kiku-agent 1.0.5 → 1.0.7

package/README.md

@@ -48,7 +48,24 @@ The same package also installs the stdio MCP server command:
 kiku-mcp
 ```
 
-For a local checkout before the package is published, use `npm pack` followed by `npm install -g ./kiku-agent-1.0.5.tgz`.
+### Updating the CLI and MCP server
+
+Kiku uses the published `kiku-agent` package on npm as its update source. Inside an interactive session:
+
+```text
+/update                         # Check the configured release channel
+/update install                 # Install the stable `latest` release
+/update install next            # Install a preview release
+/update install 1.2.3           # Install an exact published version
+/update channel next            # Persist the preferred channel
+/update rollback                # Restore the previously recorded version
+```
+
+Automatic installation is available for global npm installs. Source checkouts, local dependencies, and `npx` executions receive a safe manual command instead of having their files replaced. Major updates require `--allow-major`; deprecated releases require `--allow-deprecated`. npm commands are executed without a shell and only allow known channel names or strict semantic versions.
+
+After an update or rollback, restart `kiku-agent` and any application hosting `kiku-mcp`. Update settings and rollback history are stored in the normal Kiku user data directory as `update-state.json`; sessions, memory, `.env`, and YouTube cookies are not modified.
+
+For a local checkout before the package is published, use `npm pack` followed by `npm install -g ./kiku-agent-1.0.7.tgz`.
 
 ### ✨ Features
 - **LM Studio Integration**: Connects seamlessly to LM Studio's OpenAI-compatible server at `http://127.0.0.1:1234/v1` (default model `google/gemma-4-e4b`, with auto-detection of loaded models).
@@ -84,10 +101,11 @@ npm run kiku-agent
 
 ### Launch with Specific Cloud Provider & Model
 ```bash
-npm run kiku-agent -- --provider openai --model gpt-4o-mini
-npm run kiku-agent -- --provider gemini --model gemini-2.5-flash
-npm run kiku-agent -- --provider groq --model llama-3.1-8b-instant
-npm run kiku-agent -- --provider openrouter --model google/gemini-2.5-flash
+npm run kiku-agent -- --provider openai --model gpt-5.6-luna
+npm run kiku-agent -- --provider gemini --model gemini-3.7-flash
+npm run kiku-agent -- --provider groq --model openai/gpt-oss-120b
+npm run kiku-agent -- --provider deepseek --model deepseek-v4-flash
+npm run kiku-agent -- --provider openrouter --model openrouter/auto
 ```
 
 ### One-shot Question Query
@@ -104,9 +122,10 @@ npm run kiku-agent -- --model google/gemma-4-e4b-Q4_K_M --url http://127.0.0.1:1
 | Flag | Description |
 |---|---|
 | `-h`, `--help` | Show usage help |
+| `-V`, `--version` | Print the installed package version and exit |
 | `-P`, `--provider <name>` | Select active provider (`lmstudio`, `openai`, `gemini`, `groq`, `deepseek`, `openrouter`, `custom`) |
 | `-k`, `--api-key <key>` | Specify API key for cloud provider |
-| `-m`, `--model <id>` | Override model ID (e.g. `gpt-4o-mini`, `google/gemma-4-e4b-Q4_K_M`) |
+| `-m`, `--model <id>` | Override model ID (e.g. `gpt-5.6-luna`, `google/gemma-4-e4b-Q4_K_M`) |
 | `-u`, `--url <url>` | Override provider base URL |
 | `-t`, `--tools` | List all registered tools |
 | `-s`, `--stats` | Print catalog overview statistics |
@@ -131,8 +150,11 @@ Run `/help` inside a session for the full list. Commonly used:
 - `/model use <model-id>` — Switch runtime model without restarting the CLI
 - `/model reconnect` — Recheck server and leave catalog-only mode after a chat model is loaded
 - `/doctor` — Check Node, writable app data, LM Studio, model availability, FFmpeg, speaker, and yt-dlp
+- `/update [check|install|channel|rollback]` — Manage npm-backed CLI and MCP package updates
 - `/tools` — View all tool names and parameter descriptions
-- `/details [Qn]` — Expand recorded thinking and tool results for a question
+- `/details [Qn]` — Open a compact diagnostic overview for the latest or selected question (F2 opens latest)
+- `/details [Qn] answer|thinking|tools` — Inspect one diagnostic section without dumping every raw result
+- `/details [Qn] tool <n>` — Inspect bounded raw arguments and result for one tool call
 - `/stats` — Print summary counts of dramas, songs, brands, and awards
 - `/thinking [show|hide]` — Toggle reasoning process visibility
 - `/save` — Export session transcript to Markdown file
@@ -206,30 +228,33 @@ Run `/help` inside a session for the full list. Commonly used:
 | Provider | Recommended Active Models | Pricing / Cost Tier | Doc & API Reference | Required ENV Variable |
 |---|---|---|---|---|
 | **LM Studio** (Local) | `google/gemma-4-e4b` *(Q4_K_M, Q6_K, Q8_0)* | **100% Free / Local** (Zero token cost) | [Docs](https://lmstudio.ai/docs) • [API Ref](https://lmstudio.ai/docs/api/local-server) | `LMSTUDIO_BASE_URL` (default: `http://127.0.0.1:1234/v1`) |
-| **OpenAI** | `gpt-4o-mini` *(default)*, `gpt-4o`, `o3-mini` | **Ultra-Cheap** (~$0.15 / 1M input) | [Docs](https://platform.openai.com/docs) • [API Ref](https://platform.openai.com/docs/api-reference) | `OPENAI_API_KEY` |
-| **Google Gemini** | `gemini-2.5-flash` *(default)*, `gemini-2.5-flash-lite`, `gemini-2.5-pro`, `gemini-3.1-pro` | **Ultra-Cheap** ($0.10 / 1M in, 1M-2M context) | [Docs](https://ai.google.dev/docs) • [API Ref](https://ai.google.dev/api) | `GEMINI_API_KEY` |
-| **Groq** | `llama-3.3-70b-versatile` *(default)*, `llama-3.1-8b-instant`, `gpt-oss-120b` | **LPU Ultra-Fast** ($0.05 - $0.59 / 1M in) | [Docs](https://console.groq.com/docs/quickstart) • [API Ref](https://console.groq.com/docs/api-reference) | `GROQ_API_KEY` |
-| **DeepSeek** | `deepseek-chat` *(default, V3)*, `deepseek-reasoner` *(R1 CoT)* | **Ultra-Low Cost** ($0.14 - $0.27 / 1M in) | [Docs](https://api-docs.deepseek.com) • [API Ref](https://api-docs.deepseek.com/api/create-chat-completion) | `DEEPSEEK_API_KEY` |
-| **OpenRouter** | `google/gemini-2.5-flash`, `deepseek/deepseek-chat`, `deepseek/deepseek-r1`, `meta-llama/llama-3.3-70b-instruct`, `qwen/qwen-2.5-72b-instruct`, `:free` | **Flexible Aggregator** (Free & Paid) | [Docs](https://openrouter.ai/docs/quick-start) • [API Ref](https://openrouter.ai/docs/api_reference/overview) | `OPENROUTER_API_KEY` |
+| **OpenAI** | `gpt-5.6-luna` *(default)*, `gpt-5.6-terra`, `gpt-5.6-sol` | **Cost-sensitive → Flagship** ($0.20 - $4.00 / 1M input) | [Docs](https://developers.openai.com/api/docs/models) • [API Ref](https://developers.openai.com/api/reference) | `OPENAI_API_KEY` |
+| **Google Gemini** | `gemini-3.7-flash` *(default)*, `gemini-3.5-flash-lite`, `gemini-3.6-flash` | **Free tier / Stable Flash** ($0.30 - $0.75 / 1M input) | [Docs](https://ai.google.dev/gemini-api/docs/models) • [API Ref](https://ai.google.dev/api/generate-content) | `GEMINI_API_KEY` |
+| **Groq** | `openai/gpt-oss-120b` *(default)*, `openai/gpt-oss-20b` | **LPU Ultra-Fast** ($0.075 - $0.15 / 1M input) | [Docs](https://console.groq.com/docs/quickstart) • [API Ref](https://console.groq.com/docs/api-reference) | `GROQ_API_KEY` |
+| **DeepSeek** | `deepseek-v4-flash` *(default)*, `deepseek-v4-pro` | **V4 / 1M context** ($0.22 - $1.32 / 1M uncached input) | [Docs](https://api-docs.deepseek.com/quick_start/pricing/) • [API Ref](https://api-docs.deepseek.com/api/create-chat-completion) | `DEEPSEEK_API_KEY` |
+| **OpenRouter** | `openrouter/auto` *(default, dynamic catalog)* | **Automatic routing** (billed at routed model rate) | [Docs](https://openrouter.ai/docs/guides/routing/routers/auto-router) • [API Ref](https://openrouter.ai/docs/api/reference/overview) | `OPENROUTER_API_KEY` |
 | **Custom Gateway** | Any OpenAI-compatible server (vLLM, Ollama, TGI, Azure) | **Custom** | [OpenAI Spec](https://platform.openai.com/docs/api-reference) | `LLM_BASE_URL`, `LLM_API_KEY` |
 
 ```bash
 # Launch with a specific cloud provider and model
-npm run kiku-agent -- --provider openai --model gpt-4o-mini
-npm run kiku-agent -- --provider gemini --model gemini-2.5-flash
-npm run kiku-agent -- --provider groq --model llama-3.1-8b-instant
-npm run kiku-agent -- --provider openrouter --model google/gemini-2.5-flash
+npm run kiku-agent -- --provider openai --model gpt-5.6-luna
+npm run kiku-agent -- --provider gemini --model gemini-3.7-flash
+npm run kiku-agent -- --provider groq --model openai/gpt-oss-120b
+npm run kiku-agent -- --provider deepseek --model deepseek-v4-flash
+npm run kiku-agent -- --provider openrouter --model openrouter/auto
 
 # Manage providers interactively within the agent CLI
 /provider list                                     # View all 7 providers and key status
-/provider use groq llama-3.1-8b-instant            # Switch provider and model in one command
+/provider use groq openai/gpt-oss-120b             # Switch provider and model in one command
 /provider models gemini                            # List curated cost-effective models with pricing & context
 /provider info openai                              # Show developer documentation & API details
 /provider test groq                                # Test live endpoint connection and measure latency
 /provider setkey groq gsk-...                      # Set or update API key dynamically at runtime
 /model recommend                                   # View recommended models for active provider
-/model use gpt-4o-mini                             # Switch active model on current provider
-```
+/model use gpt-5.6-luna                            # Switch active model on current provider
+```
+
+At interactive startup, provider switches, and before a one-shot query, the CLI checks the provider's live `/models` catalog. If a configured model has been retired, it selects the first available model from that provider's maintained recommendation list and reports the fallback. Legacy IDs for Groq, DeepSeek, and selected renamed models are normalized automatically.
 
 ---
 
@@ -252,6 +277,8 @@ The production deployment exposes a stateless Streamable HTTP endpoint at `https
 /mcp tools
 ```
 
+`/mcp` shows the connection summary plus every available MCP tool command in a compact list. `/mcp tools` expands the same list with parameters and descriptions. Long command panels open at their first page in the classic TUI; use Page Down or End to continue through the result.
+
 ---
 
 ## ⚙️ Environment Variables

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "kiku-agent",
-  "version": "1.0.5",
+  "version": "1.0.7",
   "description": "Kiku AI terminal agent and MCP server for Ju Jingyi knowledge, media, and career data.",
   "type": "module",
   "bin": {
@@ -23,6 +23,16 @@
   "engines": {
     "node": ">=20.0.0"
   },
+  "kikuUpdate": {
+    "channel": "stable",
+    "releaseSummary": [
+      "Updated production model defaults and live fallback resolution across cloud providers",
+      "Hardened REST API key validation and isolated rate-limit probes",
+      "Fixed discography filtering, MCP configuration isolation, and environment loading order",
+      "Secured YouTube iframe messaging and expanded provider and API regression coverage"
+    ],
+    "releaseUrl": "https://www.npmjs.com/package/kiku-agent"
+  },
   "keywords": [
     "kiku",
     "ju-jingyi",

package/server/agent/agentCore.js

@@ -9,6 +9,7 @@ import { sanitizeQuery, isLikelySafe } from './sanitizer.js';
 import { createAbortError, isAbortError } from './errors.js';
 import { ResearchSubagent } from './subagent.js';
 import { ResearchOrchestrator } from './research.js';
+import { SuggestionEngine } from './suggestions.js';
 import { McpClientManager } from './mcpClient.js';
 import { DynamicToolRouter } from './toolRouter.js';
 import { contextCompactor } from './contextCompactor.js';
@@ -62,6 +63,7 @@ export class KikuAgent {
     this.mcpManager = options.mcpManager || new McpClientManager();
     this.router = new DynamicToolRouter(this.getAllToolsSchema());
     this.research = new ResearchOrchestrator({ client: this.client });
+    this.suggestions = new SuggestionEngine({ client: this.client, memory: this.memory });
     this.systemPrompt = this.buildSystemPrompt();
     this.initHistory();
     this.restoredMessageCount = 0;
@@ -88,7 +90,7 @@ export class KikuAgent {
         await this.mcpManager.loadPersistedServers();
         this.router.setTools(this.getAllToolsSchema());
         this.systemPrompt = this.buildSystemPrompt();
-      } catch {}
+      } catch { }
     }
     if (!this.conversationHistory || this.conversationHistory.length === 0) {
       this.initHistory();
@@ -128,6 +130,20 @@ export class KikuAgent {
     return this.research.cancel();
   }
 
+  async generateSuggestions(options = {}) {
+    await this._ready;
+    return this.suggestions.generateSuggestions(options);
+  }
+
+  async runSuggestionBatch(targets, options = {}) {
+    await this._ready;
+    return this.suggestions.runBatchSubagents(targets, options);
+  }
+
+  getLastSuggestions() {
+    return this.suggestions.getLastSuggestions();
+  }
+
   initHistory() {
     this.conversationHistory = [
       {
@@ -139,12 +155,12 @@ export class KikuAgent {
 
   clearMemory() {
     if (this.session.messages.length > 0) {
-      this.memory.archiveSession(this.session.toSnapshot()).catch(() => {});
-      this.memory.consolidateSession(this.session.toSnapshot()).catch(() => {});
+      this.memory.archiveSession(this.session.toSnapshot()).catch(() => { });
+      this.memory.consolidateSession(this.session.toSnapshot()).catch(() => { });
     }
     this.initHistory();
     this.session.reset();
-    this.session.clearDisk().catch(() => {});
+    this.session.clearDisk().catch(() => { });
   }
 
   async loadSession() {
@@ -156,7 +172,7 @@ export class KikuAgent {
           content: this.systemPrompt
         }
       ];
-      
+
       for (const msg of this.session.messages) {
         if (msg.role === 'user') {
           const textPart = msg.parts.find(p => p.type === 'text');
@@ -224,6 +240,47 @@ export class KikuAgent {
     return result;
   }
 
+  async removeMcp(name) {
+    const result = await this.mcpManager.remove(name);
+    this.router.setTools(this.getAllToolsSchema());
+    this.systemPrompt = this.buildSystemPrompt();
+    if (this.conversationHistory && this.conversationHistory[0]?.role === 'system') {
+      this.conversationHistory[0].content = this.getEnrichedSystemPrompt();
+    }
+    return result;
+  }
+
+  async reconnectMcp(name) {
+    const result = name && name !== 'all'
+      ? await this.mcpManager.reconnect(name)
+      : await this.mcpManager.reconnectAll();
+    this.router.setTools(this.getAllToolsSchema());
+    this.systemPrompt = this.buildSystemPrompt();
+    if (this.conversationHistory && this.conversationHistory[0]?.role === 'system') {
+      this.conversationHistory[0].content = this.getEnrichedSystemPrompt();
+    }
+    return result;
+  }
+
+  async saveMcpConfig(target = 'global') {
+    if (target === 'workspace' || target === 'local') {
+      return this.mcpManager.saveToWorkspace();
+    }
+    return this.mcpManager.savePersistedConfig();
+  }
+
+  async loadMcpConfig(customPath) {
+    const results = customPath
+      ? await this.mcpManager.loadConfigFile(customPath)
+      : await this.mcpManager.loadPersistedServers();
+    this.router.setTools(this.getAllToolsSchema());
+    this.systemPrompt = this.buildSystemPrompt();
+    if (this.conversationHistory && this.conversationHistory[0]?.role === 'system') {
+      this.conversationHistory[0].content = this.getEnrichedSystemPrompt();
+    }
+    return results;
+  }
+
   buildSystemPrompt() {
     const allTools = this.getAllToolsSchema();
     const toolsSummary = allTools.map(t => {
@@ -280,9 +337,9 @@ ${toolsSummary}
    - Directory listing ➔ list_directory
    - Screenshot ➔ screenshot_webpage
 
-6. AGGRESSIVE WEB SEARCH FALLBACK:
-   - If local tools return insufficient results, IMMEDIATELY call web_search or web_extract.
-   - Combine local data WITH web search for complete, accurate answers.
+6. AGGRESSIVE WEB SEARCH FALLBACK (CRITICAL):
+   - If a local catalog tool (e.g. search_works, search_discography, get_brand_endorsements, get_awards) returns not_found, empty, or 0 results, DO NOT give up or guess. You MUST IMMEDIATELY call web_search or web_extract in the next step to look up live external facts online.
+   - Combine local catalog data WITH live web search for complete, accurate, and up-to-date answers.
    - Use web_extract for comprehensive research on any topic.
 
 7. LEGAL & CAREER KNOWLEDGE:
@@ -367,18 +424,18 @@ ${toolsSummary}
     for (const topic of topics) {
       this.memory.trackTopicInterest(topic);
     }
-    
+
     const {
-      onStepStart = () => {},
-      onReasoningStart = () => {},
-      onReasoningDelta = () => {},
-      onReasoningEnd = () => {},
-      onToolStart = () => {},
-      onToolResult = () => {},
-      onTokenStart = () => {},
-      onTokenDelta = () => {},
-      onFinalAnswer = () => {},
-      onError = () => {},
+      onStepStart = () => { },
+      onReasoningStart = () => { },
+      onReasoningDelta = () => { },
+      onReasoningEnd = () => { },
+      onToolStart = () => { },
+      onToolResult = () => { },
+      onTokenStart = () => { },
+      onTokenDelta = () => { },
+      onFinalAnswer = () => { },
+      onError = () => { },
       signal = null
     } = callbacks;
 
@@ -509,6 +566,15 @@ ${toolsSummary}
       let toolCalls = assistantMessage.tool_calls || [];
       if (toolCalls.length === 0 && assistantMessage.content) {
         toolCalls = this.parseTextToolCalls(assistantMessage.content);
+       
+        if (toolCalls.length > 0) {
+          let cleaned = assistantMessage.content;
+
+          cleaned = cleaned.replace(/```(?:json)?\s*\{[\s\S]*?\}\s*```/g, '');
+        
+          cleaned = cleaned.replace(/Action:\s*[a-zA-Z0-9_]+\s*\nAction Input:\s*(?:\{[\s\S]*?\}|[^\n]+)/g, '');
+          assistantMessage.content = cleaned.trim();
+        }
       }
 
       this.conversationHistory.push(assistantMessage);
@@ -519,8 +585,8 @@ ${toolsSummary}
 
         const finalAnswer = assistantMessage.content || (
           dominantLang === 'zh' ? "抱歉，无法处理您的回答。" :
-          dominantLang === 'en' ? "Sorry, unable to process a response." :
-          "ขออภัยค่ะ ไม่สามารถประมวลผลคำตอบได้"
+            dominantLang === 'en' ? "Sorry, unable to process a response." :
+              "ขออภัยค่ะ ไม่สามารถประมวลผลคำตอบได้"
         );
         onFinalAnswer(finalAnswer, { durationMs: totalDurationMs, language: dominantLang });
         this._extractFactsFromFinalAnswer(finalAnswer);
@@ -532,11 +598,8 @@ ${toolsSummary}
         return finalAnswer;
       }
 
-      for (const [toolCallIndex, tc] of toolCalls.entries()) {
-        if (signal?.aborted) {
-          this.stateMachine.reset();
-          throw createAbortError();
-        }
+      const executionPromises = toolCalls.map(async (tc, toolCallIndex) => {
+        if (signal?.aborted) throw createAbortError();
 
         const toolName = tc?.function?.name || '';
         const toolCallId = tc?.id || `call_${queryId}_${currentIteration}_${toolCallIndex}`;
@@ -560,20 +623,21 @@ ${toolsSummary}
             validationWarnings: prepared.warnings,
           };
           eventBus.emit(Events.AGENT_TOOL_ERROR, {
-            queryId, toolName, args: tc.function.arguments,
+            queryId, toolName, args: tc?.function?.arguments,
             error: prepared.error,
           });
           onToolStart(toolName, {});
           onToolResult(toolName, rejectedResult, 0);
-          const toolMessage = {
-            role: 'tool',
-            tool_call_id: toolCallId,
-            name: toolName,
-            content: JSON.stringify(rejectedResult)
+          return {
+            toolMessage: {
+              role: 'tool',
+              tool_call_id: toolCallId,
+              name: toolName,
+              content: JSON.stringify(rejectedResult)
+            }
           };
-          this.conversationHistory.push(toolMessage);
-          continue;
         }
+
         if (prepared.warnings.length > 0) {
           eventBus.emit(Events.AGENT_TOOL_ERROR, {
             queryId, toolName, warnings: prepared.warnings,
@@ -581,7 +645,6 @@ ${toolsSummary}
         }
 
         const finalArgs = prepared.args;
-
         this.stateMachine.transition(AgentStates.FETCHING_TOOL, { toolName });
 
         const toolPart = this.session.addToolPart(currentAssistantSessionMsg, toolName, finalArgs);
@@ -610,30 +673,40 @@ ${toolsSummary}
 
         this._extractFactsFromToolResult(toolName, finalArgs, result);
 
-        const toolMessage = {
-          role: 'tool',
-          tool_call_id: toolCallId,
-          name: toolName,
-          content: JSON.stringify({
-            authoritative: true,
-            result,
-            instruction: result.status === 'not_found'
-              ? 'No matching data was found. Do not invent records; say that the catalog has no match or use an appropriate fallback tool.'
-              : result.status === 'error'
-                ? 'The tool failed. Do not claim that the requested facts were verified.'
-                : 'Use only facts present in this result when answering.'
-          })
+     
+        const cappedResult = this._capToolResultForHistory(result);
+
+        return {
+          toolMessage: {
+            role: 'tool',
+            tool_call_id: toolCallId,
+            name: toolName,
+            content: JSON.stringify({
+              authoritative: true,
+              result: cappedResult,
+              instruction: result.status === 'not_found'
+                ? 'No matching data was found. Do not invent records; say that the catalog has no match or use an appropriate fallback tool.'
+                : result.status === 'error'
+                  ? 'The tool failed. Do not claim that the requested facts were verified.'
+                  : 'Use only facts present in this result when answering. Do not echo raw JSON data — summarize in natural language.'
+            })
+          }
         };
+      });
 
-        this.conversationHistory.push(toolMessage);
+      const toolResults = await Promise.all(executionPromises);
+      for (const item of toolResults) {
+        if (item?.toolMessage) {
+          this.conversationHistory.push(item.toolMessage);
+        }
       }
     }
 
-    const langPrompt = dominantLang === 'zh' 
+    const langPrompt = dominantLang === 'zh'
       ? "请根据上述工具查询结果，直接回答（不要开头问候语），用流畅优雅的中文，先1-2句概述，再列要点。包含具体的 YouTube 真实链接（如 Direct Watch: https://www.youtube.com/watch?v=... หรือ Search: https://www.youtube.com/results?search_query=...）及事实数据。"
       : dominantLang === 'en'
-      ? "Based on the tool results above, answer directly (no opening greeting or template phrase). Give a 1-2 sentence summary, then concise bullet points in polished English. Include verified YouTube direct and search links (https://www.youtube.com/results?search_query=...)."
-      : "สรุปคำตอบจากผลลัพธ์เครื่องมือข้างต้น โดยไม่ต้องทักทาย ตอบตรงประเด็น 1-2 ประโยค แล้วตามด้วย bullet points สั้นๆ เป็นภาษาไทย พร้อมแสดงลิงก์ YouTube จริง (Direct Watch Link และ https://www.youtube.com/results?search_query=...) และข้อมูลผลงานที่ถูกต้องแม่นยำ";
+        ? "Based on the tool results above, answer directly (no opening greeting or template phrase). Give a 1-2 sentence summary, then concise bullet points in polished English. Include verified YouTube direct and search links (https://www.youtube.com/results?search_query=...)."
+        : "สรุปคำตอบจากผลลัพธ์เครื่องมือข้างต้น โดยไม่ต้องทักทาย ตอบตรงประเด็น 1-2 ประโยค แล้วตามด้วย bullet points สั้นๆ เป็นภาษาไทย พร้อมแสดงลิงก์ YouTube จริง (Direct Watch Link และ https://www.youtube.com/results?search_query=...) และข้อมูลผลงานที่ถูกต้องแม่นยำ";
 
     this.conversationHistory.push({ role: 'user', content: langPrompt });
 
@@ -677,19 +750,55 @@ ${toolsSummary}
     const defaultDone = dominantLang === 'zh' ? "查询完毕" : dominantLang === 'en' ? "Search complete." : "เสร็จสิ้นการค้นหาข้อมูล";
     const output = finalRes?.content || defaultDone;
     onFinalAnswer(output, { durationMs: totalDurationMs, language: dominantLang });
-    
+
     this._extractFactsFromFinalAnswer(output);
 
     this.stateMachine.transition(AgentStates.COMPLETE);
     eventBus.emit(Events.AGENT_FINAL_ANSWER, { queryId, answer: output, durationMs: totalDurationMs });
     eventBus.emit(Events.AGENT_QUERY_END, { queryId, durationMs: totalDurationMs });
     this.stateMachine.reset();
-    
+
     this.saveSession().catch(err => console.error('Session save failed:', err));
-    
+
     return output;
   }
 
+  _capToolResultForHistory(result) {
+    if (!result || typeof result !== 'object') return result;
+
+    const MAX_RESULT_BYTES = 4096;
+    const serialized = JSON.stringify(result);
+    if (serialized.length <= MAX_RESULT_BYTES) return result;
+
+
+    const truncate = (obj) => {
+      if (Array.isArray(obj)) {
+        if (obj.length > 8) {
+          const kept = obj.slice(0, 8).map(truncate);
+          kept.push({ _truncated: true, totalItems: obj.length, showing: 8 });
+          return kept;
+        }
+        return obj.map(truncate);
+      }
+      if (obj && typeof obj === 'object') {
+        const out = {};
+        for (const [key, val] of Object.entries(obj)) {
+          out[key] = truncate(val);
+        }
+        return out;
+      }
+    
+      if (typeof obj === 'string' && obj.length > 500) {
+        return obj.slice(0, 500) + '…[truncated]';
+      }
+      return obj;
+    };
+
+    const capped = truncate(result);
+    capped._cappedForHistory = true;
+    return capped;
+  }
+
   _extractFactsFromToolResult(toolName, toolArgs, result) {
     if (!result || result.status === 'error' || result.status === 'not_found') return;