLLM(enhancedPrompt, query, mode, agentOnChunk, model, agentRenderer);
+            if (!thought)
+                break; // error or empty
+            // === Step 2: Build Action ===
+            const action = this.buildAction(thought);
+            if (action.reasoning && !onChunk) {
+                console.log(chalk_1.default.gray(`\n🤔 Reasoning: ${action.reasoning}`));
             }
-            // Add ranked context files
-            for (const item of enhancedContext.rankedItems) {
-                messages.push({
-                    role: 'user',
-                    content: `@${item.path} (相关度: ${(item.relevance * 100).toFixed(0)}%)\n${item.summary || item.content || ''}`
-                });
+            if (thought.usedRouter) {
+                console.log(chalk_1.default.gray(`[Router] 🤖 Model: ${thought.modelName}`));
             }
-            // Add user input
-            if (userInput) {
-                messages.push({
-                    role: 'user',
-                    content: userInput
-                });
+            // === Step 3: Handle answer type ===
+            if (thought.isDone || action.type === "answer") {
+                // Command 模式下 AI 返回 answer：引导其使用工具
+                if (mode === "command") {
+                    const content = action.payload.content || action.payload.text || '';
+                    this.context.addMessage('system', `在命令执行模式下，请直接使用工具执行命令，不要回复对话。请使用 list_directory_tree 查看文件列表，然后用 shell_cmd 执行命令来查找最大文件。任务: "${userInput}"`);
+                    continue; // 不 break，让 AI 重试使用工具
+                }
+                await this.handleAnswer(action, thought, userInput, mode, renderer, agentRenderer);
+                break;
             }
-            // Build dynamic context (if previous step had error)
-            const dynamicContext = await (0, dynamicPrompt_1.buildDynamicContext)(lastError);
-            // 构建基础prompt（包括治理策略）
-            const basePrompt = governance_1.GovernanceService.getPolicyManual();
-            // 注入动态上下文
-            const enhancedPrompt = (0, dynamicPrompt_1.injectDynamicContext)(basePrompt, dynamicContext);
-            // Create renderer if not provided but onChunk is available
-            let agentRenderer = renderer;
-            let agentOnChunk = onChunk;
-            if (!agentRenderer && agentOnChunk) {
-                agentRenderer = new renderer_1.StreamMarkdownRenderer(chalk_1.default.bgHex('#3b82f6').white.bold(' 🤖 Agent ') + ' ', undefined, { autoFinish: false });
-                agentOnChunk = agentRenderer.startChunking();
+            // === Step 4: Causal ACK Check ===
+            if (!this.verifyAckCausality(thought))
+                continue;
+            // === Step 5: Governance ===
+            if (!await this.passGovernance(action))
+                continue;
+            // === Step 6: Record KG Edge ===
+            await this.recordKnowledgeGraphEdge(thought, action);
+            // === Step 7: Pre-execution Checks ===
+            const cachedResult = await this.checkExecutionBlock(action, writeModeState);
+            if (cachedResult === 'blocked')
+                continue;
+            // === Step 8: Execute ===
+            const result = cachedResult && typeof cachedResult === 'object'
+                ? cachedResult
+                : await this.executeAction(action, writeModeState);
+            if (result.success) {
+                lastError = undefined;
+                lastToolCall = this.handleSuccessfulExecution(result, action, lastToolCall, userInput, writeModeState, mode, thought, agentRenderer);
+                if (lastToolCall === null)
+                    break; // null signals break (stabilization or duplicate)
+            }
+            else {
+                lastError = await this.handleFailedExecution(result, action, mode, thought, userInput);
+            }
+        }
+        if (turnCount >= this.maxTurns) {
+            console.log(chalk_1.default.red(`\n⚠️ Max turns (${this.maxTurns}) reached.`));
+        }
+    }
+    // === Private Helper Methods ===
+    prepareRenderer(renderer, onChunk) {
+        let agentRenderer = renderer;
+        let agentOnChunk = onChunk;
+        if (!agentRenderer && agentOnChunk) {
+            agentRenderer = new renderer_1.StreamMarkdownRenderer(chalk_1.default.bgHex('#3b82f6').white.bold(' 🤖 Agent ') + ' ', undefined, { autoFinish: false });
+            agentOnChunk = agentRenderer.startChunking();
+        }
+        return { agentRenderer, agentOnChunk };
+    }
+    async buildPrompt(userInput, lastError) {
+        const dynamicContext = await (0, dynamicPrompt_1.buildDynamicContext)(lastError);
+        const basePrompt = governance_1.GovernanceService.getPolicyManual();
+        return {
+            prompt: (0, dynamicPrompt_1.injectDynamicContext)(basePrompt, dynamicContext),
+            userInput
+        };
+    }
+    async callLLM(enhancedPrompt, userInput, mode, onChunk, model, renderer) {
+        const messages = [];
+        const enhancedContext = await this.context.getEnhancedContext({
+            query: userInput || enhancedPrompt,
+            minRelevance: 0.3,
+            maxTokens: 8000,
+            enableSmartSummary: true
+        });
+        if (enhancedContext.summary) {
+            messages.push({ role: 'system', content: enhancedContext.summary });
+        }
+        for (const item of enhancedContext.rankedItems) {
+            messages.push({
+                role: 'user',
+                content: `@${item.path} (相关度: ${(item.relevance * 100).toFixed(0)}%)\n${item.summary || item.content || ''}`
+            });
+        }
+        // 关键：确保用户输入被包含为最后一条消息
+        if (userInput) {
+            messages.push({ role: 'user', content: userInput });
+        }
+        else {
+            messages.push({ role: 'user', content: enhancedPrompt });
+        }
+        try {
+            const thought = await llmAdapter_1.LLMAdapter.think(messages, mode, onChunk, model, enhancedPrompt, this.context);
+            if (!thought.raw || thought.raw.trim() === '') {
+                console.log(chalk_1.default.red('\n⚠️ AI 返回了空响应，请检查网络连接或模型配置。'));
+                return null;
             }
-            let thought;
+            return thought;
+        }
+        catch (error) {
+            this.handleLLMError(error);
+            return null;
+        }
+    }
+    handleLLMError(error) {
+        let errorMessage = '未知内部错误';
+        let statusCode = 0;
+        if (error instanceof llm_1.AIError) {
+            errorMessage = error.message;
+            statusCode = error.statusCode;
+        }
+        else if (error instanceof Error) {
+            errorMessage = error.message;
+            statusCode = error.statusCode || 0;
+        }
+        else if (typeof error === 'string') {
+            errorMessage = error;
+        }
+        const statusInfo = statusCode ? ` (状态码: ${statusCode})` : '';
+        console.log(chalk_1.default.red(`\n❌ AI 思考过程发生错误: ${errorMessage}${statusInfo}`));
+        const isTransient = statusCode === 429 || statusCode >= 500
+            || errorMessage.includes('timeout') || errorMessage.includes('network') || errorMessage.includes('ETIMEDOUT');
+        if (isTransient) {
+            console.log(chalk_1.default.yellow('⚠️ 检测到瞬时错误，自动跳过此轮'));
+            this.context.addMessage("system", `AI 调用失败${statusInfo}，请稍后重试`);
+            return;
+        }
+        this.context.addMessage("system", `思考过程中发生错误${statusInfo}: ${errorMessage}`);
+        if (statusCode === 401 || statusCode === 403 || errorMessage.includes('401') || errorMessage.includes('403')) {
+            console.log(chalk_1.default.yellow('💡 检测到权限或授权错误，请检查 API 配置。'));
+        }
+    }
+    buildAction(thought) {
+        return {
+            id: (0, crypto_1.randomUUID)(),
+            type: thought.type || "answer",
+            payload: thought.payload || { text: thought.raw },
+            riskLevel: "low",
+            reasoning: thought.reasoning || "",
+        };
+    }
+    async handleAnswer(action, thought, userInput, mode, externalRenderer, agentRenderer) {
+        const result = await executor_1.ToolExecutor.execute(action);
+        if (!externalRenderer && agentRenderer) {
+            for (let i = 0; i < result.output.length; i += 10) {
+                agentRenderer.onChunk(result.output.slice(i, i + 10));
+            }
+            agentRenderer.finish();
+        }
+        else if (!externalRenderer) {
+            const rendered = marked.parse(result.output);
+            console.log(chalk_1.default.green(`\n🤖 AI：\n`) + rendered);
+        }
+        this.context.addMessage("assistant", result.output);
+        await this.learnFromExecution(userInput, mode, thought);
+    }
+    verifyAckCausality(thought) {
+        const lastObs = this.context.getLastAckableObservation();
+        const ackText = thought.parsedPlan?.acknowledged_observation;
+        if (lastObs && ackText && ackText !== 'NONE') {
+            if (lastObs.content.trim() !== ackText.trim()) {
+                console.log(chalk_1.default.red(`[CAUSAL BREAK] ❌ ACK mismatch!`));
+                console.log(chalk_1.default.red(`  Expected: ${lastObs.content.trim().substring(0, 100)}...`));
+                console.log(chalk_1.default.red(`  Received: ${ackText.trim().substring(0, 100)}...`));
+                this.context.addMessage("system", `CAUSAL BREAK: ACK does not match physical Observation. Cannot proceed without acknowledging reality.`);
+                return false;
+            }
+            console.log(chalk_1.default.green(`[CAUSAL LOCK] ✅ ACK verified`));
+        }
+        return true;
+    }
+    async passGovernance(action) {
+        const preCheck = (0, core_1.evaluateProposal)(action, governance_1.GovernanceService.getRules(), governance_1.GovernanceService.getLedgerSnapshot());
+        if (preCheck.effect === "deny") {
+            console.log(chalk_1.default.red(`[PRE-FLIGHT] 🛡️ Policy Blocked: ${preCheck.reason}`));
+            this.context.addMessage("system", `POLICY DENIED: ${preCheck.reason}. Find a different way.`);
+            return false;
+        }
+        const decision = await governance_1.GovernanceService.adjudicate(action);
+        if (decision.status === "rejected") {
+            console.log(chalk_1.default.red(`[GOVERNANCE] ❌ Rejected: ${decision.reason}`));
+            this.context.addMessage("system", `Rejected by Governance: ${decision.reason}`);
+            return false;
+        }
+        return true;
+    }
+    async recordKnowledgeGraphEdge(thought, action) {
+        const lastObs = this.context.getLastAckableObservation();
+        const ackText = thought.parsedPlan?.acknowledged_observation;
+        if (lastObs && lastObs.metadata?.obsId && ackText && ackText !== 'NONE') {
             try {
-                thought = await llmAdapter_1.LLMAdapter.think(messages, mode, agentOnChunk, model, enhancedPrompt, this.context);
-                if (!thought.raw || thought.raw.trim() === '') {
-                    console.log(chalk_1.default.red('\n⚠️ AI 返回了空响应，请检查网络连接或模型配置。'));
-                    break;
-                }
+                const { recordEdge } = await Promise.resolve().then(() => __importStar(require('../engine/agent/knowledgeGraph')));
+                recordEdge({
+                    from: lastObs.metadata.obsId,
+                    to: action.id,
+                    type: 'ACKNOWLEDGED_BY',
+                    metadata: { verified: true, timestamp: Date.now() }
+                });
+                console.log(chalk_1.default.gray(`[KG] ⚓ Causal edge recorded`));
             }
             catch (error) {
-                let errorMessage = '未知内部错误';
-                let statusCode = 0;
-                if (error instanceof llm_1.AIError) {
-                    errorMessage = error.message;
-                    statusCode = error.statusCode;
-                }
-                else if (error instanceof Error) {
-                    errorMessage = error.message;
-                    statusCode = error.statusCode || 0;
-                }
-                else if (typeof error === 'string') {
-                    errorMessage = error;
-                }
-                const statusInfo = statusCode ? ` (状态码: ${statusCode})` : '';
-                console.log(chalk_1.default.red(`\n❌ AI 思考过程发生错误: ${errorMessage}${statusInfo}`));
-                this.context.addMessage("system", `思考过程中发生错误${statusInfo}: ${errorMessage}`);
-                if (statusCode === 401 || statusCode === 403 || errorMessage.includes('401') || errorMessage.includes('403')) {
-                    console.log(chalk_1.default.yellow('💡 检测到权限或授权错误，请检查 API 配置。'));
-                    break;
-                }
-                if (statusCode === 429) {
-                    console.log(chalk_1.default.yellow('💡 API 调用频率过高，请稍后再试。'));
-                }
-                break;
+                console.warn(chalk_1.default.yellow(`[KG] Warning: Failed to record causal edge: ${error.message}`));
             }
-            try {
-                const action = {
-                    id: (0, crypto_1.randomUUID)(),
-                    type: thought.type || "answer",
-                    payload: thought.payload || { text: thought.raw },
-                    riskLevel: "low",
-                    reasoning: thought.reasoning || "",
-                };
-                if (action.reasoning && !onChunk) {
-                    console.log(chalk_1.default.gray(`\n🤔 Reasoning: ${action.reasoning}`));
-                }
-                if (thought.usedRouter) {
-                    console.log(chalk_1.default.gray(`[Router] 🤖 Model: ${thought.modelName}`));
-                }
-                // 如果 LLM 认为已经完成或者当前的动作就是回答
-                if (thought.isDone || action.type === "answer") {
-                    const result = await executor_1.ToolExecutor.execute(action);
-                    // ... rest of the logic
-                    // 如果没有 renderer，使用内部创建的
-                    if (!renderer && agentRenderer) {
-                        // Stream final answer through internal renderer
-                        for (let i = 0; i < result.output.length; i += 10) {
-                            const chunk = result.output.slice(i, i + 10);
-                            agentRenderer.onChunk(chunk);
-                        }
-                        agentRenderer.finish();
-                    }
-                    else if (!renderer) {
-                        // Fallback to marked if no renderer
-                        const rendered = marked.parse(result.output);
-                        console.log(chalk_1.default.green(`\n🤖 AI：\n`) + rendered);
-                    }
-                    // 如果外部传入了 renderer，由外部调用 finish()
-                    this.context.addMessage("assistant", result.output);
-                    // Learn from successful chat
-                    try {
-                        const { createExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionRecord')));
-                        const { inferCapabilityRequirement } = await Promise.resolve().then(() => __importStar(require('../core/capabilityInference')));
-                        const { saveExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
-                        const record = createExecutionRecord('agent-chat', { required: [], preferred: [] }, {
-                            aiProxyUrl: { value: '', source: 'built-in' },
-                            defaultModel: { value: '', source: 'built-in' },
-                            accountType: { value: 'free', source: 'built-in' }
-                        }, { selected: null, candidates: [], fallbackOccurred: false }, { success: true }, undefined, userInput, 'chat');
-                        record.llmResult = { plan: thought.parsedPlan };
-                        record.input = { rawInput: userInput };
-                        const savedRecordId = saveExecutionRecord(record);
-                        const { loadExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
-                        const savedRecord = loadExecutionRecord(savedRecordId);
-                        if (savedRecord) {
-                            const { learnSkillFromRecord } = await Promise.resolve().then(() => __importStar(require('./skills')));
-                            learnSkillFromRecord(savedRecord, true);
-                        }
-                    }
-                    catch (error) {
-                        console.warn(chalk_1.default.yellow(`[Skill Learning] Failed: ${error}`));
-                    }
-                    break;
-                }
-                // === 强制 ACK 校验（Causal Lock） ===
-                const lastObs = this.context.getLastAckableObservation();
-                const ackText = thought.parsedPlan?.acknowledged_observation;
-                if (lastObs && ackText && ackText !== 'NONE') {
-                    const actualContent = lastObs.content.trim();
-                    const ackedContent = ackText.trim();
-                    if (actualContent !== ackedContent) {
-                        console.log(chalk_1.default.red(`[CAUSAL BREAK] ❌ ACK mismatch!`));
-                        console.log(chalk_1.default.red(`  Expected: ${actualContent.substring(0, 100)}...`));
-                        console.log(chalk_1.default.red(`  Received: ${ackedContent.substring(0, 100)}...`));
-                        this.context.addMessage("system", `CAUSAL BREAK: ACK does not match physical Observation. Cannot proceed without acknowledging reality.`);
-                        continue;
+        }
+    }
+    async checkExecutionBlock(action, writeModeState) {
+        if (action.type !== 'tool_call')
+            return null;
+        const toolName = action.payload.tool_name;
+        // Write Mode cache
+        if (writeModeState && toolName === 'read_file') {
+            const filePath = action.payload.parameters.path;
+            if (filePath === writeModeState.filePath && writeModeState.content) {
+                console.log(chalk_1.default.yellow(`[CACHED] 📋 文件 ${filePath} 内容已缓存，直接返回`));
+                return { success: true, output: writeModeState.content, artifacts: [filePath] };
+            }
+        }
+        const blockCheck = errorTracker_1.ErrorTracker.shouldBlockExecution(toolName, action.payload.parameters);
+        if (blockCheck.blocked) {
+            console.log(chalk_1.default.red(`[BLOCKED] 🚫 ${blockCheck.reason}`));
+            if (blockCheck.existingError) {
+                console.log(chalk_1.default.yellow(`[Error History] 此错误已发生 ${blockCheck.existingError.attemptCount} 次`));
+                console.log(chalk_1.default.gray(`上次错误: ${blockCheck.existingError.errorMessage}`));
+            }
+            this.context.addMessage("system", `BLOCKED: ${blockCheck.reason}. 建议尝试不同的方法。`);
+            return 'blocked';
+        }
+        return null;
+    }
+    async executeAction(action, _writeModeState) {
+        console.log(chalk_1.default.yellow(`[EXECUTING] ⚙️ ${action.type}...`));
+        return await executor_1.ToolExecutor.execute(action);
+    }
+    handleSuccessfulExecution(result, action, lastToolCall, userInput, writeModeState, mode, thought, agentRenderer) {
+        const actualToolName = action.payload?.tool_name || action.type;
+        this.context.addToolResult(actualToolName, result.output);
+        const preview = result.output.length > 300 ? result.output.substring(0, 300) + '...' : result.output;
+        console.log(chalk_1.default.green(`[SUCCESS] Result:\n${preview}`));
+        // Stabilization detection: 检查输出是否稳定（排除错误输出）
+        const normalizedOutput = result.output?.trim()?.split('\n').filter(Boolean)[0] || '';
+        const isErrorMessage = /invalid option|unknown|error|failed|no such|cannot|usage:|not found/i.test(normalizedOutput);
+        if (!isErrorMessage) {
+            // 方案3: 语义完成检测 — 输出已经是直接答案（适用于首轮查询，必须在 prevOutput 检测之前）
+            // 只有当完整输出都很短时才触发，避免在截断的多行输出上误触发
+            if (result.output.length < 100 && this.isSemanticComplete(normalizedOutput, userInput)) {
+                console.log(chalk_1.default.yellow('[Semantic Complete] 输出已经是直接答案，自动完成'));
+                console.log(chalk_1.default.cyan(`\n✓ 结果: ${normalizedOutput}\n`));
+                this.context.addMessage("assistant", result.output);
+                agentRenderer && (() => { agentRenderer.buffer = ''; agentRenderer.quietMode = true; agentRenderer.finish(); })();
+                return null; // signal break
+            }
+            // 方案4: 截断输出检测 — 结果被截断但已包含数据，引导 AI 完成
+            if (result.output.includes('[⚠️ OUTPUT TRUNCATED]') && normalizedOutput.length > 0) {
+                this.context.addMessage('system', `输出已截断，但已包含足够的数据。请直接使用 answer 类型向用户呈现你看到的结果，不要再运行更多命令。`);
+            }
+            // 方案1: 连续两次输出完全相同
+            const prevOutput = lastToolCall?.lastOutput;
+            const isOutputStable = prevOutput && normalizedOutput && prevOutput === normalizedOutput;
+            // 方案2: 最近2次输出包含相同的文件/路径名（即使命令不同）
+            const extractPath = (line) => {
+                const parts = line.trim().split(/\s+/);
+                return parts.length > 1 ? parts[parts.length - 1] : '';
+            };
+            const currentPath = extractPath(normalizedOutput);
+            const prevPath = prevOutput ? extractPath(prevOutput) : '';
+            const isOutputStableByName = currentPath && prevPath && currentPath === prevPath;
+            if (isOutputStable || isOutputStableByName) {
+                console.log(chalk_1.default.yellow('[Stabilization] 输出结果已稳定，自动完成'));
+                console.log(chalk_1.default.cyan(`\n✓ 结果: ${normalizedOutput}\n`));
+                this.context.addMessage("assistant", result.output);
+                agentRenderer && (() => { agentRenderer.buffer = ''; agentRenderer.quietMode = true; agentRenderer.finish(); })();
+                return null; // signal break
+            }
+        }
+        // Duplicate detection: 区分 tool_call 和 shell_cmd
+        const actualTool = action.payload?.tool_name || action.type;
+        const params = action.payload?.tool_name === 'shell_cmd' || action.type === 'shell_cmd'
+            ? { command: (action.payload?.command || action.payload?.parameters?.command || '') }
+            : (action.payload?.parameters || action.payload);
+        const currentToolCall = { tool: actualTool, params };
+        const isDuplicate = lastToolCall &&
+            lastToolCall.tool === currentToolCall.tool &&
+            JSON.stringify(lastToolCall.params) === JSON.stringify(currentToolCall.params);
+        // Build output history for stabilization detection
+        const outputHistory = [...(lastToolCall?.outputHistory || []), normalizedOutput].slice(-3);
+        if (isDuplicate && lastToolCall) {
+            const count = lastToolCall.count + 1;
+            // 如果是错误输出，提示 AI 换方案
+            if (isErrorMessage) {
+                if (count >= 2) {
+                    const failedCmd = actualTool === 'shell_cmd'
+                        ? (lastToolCall.params?.command || '')
+                        : '';
+                    console.log(chalk_1.default.yellow('[Duplicate Error] 相同命令重复失败 2 次，自动终止该路径'));
+                    let errorMsg = `命令执行失败，且已重复尝试 2 次。请换一种完全不同的方法来完成用户任务: "${userInput}"。`;
+                    if (failedCmd) {
+                        errorMsg += `\n⚠️ 命令 "${failedCmd}" 在当前系统不可用，已被列入本次对话黑名单，不要再使用此命令。`;
                     }
-                    console.log(chalk_1.default.green(`[CAUSAL LOCK] ✅ ACK verified`));
+                    errorMsg += `\n注意：当前系统可能是 macOS（BSD 工具链），不支持 Linux 特定的命令选项。请使用 macOS 兼容的命令。`;
+                    this.context.addMessage('system', errorMsg);
+                    // 重置重复计数让新命令不被当作重复
+                    lastToolCall = { tool: 'reset', params: {}, count: 0, lastOutput: '', outputHistory: [] };
+                    return lastToolCall;
                 }
-                // === 预检 (Pre-flight) ===
-                const preCheck = (0, core_1.evaluateProposal)(action, governance_1.GovernanceService.getRules(), governance_1.GovernanceService.getLedgerSnapshot());
-                if (preCheck.effect === "deny") {
-                    console.log(chalk_1.default.red(`[PRE-FLIGHT] 🛡️ Policy Blocked: ${preCheck.reason}`));
-                    this.context.addMessage("system", `POLICY DENIED: ${preCheck.reason}. Find a different way.`);
-                    continue;
-                }
-                // === 正式治理 (WASM + 人工/自动) ===
-                const decision = await governance_1.GovernanceService.adjudicate(action);
-                if (decision.status === "rejected") {
-                    console.log(chalk_1.default.red(`[GOVERNANCE] ❌ Rejected: ${decision.reason}`));
-                    this.context.addMessage("system", `Rejected by Governance: ${decision.reason}`);
-                    continue;
+                console.log(chalk_1.default.gray(`[Repeat Error] 重复失败 (${count + 1}/2)，请换方案...`));
+            }
+            else if (count >= 2) {
+                console.log(chalk_1.default.yellow('[Duplicate Detection] 达到重复限制，强制完成'));
+                console.log(chalk_1.default.cyan(`\n✓ ${action.payload?.tool_name || action.type} 已完成\n`));
+                agentRenderer && (() => { agentRenderer.buffer = ''; agentRenderer.quietMode = true; agentRenderer.finish(); })();
+                return null; // signal break
+            }
+            else {
+                console.log(chalk_1.default.gray(`[Repeat Detection] 重复调用 (${count + 1}/2)，继续...`));
+            }
+            lastToolCall.count = count;
+            lastToolCall.lastOutput = normalizedOutput;
+            lastToolCall.outputHistory = outputHistory;
+            if (!isErrorMessage) {
+                // 告诉 AI 结果已重复，应该返回答案而不是继续调用工具
+                this.context.addMessage('system', `同样的操作已经重复执行了 ${count} 次，结果没有变化。请停止继续调用工具，使用 answer 类型返回最终结果给用户。输出: "${result.output.slice(0, 200)}"`);
+            }
+        }
+        else {
+            lastToolCall = { ...currentToolCall, count: 0, lastOutput: normalizedOutput, outputHistory };
+        }
+        // Auto-complete logic for specific tool types
+        if (action.type === 'tool_call') {
+            const toolName = action.payload.tool_name;
+            if (toolName === 'write_file') {
+                console.log(chalk_1.default.gray('[Auto-Complete] 文件写入成功，自动完成任务'));
+                console.log(chalk_1.default.green(`✓ 已创建文件: ${action.payload.parameters.path}\n`));
+                this.context.addMessage("assistant", `已成功创建文件 ${action.payload.parameters.path}`);
+                agentRenderer && (() => { agentRenderer.buffer = ''; agentRenderer.quietMode = true; agentRenderer.finish(); })();
+                result.shouldBreak = true;
+                return lastToolCall;
+            }
+            if (this.readOnlyTools.includes(toolName)) {
+                const requiresAnalysis = /^(.*?)(帮我|请)?(分析|解释|说明|总结)|\b(review|explain)\s+(this|the|it)\b/i.test(userInput);
+                const requiresWrite = /替换|replace|修改|modify|添加|append|插入|insert|删除|delete|移除|remove|更新|update|改成|改成|改为/i.test(userInput);
+                // Command 模式下不自动完成，让 AI 自行决定后续操作
+                if (mode === 'command') {
+                    console.log(chalk_1.default.gray('[Command Mode] 只读工具执行成功，等待 AI 决定下一步'));
+                    return lastToolCall;
                 }
-                // === 记录因果边到 KG ===
-                if (lastObs && lastObs.metadata?.obsId && ackText && ackText !== 'NONE') {
-                    try {
-                        const { recordEdge } = await Promise.resolve().then(() => __importStar(require('../engine/agent/knowledgeGraph')));
-                        recordEdge({
-                            from: lastObs.metadata.obsId,
-                            to: action.id,
-                            type: 'ACKNOWLEDGED_BY',
-                            metadata: {
-                                verified: true,
-                                timestamp: Date.now()
-                            }
-                        });
-                        console.log(chalk_1.default.gray(`[KG] ⚓ Causal edge recorded`));
-                    }
-                    catch (error) {
-                        console.warn(chalk_1.default.yellow(`[KG] Warning: Failed to record causal edge: ${error.message}`));
-                    }
+                if (!requiresAnalysis && !requiresWrite) {
+                    console.log(chalk_1.default.gray('[Auto-Complete] 只读工具执行成功，自动完成任务'));
+                    console.log(chalk_1.default.cyan(`\n📄 ${toolName} 结果：\n`));
+                    console.log(result.output);
+                    this.context.addMessage("assistant", `已成功执行 ${toolName}，结果：\n${result.output}`);
+                    agentRenderer && (() => { agentRenderer.buffer = ''; agentRenderer.quietMode = true; agentRenderer.finish(); })();
+                    result.shouldBreak = true;
+                    return lastToolCall;
                 }
-                // === 执行前错误检查 ===
-                // 检查是否应该阻止执行（防止重复错误）
-                let cachedResult = null;
-                if (action.type === 'tool_call') {
-                    const toolName = action.payload.tool_name;
-                    // Write Mode 阻止重复 read_file 调用，但返回缓存的内容
-                    if (writeModeFileRead && toolName === 'read_file') {
-                        const filePath = action.payload.parameters.path;
-                        if (filePath === writeModeFileRead && writeModeFileContent) {
-                            console.log(chalk_1.default.yellow(`[CACHED] 📋 文件 ${filePath} 内容已缓存，直接返回（避免重复读取）`));
-                            // 使用缓存的结果，跳过实际执行
-                            cachedResult = {
-                                success: true,
-                                output: writeModeFileContent,
-                                artifacts: [filePath]
-                            };
-                        }
+                else if (requiresWrite) {
+                    console.log(chalk_1.default.gray('[Write Mode] 文件已读取，继续执行写入操作...'));
+                    if (writeModeState) {
+                        writeModeState.filePath = action.payload.parameters.path;
+                        writeModeState.content = result.output;
                     }
-                    if (!cachedResult) {
-                        const blockCheck = errorTracker_1.ErrorTracker.shouldBlockExecution(toolName, action.payload.parameters);
-                        if (blockCheck.blocked) {
-                            console.log(chalk_1.default.red(`[BLOCKED] 🚫 ${blockCheck.reason}`));
-                            // 显示错误历史
-                            if (blockCheck.existingError) {
-                                console.log(chalk_1.default.yellow(`[Error History] 此错误已发生 ${blockCheck.existingError.attemptCount} 次`));
-                                console.log(chalk_1.default.gray(`上次错误: ${blockCheck.existingError.errorMessage}`));
-                            }
-                            this.context.addMessage("system", `BLOCKED: ${blockCheck.reason}. 建议尝试不同的方法。`);
-                            continue;
-                        }
-                    }
-                }
-                // === 执行 ===
-                let result;
-                if (cachedResult) {
-                    result = cachedResult;
-                    // 添加到上下文
-                    this.context.addToolResult('read_file', result.output);
-                    const preview = result.output.length > 300
-                        ? result.output.substring(0, 300) + '...'
-                        : result.output;
-                    console.log(chalk_1.default.green(`[SUCCESS] Result (cached):\n${preview}`));
+                    this.context.addMessage('system', `文件 ${action.payload.parameters.path} 已成功读取。请根据用户要求修改内容后，调用 write_file 工具写入文件。不要重复调用 read_file！`);
                 }
                 else {
-                    console.log(chalk_1.default.yellow(`[EXECUTING] ⚙️ ${action.type}...`));
-                    result = await executor_1.ToolExecutor.execute(action);
+                    console.log(chalk_1.default.gray('[Analysis Mode] 文件已读取，继续分析...'));
                 }
-                if (result.success) {
-                    // 成功时清除错误状态
-                    lastError = undefined;
-                    // 使用实际工具名称而不是 action.type，这样 LLM 知道调用了哪个工具
-                    const actualToolName = action.payload?.tool_name || action.type;
-                    this.context.addToolResult(actualToolName, result.output);
-                    const preview = result.output.length > 300
-                        ? result.output.substring(0, 300) + '...'
-                        : result.output;
-                    console.log(chalk_1.default.green(`[SUCCESS] Result:\n${preview}`));
-                    // 通用重复检测：所有工具
-                    const currentToolCall = { tool: action.payload.tool_name || action.type, params: action.payload.parameters || action.payload };
-                    // 检查是否重复，并允许一定次数的重复（给 AI 时间完成多步骤任务）
-                    let duplicateCount = 0;
-                    const isDuplicate = lastToolCall &&
-                        lastToolCall.tool === currentToolCall.tool &&
-                        JSON.stringify(lastToolCall.params) === JSON.stringify(currentToolCall.params);
-                    if (isDuplicate) {
-                        duplicateCount = lastToolCall.count || 0;
-                        // 允许最多 2 次重复调用（给 AI 时间完成任务）
-                        if (duplicateCount >= 2) {
-                            console.log(chalk_1.default.yellow('[Duplicate Detection] 达到重复限制，强制完成'));
-                            console.log(chalk_1.default.cyan(`\n✓ ${action.payload.tool_name || action.type} 已完成\n`));
-                            if (agentRenderer) {
-                                agentRenderer.buffer = '';
-                                agentRenderer.quietMode = true;
-                                agentRenderer.finish();
-                            }
-                            break;
+            }
+        }
+        // Learn from this execution
+        this.learnFromExecution(userInput, mode, thought).catch(() => { });
+        return lastToolCall;
+    }
+    async handleFailedExecution(result, action, mode, thought, userInput) {
+        const actualToolName = action.payload?.tool_name || action.type;
+        this.context.addToolResult(actualToolName, `Error: ${result.error}`);
+        console.log(chalk_1.default.red(`[ERROR] ${result.error}`));
+        // Record to error tracker
+        if (action.type === 'tool_call') {
+            errorTracker_1.ErrorTracker.recordError(action.payload.tool_name, action.payload.parameters, result.error || 'Unknown error', { mode, model: thought.modelName, userInput });
+        }
+        else if (action.type === 'shell_cmd') {
+            errorTracker_1.ErrorTracker.recordError('shell_cmd', { command: action.payload.command }, result.error || 'Unknown error', { mode, model: thought.modelName, userInput });
+        }
+        // Auto-fix for shell_cmd failures
+        if (action.type === 'shell_cmd' && result.error) {
+            console.log(chalk_1.default.yellow('[AutoFix] 尝试自动修复命令...'));
+            try {
+                const { autoFixCommand } = await Promise.resolve().then(() => __importStar(require('../core/autofix')));
+                const { getOSProfile } = await Promise.resolve().then(() => __importStar(require('../core/os')));
+                const os = getOSProfile();
+                const fixPlan = await autoFixCommand(action.payload.command, result.error || result.output || '', os, thought.modelName);
+                if (fixPlan && fixPlan.command) {
+                    console.log(chalk_1.default.cyan(`[AutoFix] 建议修复命令: ${fixPlan.command}`));
+                    console.log(chalk_1.default.gray(`[AutoFix] 说明: ${fixPlan.plan || '无'}`));
+                    this.context.addMessage('system', `自动修复建议：${fixPlan.command}\n原因：${fixPlan.plan || '无'}`);
+                    if (fixPlan.risk === 'low') {
+                        console.log(chalk_1.default.yellow('[AutoFix] 修复方案风险低，自动执行...'));
+                        const fixedAction = {
+                            id: (0, crypto_1.randomUUID)(),
+                            type: 'shell_cmd',
+                            payload: { command: fixPlan.command },
+                            riskLevel: 'low',
+                            reasoning: `AutoFix from failed command: ${action.payload.command}`
+                        };
+                        const fixedResult = await executor_1.ToolExecutor.execute(fixedAction);
+                        if (fixedResult.success) {
+                            console.log(chalk_1.default.green('[AutoFix] 修复成功！'));
+                            this.context.addToolResult('shell_cmd', fixedResult.output);
+                            return ''; // clear error
                         }
                         else {
-                            console.log(chalk_1.default.gray(`[Repeat Detection] 重复调用 (${duplicateCount + 1}/2)，继续...`));
+                            console.log(chalk_1.default.red('[AutoFix] 修复失败，继续原始错误处理'));
                         }
                     }
-                    // 更新上次工具调用记录（包含计数）
-                    if (isDuplicate && lastToolCall) {
-                        lastToolCall.count = duplicateCount + 1;
-                    }
-                    else {
-                        lastToolCall = { ...currentToolCall, count: 0 };
-                    }
-                    // 智能完成：根据工具类型和用户意图决定是否自动完成
-                    if (action.type === 'tool_call') {
-                        const toolName = action.payload.tool_name;
-                        // 写入文件成功后自动完成
-                        if (toolName === 'write_file') {
-                            console.log(chalk_1.default.gray('[Auto-Complete] 文件写入成功，自动完成任务'));
-                            console.log(chalk_1.default.green(`✓ 已创建文件: ${action.payload.parameters.path}\n`));
-                            this.context.addMessage("assistant", `已成功创建文件 ${action.payload.parameters.path}`);
-                            // 清除 Write Mode 标记，因为写入已完成
-                            writeModeFileRead = null;
-                            writeModeFileContent = null;
-                            if (agentRenderer) {
-                                agentRenderer.buffer = '';
-                                agentRenderer.quietMode = true;
-                                agentRenderer.finish();
-                            }
-                            break;
-                        }
-                        // 只读工具处理
-                        const readOnlyTools = ['read_file', 'list_files', 'read_file_lines', 'read_file_lines_from_end', 'file_info', 'git_status', 'git_log', 'git_diff', 'list_directory_tree', 'search_in_files', 'search_symbol', 'continue_reading', 'analyze_dependencies'];
-                        if (readOnlyTools.includes(toolName)) {
-                            // 检测用户意图：
-                            // 1. 如果要求"分析"、"解释"等，则不自动完成（继续分析）
-                            // 2. 如果涉及写入操作（替换、修改、添加等），则不自动完成（继续执行写入）
-                            // 使用更精确的匹配，避免文件名中的关键词（如 git_reviews.md 中的 review）误触发
-                            const requiresAnalysis = /^(.*?)(帮我|请)?(分析|解释|说明|总结)|\b(review|explain)\s+(this|the|it)\b/i.test(userInput);
-                            const requiresWrite = /替换|replace|修改|modify|添加|append|插入|insert|删除|delete|移除|remove|更新|update|改成|改成|改为/i.test(userInput);
-                            if (!requiresAnalysis && !requiresWrite) {
-                                // 简单读取请求，直接返回结果
-                                console.log(chalk_1.default.gray('[Auto-Complete] 只读工具执行成功，自动完成任务'));
-                                console.log(chalk_1.default.cyan(`\n📄 ${toolName} 结果：\n`));
-                                console.log(result.output);
-                                this.context.addMessage("assistant", `已成功执行 ${toolName}，结果：\n${result.output}`);
-                                if (agentRenderer) {
-                                    agentRenderer.buffer = '';
-                                    agentRenderer.quietMode = true;
-                                    agentRenderer.finish();
-                                }
-                                break;
-                            }
-                            else if (requiresWrite) {
-                                // 写入请求，继续循环让 AI 执行写入操作
-                                console.log(chalk_1.default.gray('[Write Mode] 文件已读取，继续执行写入操作...'));
-                                // 标记此文件已在 Write Mode 下读取，防止重复读取
-                                writeModeFileRead = action.payload.parameters.path;
-                                // 缓存文件内容，避免重复读取
-                                writeModeFileContent = result.output;
-                                // 添加系统消息明确指导 AI 下一步操作
-                                this.context.addMessage('system', `文件 ${action.payload.parameters.path} 已成功读取。请根据用户要求修改内容后，调用 write_file 工具写入文件。不要重复调用 read_file！`);
-                            }
-                            else {
-                                // 分析请求，继续循环让 AI 分析内容
-                                console.log(chalk_1.default.gray('[Analysis Mode] 文件已读取，继续分析...'));
-                            }
-                        }
-                    }
-                    // Learn from this successful execution
-                    try {
-                        const { createExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionRecord')));
-                        const { inferCapabilityRequirement } = await Promise.resolve().then(() => __importStar(require('../core/capabilityInference')));
-                        const { saveExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
-                        const record = createExecutionRecord(`agent-${mode}`, { required: [], preferred: [] }, {
-                            aiProxyUrl: { value: '', source: 'built-in' },
-                            defaultModel: { value: '', source: 'built-in' },
-                            accountType: { value: 'free', source: 'built-in' }
-                        }, { selected: null, candidates: [], fallbackOccurred: false }, { success: true }, undefined, userInput, mode);
-                        // Attach thought/plan data for skill learning
-                        record.llmResult = { plan: thought.parsedPlan };
-                        record.input = { rawInput: userInput };
-                        const savedRecordId = saveExecutionRecord(record);
-                        const { loadExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
-                        const savedRecord = loadExecutionRecord(savedRecordId);
-                        if (savedRecord) {
-                            const { learnSkillFromRecord } = await Promise.resolve().then(() => __importStar(require('./skills')));
-                            learnSkillFromRecord(savedRecord, true);
-                        }
-                    }
-                    catch (error) {
-                        console.warn(chalk_1.default.yellow(`[Skill Learning] Failed: ${error}`));
-                    }
                 }
                 else {
-                    // 失败时记录错误，尝试自动修复
-                    lastError = result.error;
-                    const actualToolName = action.payload?.tool_name || action.type;
-                    this.context.addToolResult(actualToolName, `Error: ${result.error}`);
-                    console.log(chalk_1.default.red(`[ERROR] ${result.error}`));
-                    // 记录到错误追踪器
-                    if (action.type === 'tool_call') {
-                        errorTracker_1.ErrorTracker.recordError(action.payload.tool_name, action.payload.parameters, result.error || 'Unknown error', { mode, model: thought.modelName, userInput });
-                    }
-                    else if (action.type === 'shell_cmd') {
-                        errorTracker_1.ErrorTracker.recordError('shell_cmd', { command: action.payload.command }, result.error || 'Unknown error', { mode, model: thought.modelName, userInput });
-                    }
-                    // 尝试自动修复（仅针对 shell_cmd 失败）
-                    if (action.type === 'shell_cmd' && result.error) {
-                        console.log(chalk_1.default.yellow('[AutoFix] 尝试自动修复命令...'));
-                        try {
-                            const { autoFixCommand } = await Promise.resolve().then(() => __importStar(require('../core/autofix')));
-                            const { getOSProfile } = await Promise.resolve().then(() => __importStar(require('../core/os')));
-                            const os = getOSProfile();
-                            const fixPlan = await autoFixCommand(action.payload.command, result.error || result.output || '', os, thought.modelName);
-                            if (fixPlan && fixPlan.command) {
-                                console.log(chalk_1.default.cyan(`[AutoFix] 建议修复命令: ${fixPlan.command}`));
-                                console.log(chalk_1.default.gray(`[AutoFix] 说明: ${fixPlan.plan || '无'}`));
-                                // 将修复建议添加到上下文
-                                this.context.addMessage('system', `自动修复建议：${fixPlan.command}\n原因：${fixPlan.plan || '无'}`);
-                                // 如果修复方案风险低，直接执行
-                                if (fixPlan.risk === 'low') {
-                                    console.log(chalk_1.default.yellow('[AutoFix] 修复方案风险低，自动执行...'));
-                                    const fixedAction = {
-                                        id: (0, crypto_1.randomUUID)(),
-                                        type: 'shell_cmd',
-                                        payload: { command: fixPlan.command },
-                                        riskLevel: 'low',
-                                        reasoning: `AutoFix from failed command: ${action.payload.command}`
-                                    };
-                                    const fixedResult = await executor_1.ToolExecutor.execute(fixedAction);
-                                    if (fixedResult.success) {
-                                        console.log(chalk_1.default.green('[AutoFix] 修复成功！'));
-                                        lastError = undefined;
-                                        this.context.addToolResult('shell_cmd', fixedResult.output);
-                                        continue;
-                                    }
-                                    else {
-                                        console.log(chalk_1.default.red('[AutoFix] 修复失败，继续原始错误处理'));
-                                    }
-                                }
-                            }
-                            else {
-                                console.log(chalk_1.default.gray('[AutoFix] 无法生成修复建议'));
-                            }
-                        }
-                        catch (fixError) {
-                            console.warn(chalk_1.default.yellow(`[AutoFix] 修复过程出错: ${fixError.message}`));
-                        }
-                    }
+                    console.log(chalk_1.default.gray('[AutoFix] 无法生成修复建议'));
                 }
             }
-            catch (error) {
-                let errorMessage = '未知执行错误';
-                if (error instanceof Error) {
-                    errorMessage = error.message;
-                }
-                else if (typeof error === 'string') {
-                    errorMessage = error;
-                }
-                console.log(chalk_1.default.red(`\n❌ 任务执行失败 [Action: ${thought?.type}]: ${errorMessage}`));
-                this.context.addMessage("system", `执行引擎错误 [Action: ${thought?.type}]: ${errorMessage}`);
-                break;
+            catch (fixError) {
+                console.warn(chalk_1.default.yellow(`[AutoFix] 修复过程出错: ${fixError.message}`));
+            }
+        }
+        return result.error;
+    }
+    /**
+     * 语义完成检测：输出是否已经是直接答案
+     * 用于防止 AI 在已经获取答案后继续无意义地查询
+     */
+    isSemanticComplete(output, userInput) {
+        const trimmed = output.trim();
+        // 输出必须是短的（直接答案通常很短）
+        if (trimmed.length > 100 || trimmed.length === 0)
+            return false;
+        // 输出不能包含多行
+        if (trimmed.includes('\n'))
+            return false;
+        const q = userInput.toLowerCase();
+        // 1. 问"大小/多少字节" → 输出是纯数字 或 带单位的短文本（如 "4.0K"、"288K"、"1.5MB"）
+        if (/(大小|多少字节|多大|size|bytes?)/.test(q)) {
+            if (/^\d+(\.\d+)?$/.test(trimmed))
+                return true;
+            // 人类可读格式：数字+单位，可能带前后空格和路径
+            if (/^[\d.]+\s*[KMGT]?B?$/i.test(trimmed))
+                return true;
+            // 带路径的格式：如 "4.0K    ./package.json"
+            if (/^[\d.]+\s*[KMGT]?B?\s+\S+/i.test(trimmed))
+                return true;
+        }
+        // 2. 问"几个/多少个/数量/count" → 输出是纯数字
+        if (/(几个|多少个|数量|count|how many)/.test(q) && /^\d+$/.test(trimmed))
+            return true;
+        // 3. 问"最大/最小的文件" → 输出是 "数字 路径" 格式（如 "0 ./foo.txt"）
+        if (/(最大|最小|largest|smallest).*文件/.test(q) && /^\d+\s+\.\//.test(trimmed))
+            return true;
+        // 4. 问"最新/最早"的文件 → 输出是单个文件名或带信息的短行
+        if (/(最新|最旧|最近|最早|newest|latest|oldest)/.test(q) && trimmed.split(/\s+/).length <= 5 && trimmed.length < 80)
+            return true;
+        // 5. 问"行数" → 输出是纯数字
+        if (/(行数|多少行|line.?count|多少 line)/.test(q) && /^\d+$/.test(trimmed))
+            return true;
+        // 6. 通用：输出是短小的、看起来像答案的内容（单行，< 50 字符，不是命令输出）
+        // 排除：以 - 开头（ls -l 风格）、包含多个空格的长行
+        if (trimmed.length < 50 && !trimmed.startsWith('-') && trimmed.split(/\s+/).length <= 4) {
+            // 包含数字或路径，看起来像答案
+            if (/\d+|\.\//.test(trimmed))
+                return true;
+        }
+        return false;
+    }
+    async learnFromExecution(userInput, mode, thought) {
+        try {
+            const { createExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionRecord')));
+            const { saveExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
+            const record = createExecutionRecord(`agent-${mode}`, { required: [], preferred: [] }, { aiProxyUrl: { value: '', source: 'built-in' }, defaultModel: { value: '', source: 'built-in' }, accountType: { value: 'free', source: 'built-in' } }, { selected: null, candidates: [], fallbackOccurred: false }, { success: true }, undefined, userInput, mode);
+            record.llmResult = { plan: thought.parsedPlan };
+            record.input = { rawInput: userInput };
+            const savedRecordId = saveExecutionRecord(record);
+            const { loadExecutionRecord } = await Promise.resolve().then(() => __importStar(require('../core/executionStore')));
+            const savedRecord = loadExecutionRecord(savedRecordId);
+            if (savedRecord) {
+                const { learnSkillFromRecord } = await Promise.resolve().then(() => __importStar(require('./skills')));
+                learnSkillFromRecord(savedRecord, true);
             }
         }
-        if (turnCount >= maxTurns) {
-            console.log(chalk_1.default.red(`\n⚠️ Max turns (${maxTurns}) reached.`));
+        catch (error) {
+            console.warn(chalk_1.default.yellow(`[Skill Learning] Failed: ${error}`));
         }
     }
 }