diff --git a/PATCHNOTES.md b/PATCHNOTES.md
index bfe526c..a0cc107 100644
--- a/PATCHNOTES.md
+++ b/PATCHNOTES.md
@@ -1,3 +1,12 @@
+# Patch Notes - v2.33.9 (2026-05-01)
+
+## 🚀 Core Engine Upgrade
+- **Agent Workflow Optimization:** Enhanced `AgentWorkflowManager` and `AgentFactory` for more robust multi-agent orchestration.
+- **Utility Refinement:** Core utilities and configuration logic updated for improved reliability and performance.
+- **UI/UX Sync:** Further alignment between sidebar provider and core agent logic for a seamless experience.
+
+---
+
 # Patch Notes - v2.33.8 (2026-05-01)
 
 ## 🛠️ Performance & Stability
diff --git a/package.json b/package.json
index 66c257b..07507d0 100644
--- a/package.json
+++ b/package.json
@@ -2,7 +2,7 @@
   "name": "g1nation",
   "displayName": "G1nation",
   "description": "High-performance autonomous local AI coding agent for VS Code. Features vectorized inference, asynchronous task management, and 100% offline processing.",
-  "version": "2.33.8",
+  "version": "2.33.9",
   "publisher": "connectailab",
   "license": "MIT",
   "icon": "assets/icon.png",
@@ -100,53 +100,6 @@
           "default": false,
           "description": "Enable Multi-Agent Workflow (Planner -> Researcher -> Writer) for complex tasks."
         },
-        "g1nation.multiAgentWorkflow": {
-          "type": "array",
-          "default": [
-            {
-              "id": "planner",
-              "name": "Planner",
-              "input": "original",
-              "prompt": "You are the [Master Strategist & Planner].\nTransform the user request into a precise execution blueprint.\nOutput Markdown with Objective, Core Challenges, Data Requirements, and Step-by-Step Tasks."
-            },
-            {
-              "id": "researcher",
-              "name": "Researcher",
-              "input": "combined",
-              "prompt": "You are the [Senior Technical Researcher].\nExtract, filter, and synthesize high-signal facts from the plan and available context.\nOutput Key Facts, Technical Deep-Dive, Gaps, and Summary of Knowledge."
-            },
-            {
-              "id": "writer",
-              "name": "Writer",
-              "input": "combined",
-              "prompt": "You are the [Lead Synthesis Writer & Editor].\nProduce the final response in the user language with a clear conclusion and actionable recommendation.\nPreserve important technical nuance without padding."
-            }
-          ],
-          "items": {
-            "type": "object",
-            "properties": {
-              "id": {
-                "type": "string",
-                "description": "Stable stage id."
-              },
-              "name": {
-                "type": "string",
-                "description": "Stage label shown in progress UI."
-              },
-              "input": {
-                "type": "string",
-                "enum": ["original", "previous", "combined"],
-                "default": "combined",
-                "description": "Which input this stage receives: original user request, previous stage output, or both."
-              },
-              "prompt": {
-                "type": "string",
-                "description": "Role and instructions for this stage."
-              }
-            }
-          },
-          "description": "Configurable Multi-Agent workflow. Add, remove, reorder, or rewrite stages in settings JSON."
-        },
         "g1nation.memoryEnabled": {
           "type": "boolean",
           "default": true,
diff --git a/src/agent.ts b/src/agent.ts
index ae311de..59b4574 100644
--- a/src/agent.ts
+++ b/src/agent.ts
@@ -97,6 +97,14 @@ export class AgentExecutor {
         return { problem, goal, reasoning };
     }
 
+    private sanitizeAssistantContent(text: string): string {
+        return text
+            .replace(/<rationale>[\s\S]*?<\/rationale>/gi, '')
+            .replace(/^\s*\[PROBLEM\][\s\S]*?\[GOAL\][\s\S]*?\[REASONING\][^\n]*(?:\n+|$)/i, '')
+            .replace(/^\s*\[PROBLEM\][\s\S]*?(?:\n\s*\n|$)/i, '')
+            .trim();
+    }
+
     private async restoreLastSession() {
         try {
             const lastSession = this.sessionManager.loadLastActiveSession();
@@ -394,12 +402,13 @@ export class AgentExecutor {
 
             // 5. Execute Actions
             const rationale = this.parseRationale(aiResponseText);
-            const assistantMessage: ChatMessage = { role: 'assistant', content: aiResponseText, internal: false, rationale };
+            const assistantContent = this.sanitizeAssistantContent(aiResponseText);
+            const assistantMessage: ChatMessage = { role: 'assistant', content: assistantContent, internal: false, rationale };
             this.chatHistory.push(assistantMessage);
             
             this.statusBarManager.updateStatus(AgentStatus.Executing);
             const report = await this.executeActions(aiResponseText, rootPath);
-            if (!aiResponseText.trim() && report.length === 0) {
+            if (!assistantContent.trim() && report.length === 0) {
                 this.chatHistory.pop();
                 logError('Model returned an empty response without actions.', { model: actualModel, engine, apiUrl, loopDepth });
                 this.webview.postMessage({
@@ -414,7 +423,7 @@ export class AgentExecutor {
                 return;
             }
 
-            if (report.length === 0 && loopDepth === 0 && this.isUnproductiveWaitingReply(aiResponseText)) {
+            if (report.length === 0 && loopDepth === 0 && this.isUnproductiveWaitingReply(assistantContent)) {
                 assistantMessage.internal = false;
                 const correctedReply = await this.buildUnproductiveReplyCorrection(prompt || '');
                 assistantMessage.content = correctedReply;
@@ -453,7 +462,7 @@ export class AgentExecutor {
 
             this.emitHistoryChanged();
             this.statusBarManager.updateStatus(AgentStatus.Success);
-            this.webview.postMessage({ type: 'streamChunk', value: aiResponseText });
+            this.webview.postMessage({ type: 'streamChunk', value: assistantContent });
 
         } catch (error: any) {
             this.statusBarManager.updateStatus(AgentStatus.Error, error.message);
@@ -505,7 +514,6 @@ export class AgentExecutor {
                 logError('Failed to load brain context for agents', ctxErr);
             }
 
-            const workflow = getConfig().multiAgentWorkflow;
             const selectedAgentContext = options.agentSkillContext
                 ? `\nSelected Agent Reference:\n${options.agentSkillContext}`
                 : '';
@@ -520,8 +528,7 @@ export class AgentExecutor {
                     this.webview?.postMessage({ type: 'autoContinue', value: `${step}: ${msg}` });
                     // 각 단계별 시작을 알림
                     this.webview?.postMessage({ type: 'streamChunk', value: `\n\n> **[${step}]** ${msg}\n\n` });
-                },
-                workflow
+                }
             );
 
             if (signal.aborted || !this.webview) return;
diff --git a/src/agents/AgentWorkflowManager.ts b/src/agents/AgentWorkflowManager.ts
index 5b56052..9e31b1c 100644
--- a/src/agents/AgentWorkflowManager.ts
+++ b/src/agents/AgentWorkflowManager.ts
@@ -1,5 +1,5 @@
-import { ConfigurableWorkflowAgent } from './factory';
-import { DEFAULT_MULTI_AGENT_WORKFLOW, MultiAgentStageConfig } from '../config';
+import { PlannerAgent, ResearcherAgent, WriterAgent } from './factory';
+import { AgentEngine, PipelineStage } from '../lib/engine';
 
 export class AgentWorkflowManager {
     /**
@@ -10,28 +10,24 @@ export class AgentWorkflowManager {
         modelName: string,
         brainContext: string,
         signal: AbortSignal,
-        onProgress: (step: string, message: string) => void,
-        workflow: MultiAgentStageConfig[] = DEFAULT_MULTI_AGENT_WORKFLOW
+        onProgress: (step: string, message: string) => void
     ): Promise<string> {
-        const stages = workflow.length > 0 ? workflow : DEFAULT_MULTI_AGENT_WORKFLOW;
-        const stageOutputs: string[] = [];
+        const planner = new PlannerAgent(modelName);
+        const researcher = new ResearcherAgent(modelName);
+        const writer = new WriterAgent(modelName);
+        const engine = new AgentEngine(planner, researcher, writer);
+        const missionId = `mission_${Date.now()}`;
+
         try {
-            for (const [index, stage] of stages.entries()) {
-                if (signal.aborted) throw new Error('AbortError');
-
-                const agent = new ConfigurableWorkflowAgent(modelName, stage);
-                const stepName = stage.name || `Stage ${index + 1}`;
-                onProgress(stepName, `Running ${index + 1}/${stages.length}`);
-
-                const stageInput = this.buildStageInput(stage, prompt, stageOutputs);
-                const result = await agent.execute(stageInput, brainContext, signal);
-                this.validateResult(result, stepName);
-                stageOutputs.push(result);
-
-                onProgress(stepName, `Completed ${index + 1}/${stages.length}`);
-            }
-
-            return stageOutputs[stageOutputs.length - 1];
+            return await engine.runMission(
+                missionId,
+                prompt,
+                brainContext,
+                signal,
+                (stage: PipelineStage, message: string) => {
+                    onProgress(this.mapStageToUI(stage), message);
+                }
+            );
         } catch (error: any) {
             if (error.name === 'AbortError' || error.message.includes('cancelled')) {
                 throw error;
@@ -40,28 +36,15 @@ export class AgentWorkflowManager {
         }
     }
 
-    private static buildStageInput(stage: MultiAgentStageConfig, originalPrompt: string, stageOutputs: string[]): string {
-        const previous = stageOutputs[stageOutputs.length - 1] || '';
-        if (stage.input === 'original') {
-            return originalPrompt;
-        }
-
-        if (stage.input === 'previous') {
-            return previous || originalPrompt;
-        }
-
-        return [
-            '## Original User Request',
-            originalPrompt,
-            '',
-            '## Previous Stage Output',
-            previous || 'No previous stage output.'
-        ].join('\n');
-    }
-
-    private static validateResult(data: string, step: string) {
-        if (!data || data.trim().length < 10) {
-            throw new Error(`${step} agent did not return a usable response.`);
-        }
+    private static mapStageToUI(stage: PipelineStage): string {
+        const maps: Record<PipelineStage, string> = {
+            idle: '대기',
+            planner: 'Planner',
+            researcher: 'Researcher',
+            writer: 'Writer',
+            completed: '완료',
+            error: '오류'
+        };
+        return maps[stage] || '진행 중';
     }
 }
diff --git a/src/agents/factory.ts b/src/agents/factory.ts
index d0d200e..b46c33a 100644
--- a/src/agents/factory.ts
+++ b/src/agents/factory.ts
@@ -1,5 +1,5 @@
 import * as vscode from 'vscode';
-import { MultiAgentStageConfig, getConfig } from '../config';
+import { getConfig } from '../config';
 
 export abstract class BaseAgent {
     constructor(protected readonly modelName: string) {}
@@ -145,29 +145,3 @@ Your goal is to produce a state-of-the-art final report that wows the user.
         return this.callLLM(this.persona, wrappedInput, signal);
     }
 }
-
-export class ConfigurableWorkflowAgent extends BaseAgent {
-    constructor(
-        modelName: string,
-        private readonly stage: MultiAgentStageConfig
-    ) {
-        super(modelName);
-    }
-
-    async execute(input: string, context?: string, signal?: AbortSignal): Promise<string> {
-        const wrappedInput = [
-            `### Stage: ${this.stage.name}`,
-            '',
-            '### Available Context',
-            context || 'No specific context available.',
-            '',
-            '### Stage Input',
-            input,
-            '',
-            '### Mission',
-            'Complete this stage according to your role instructions. Be concise, concrete, and preserve details needed by later stages.'
-        ].join('\n');
-
-        return this.callLLM(this.stage.prompt, wrappedInput, signal);
-    }
-}
diff --git a/src/config.ts b/src/config.ts
index 09729cb..58c9d17 100644
--- a/src/config.ts
+++ b/src/config.ts
@@ -12,15 +12,6 @@ export interface BrainProfile {
     description?: string;
 }
 
-export type MultiAgentStageInput = 'original' | 'previous' | 'combined';
-
-export interface MultiAgentStageConfig {
-    id: string;
-    name: string;
-    prompt: string;
-    input?: MultiAgentStageInput;
-}
-
 // ─── 에이전트 설정 인터페이스 (통합 버전) ───
 export interface IAgentConfig {
     ollamaUrl: string;
@@ -35,46 +26,12 @@ export interface IAgentConfig {
     maxAutoSteps: number;
     dryRun: boolean;
     multiAgentEnabled: boolean;
-    multiAgentWorkflow: MultiAgentStageConfig[];
     memoryEnabled: boolean;
     memoryShortTermMessages: number;
     memoryMediumTermSessions: number;
     memoryLongTermFiles: number;
 }
 
-export const DEFAULT_MULTI_AGENT_WORKFLOW: MultiAgentStageConfig[] = [
-    {
-        id: 'planner',
-        name: 'Planner',
-        input: 'original',
-        prompt: [
-            'You are the [Master Strategist & Planner].',
-            'Transform the user request into a precise execution blueprint.',
-            'Output Markdown with Objective, Core Challenges, Data Requirements, and Step-by-Step Tasks.'
-        ].join('\n')
-    },
-    {
-        id: 'researcher',
-        name: 'Researcher',
-        input: 'combined',
-        prompt: [
-            'You are the [Senior Technical Researcher].',
-            'Extract, filter, and synthesize high-signal facts from the plan and available context.',
-            'Output Key Facts, Technical Deep-Dive, Gaps, and Summary of Knowledge.'
-        ].join('\n')
-    },
-    {
-        id: 'writer',
-        name: 'Writer',
-        input: 'combined',
-        prompt: [
-            'You are the [Lead Synthesis Writer & Editor].',
-            'Produce the final response in the user language with a clear conclusion and actionable recommendation.',
-            'Preserve important technical nuance without padding.'
-        ].join('\n')
-    }
-];
-
 // ─── 경로 정규화 유틸리티 ───
 function normalizePath(p: string): string {
     if (!p) return p;
@@ -97,21 +54,6 @@ function toBrainProfile(raw: Partial<BrainProfile> | undefined, fallbackIndex: n
     };
 }
 
-function toMultiAgentStage(raw: Partial<MultiAgentStageConfig> | undefined, fallbackIndex: number): MultiAgentStageConfig | null {
-    if (!raw) return null;
-    const prompt = typeof raw.prompt === 'string' ? raw.prompt.trim() : '';
-    if (!prompt) return null;
-    const input = raw.input === 'original' || raw.input === 'previous' || raw.input === 'combined'
-        ? raw.input
-        : 'combined';
-    return {
-        id: (raw.id || `stage-${fallbackIndex + 1}`).trim(),
-        name: (raw.name || `Stage ${fallbackIndex + 1}`).trim(),
-        prompt,
-        input
-    };
-}
-
 // ─── VS Code 설정에서 읽어오는 값 (통합 구현) ───
 export function getConfig(): IAgentConfig {
     const cfg = vscode.workspace.getConfiguration('g1nation');
@@ -120,13 +62,9 @@ export function getConfig(): IAgentConfig {
     const legacyBrainPath = cfg.get<string>('localBrainPath', '');
     const legacyBrainRepo = cfg.get<string>('secondBrainRepo', '');
     const configuredProfiles = cfg.get<Partial<BrainProfile>[]>('brainProfiles', []);
-    const configuredWorkflow = cfg.get<Partial<MultiAgentStageConfig>[]>('multiAgentWorkflow', DEFAULT_MULTI_AGENT_WORKFLOW);
     const profiles = configuredProfiles
         .map((profile, index) => toBrainProfile(profile, index))
         .filter((profile): profile is BrainProfile => !!profile);
-    const multiAgentWorkflow = configuredWorkflow
-        .map((stage, index) => toMultiAgentStage(stage, index))
-        .filter((stage): stage is MultiAgentStageConfig => !!stage);
 
     if (profiles.length === 0) {
         const fallbackPath = normalizePath(legacyBrainPath) || path.join(os.homedir(), '.g1nation-brain');
@@ -144,15 +82,6 @@ export function getConfig(): IAgentConfig {
     const activeBrainId = cfg.get<string>('activeBrainId', profiles[0].id) || profiles[0].id;
     const activeBrain = profiles.find((profile) => profile.id === activeBrainId) || profiles[0];
 
-    const rationaleProtocol = `
-3. Always explain your thought process using the <rationale> tag BEFORE performing any actions. Use the following structure:
-<rationale>
-[PROBLEM] Description of the issue or need found in the context.
-[GOAL] What you intend to achieve with your proposed changes.
-[REASONING] Detailed logical basis for choosing specific actions or architecture.
-</rationale>
-`;
-
     return {
         ollamaUrl: cfg.get<string>('ollamaUrl', 'http://127.0.0.1:11434') || 'http://127.0.0.1:11434',
         defaultModel: cfg.get<string>('defaultModel', 'gemma4:e2b') || 'gemma4:e2b',
@@ -166,7 +95,6 @@ export function getConfig(): IAgentConfig {
         maxAutoSteps: cfg.get<number>('maxAutoSteps', 50),
         dryRun: cfg.get<boolean>('dryRun', false),
         multiAgentEnabled: cfg.get<boolean>('multiAgentEnabled', false),
-        multiAgentWorkflow: multiAgentWorkflow.length > 0 ? multiAgentWorkflow : DEFAULT_MULTI_AGENT_WORKFLOW,
         memoryEnabled: cfg.get<boolean>('memoryEnabled', true),
         memoryShortTermMessages: Math.max(0, cfg.get<number>('memoryShortTermMessages', 8)),
         memoryMediumTermSessions: Math.max(0, cfg.get<number>('memoryMediumTermSessions', 5)),
diff --git a/src/sidebarProvider.ts b/src/sidebarProvider.ts
index ce4ff69..446a1ae 100644
--- a/src/sidebarProvider.ts
+++ b/src/sidebarProvider.ts
@@ -1728,45 +1728,6 @@ export class SidebarChatProvider implements vscode.WebviewViewProvider, BridgeIn
         .step.active .step-label { color: var(--accent); }
         .step.complete .step-label { color: var(--success); }
 
-        /* --- Rationale View (Thought Process) --- */
-        .rationale-container {
-            margin: 12px 0;
-            padding: 16px;
-            background: rgba(255, 255, 255, 0.03);
-            border-left: 3px solid var(--accent);
-            border-radius: 4px 12px 12px 4px;
-            font-size: 12px;
-            line-height: 1.5;
-            animation: slideIn 0.3s ease-out;
-            backdrop-filter: blur(10px);
-        }
-        .rationale-header {
-            font-weight: 800;
-            text-transform: uppercase;
-            letter-spacing: 0.05em;
-            color: var(--accent);
-            margin-bottom: 12px;
-            display: flex;
-            align-items: center;
-            gap: 8px;
-            font-size: 11px;
-        }
-        .rationale-section {
-            margin-bottom: 10px;
-        }
-        .rationale-label {
-            display: flex;
-            align-items: center;
-            gap: 6px;
-            font-weight: 700;
-            color: var(--text-bright);
-            margin-bottom: 4px;
-            font-size: 11px;
-        }
-        .rationale-content {
-            color: var(--text-dim);
-            padding-left: 20px;
-        }
         @media (min-width: 360px) {
             .header-controls {
                 grid-template-columns: minmax(0, 1fr);
@@ -2080,24 +2041,6 @@ export class SidebarChatProvider implements vscode.WebviewViewProvider, BridgeIn
             msgEl.className = 'msg ' + (isUser ? 'msg-user' : 'msg-ai');
             msgEl._raw = text;
 
-            // If rationale exists and it's an AI message, add the Rationale View
-            if (!isUser && rationale && (rationale.problem || rationale.goal || rationale.reasoning)) {
-                const ratDiv = document.createElement('div');
-                ratDiv.className = 'rationale-container';
-                let ratHtml = '<div class="rationale-header"><span>🧠</span> Thought Process</div>';
-                if (rationale.problem) {
-                    ratHtml += '<div class="rationale-section"><div class="rationale-label"><span>⚠️</span> Problem</div><div class="rationale-content">' + rationale.problem + '</div></div>';
-                }
-                if (rationale.goal) {
-                    ratHtml += '<div class="rationale-section"><div class="rationale-label"><span>💡</span> Goal</div><div class="rationale-content">' + rationale.goal + '</div></div>';
-                }
-                if (rationale.reasoning) {
-                    ratHtml += '<div class="rationale-section"><div class="rationale-label"><span>✅</span> Rationale</div><div class="rationale-content">' + rationale.reasoning + '</div></div>';
-                }
-                ratDiv.innerHTML = ratHtml;
-                chat.appendChild(ratDiv);
-            }
-
             const head = document.createElement('div');
             head.className = 'msg-head';
             head.innerHTML = isUser ? '<div class="av av-user">U</div> You' : '<div class="av av-ai">✦</div> G1nation';
diff --git a/src/utils.ts b/src/utils.ts
index a22777f..4926e77 100644
--- a/src/utils.ts
+++ b/src/utils.ts
@@ -147,12 +147,9 @@ Core behavior:
 - Use the active Local Brain only when it is relevant to the user's question. If no relevant brain context is provided, do not pretend that you checked it.
 - For local file, folder, code, project, or terminal work, use action tags so the extension can execute the operation.
 - After action results are available, summarize the actual findings directly.
-- ALWAYS explain your thought process using the <rationale> tag BEFORE performing any actions. Use the following structure:
-<rationale>
-[PROBLEM] Description of the issue or need found in the context.
-[GOAL] What you intend to achieve with your proposed changes.
-[REASONING] Detailed logical basis for choosing specific actions or architecture.
-</rationale>
+- Do not output hidden reasoning labels such as [PROBLEM], [GOAL], [REASONING], Phase 0, Fidelity Lock-in, or process manifestos.
+- For substantial answers, write readable Markdown using ## and ### headings, short paragraphs, bullets, and tables where useful.
+- Avoid wall-of-text output. Make the answer scannable before adding detail.
 
 Available action tags: