idea-manager 0.4.0 → 0.5.0

package/src/lib/ai/cleanup.ts

@@ -1,87 +0,0 @@
-import { runClaude, extractJson, type IStructuredItem } from './client';
-import { replaceItems, getItemTree } from '../db/queries/items';
-import type { IItemTree } from '@/types';
-
-function serializeItems(items: IItemTree[], depth = 0): string {
-  const lines: string[] = [];
-  for (const item of items) {
-    const indent = '  '.repeat(depth);
-    const status = item.status || 'pending';
-    lines.push(`${indent}- [${item.item_type}/${item.priority}/${status}] ${item.title}: ${item.description || ''}`);
-    if (item.children && item.children.length > 0) {
-      lines.push(serializeItems(item.children, depth + 1));
-    }
-  }
-  return lines.join('\n');
-}
-
-function countItems(items: IItemTree[]): number {
-  let count = 0;
-  for (const item of items) {
-    count++;
-    if (item.children) count += countItems(item.children);
-  }
-  return count;
-}
-
-function mapToDbFormat(items: IStructuredItem[]): Parameters<typeof replaceItems>[2] {
-  return items.map((item) => ({
-    parent_id: null,
-    title: item.title,
-    description: item.description,
-    item_type: item.item_type,
-    priority: item.priority,
-    status: item.status,
-    children: item.children ? mapToDbFormat(item.children) : undefined,
-  }));
-}
-
-export async function cleanupItems(
-  projectId: string,
-  brainstormId: string,
-  items: IItemTree[],
-  brainstormContent: string,
-): Promise<{ items: IItemTree[]; changed: boolean }> {
-  const serialized = serializeItems(items);
-  const beforeCount = countItems(items);
-
-  const prompt = `You are a JSON-only deduplication machine. You NEVER respond with text, explanations, or conversation.
-You ALWAYS output ONLY a raw JSON array, nothing else.
-
-Your job: clean up the structured item tree below by removing duplicates and merging similar items.
-
-Schema per item:
-{ "title": string, "description": string, "item_type": "feature"|"task"|"bug"|"idea"|"note", "priority": "high"|"medium"|"low", "status": "pending"|"in_progress"|"done", "children": [same schema] }
-
-Rules:
-- Output MUST start with [ and end with ]
-- No markdown fences, no explanation, no text before or after the JSON
-- MERGE items that describe the same concept (combine their descriptions, keep the more specific title)
-- REMOVE exact or near-exact duplicates (keep the one with more detail)
-- PRESERVE the status of items — if one copy is "done" and another is "pending", keep "done"
-- PRESERVE the hierarchy — keep parent-child relationships logical
-- Keep titles concise (under 50 chars)
-- Do NOT add new items that weren't in the original
-- Do NOT remove items just because they seem unimportant — only remove TRUE duplicates
-- If the brainstorming context is provided, use it to understand which items are actually the same concept
-
-${brainstormContent ? `사용자의 브레인스토밍 메모:\n${brainstormContent}\n\n` : ''}현재 구조화된 항목 (중복 제거 및 병합하세요):
-${serialized}`;
-
-  const resultText = await runClaude(prompt);
-  const json = extractJson(resultText, 'array');
-  const cleaned = JSON.parse(json) as IStructuredItem[];
-
-  const afterCount = cleaned.reduce((sum, item) => sum + 1 + countStructuredChildren(item), 0);
-  const changed = afterCount !== beforeCount;
-
-  const dbItems = mapToDbFormat(cleaned);
-  const tree = replaceItems(projectId, brainstormId, dbItems);
-
-  return { items: tree, changed };
-}
-
-function countStructuredChildren(item: IStructuredItem): number {
-  if (!item.children) return 0;
-  return item.children.reduce((sum, child) => sum + 1 + countStructuredChildren(child), 0);
-}

package/src/lib/ai/prompter.ts

@@ -1,78 +0,0 @@
-import { runClaude } from './client';
-import { getPrompt, createPrompt } from '../db/queries/prompts';
-import { getBrainstorm } from '../db/queries/brainstorms';
-import { getRecentConversations } from '../db/queries/conversations';
-import { getProjectContextSummary } from '../db/queries/context';
-import type { IPrompt, IItem } from '@/types';
-
-export async function generatePrompt(
-  item: IItem,
-  projectContext?: { brainstormContent?: string; conversationHistory?: string; projectDocs?: string },
-): Promise<IPrompt> {
-  const existing = getPrompt(item.id);
-  if (existing?.prompt_type === 'manual') {
-    return existing;
-  }
-
-  const systemPrompt = `You are a prompt engineering expert. Generate a clear, actionable prompt for a coding assistant (like Cursor or Claude Code) to implement the given task.
-
-Rules:
-- Output ONLY the prompt text, nothing else
-- Write in Korean
-- Be specific and actionable
-- Include conditions, constraints, and requirements
-- Include relevant context from the brainstorming, conversation, and project documentation
-- The prompt should be ready to paste into a coding tool
-- Keep it concise but complete (under 500 chars)
-- Do NOT include markdown fences or extra formatting`;
-
-  const CONTEXT_LIMIT = 30_000; // 30KB max for prompt generation context
-  let context = '';
-  if (projectContext?.brainstormContent) {
-    context += `\n\n브레인스토밍 원문:\n${projectContext.brainstormContent.slice(0, 3000)}`;
-  }
-  if (projectContext?.conversationHistory) {
-    context += `\n\nAI 대화 이력:\n${projectContext.conversationHistory.slice(0, 5000)}`;
-  }
-  if (projectContext?.projectDocs) {
-    const remaining = CONTEXT_LIMIT - context.length;
-    if (remaining > 1000) {
-      context += `\n\n프로젝트 문서:\n${projectContext.projectDocs.slice(0, remaining)}`;
-    }
-  }
-
-  const prompt = `${systemPrompt}\n\n다음 항목에 대한 실행 프롬프트를 생성하세요:
-
-제목: ${item.title}
-설명: ${item.description}
-유형: ${item.item_type}
-우선순위: ${item.priority}${context}`;
-
-  const resultText = await runClaude(prompt);
-
-  return createPrompt({
-    project_id: item.project_id,
-    item_id: item.id,
-    content: resultText.trim(),
-    prompt_type: 'auto',
-  });
-}
-
-export async function generatePromptForItem(
-  item: IItem,
-): Promise<IPrompt> {
-  const brainstorm = getBrainstorm(item.project_id);
-  const conversations = getRecentConversations(item.project_id, 20);
-
-  const conversationHistory = conversations.length > 0
-    ? conversations.map(c => `${c.role === 'user' ? '사용자' : 'AI'}: ${c.content}`).join('\n')
-    : undefined;
-
-  const projectDocs = getProjectContextSummary(item.project_id) || undefined;
-
-  return generatePrompt(item, {
-    brainstormContent: brainstorm?.content,
-    conversationHistory,
-    projectDocs,
-  });
-}

package/src/lib/ai/refiner.ts

@@ -1,128 +0,0 @@
-import { runClaude } from './client';
-import { updateItem, addChildItems, getItemTree, deleteItem } from '../db/queries/items';
-import { getBrainstorm } from '../db/queries/brainstorms';
-import { getDb } from '../db/index';
-import type { IItem, IItemTree } from '@/types';
-
-interface RefineChild {
-  title: string;
-  description: string;
-  item_type: 'feature' | 'task' | 'bug' | 'idea' | 'note';
-  priority: 'high' | 'medium' | 'low';
-  children?: RefineChild[];
-}
-
-interface RefineResult {
-  title: string;
-  description: string;
-  children?: RefineChild[];
-  remove_children?: boolean;
-}
-
-export async function refineItem(
-  item: IItem,
-  userMessage: string,
-): Promise<{ title: string; description: string; tree: IItemTree[] }> {
-  const brainstorm = getBrainstorm(item.project_id);
-
-  // Get existing children info
-  const db = getDb();
-  const existingChildren = db.prepare(
-    'SELECT id, title, description, item_type, priority FROM items WHERE parent_id = ?'
-  ).all(item.id) as Pick<IItem, 'id' | 'title' | 'description' | 'item_type' | 'priority'>[];
-
-  const childrenInfo = existingChildren.length > 0
-    ? `\n\n현재 하위 항목 (${existingChildren.length}개):\n${existingChildren.map((c, i) => `${i + 1}. [${c.item_type}] ${c.title}: ${c.description}`).join('\n')}`
-    : '\n\n현재 하위 항목: 없음';
-
-  const systemPrompt = `You are a task refinement assistant. The user wants to modify a task item and potentially its sub-items.
-You ALWAYS output ONLY a raw JSON object, nothing else.
-
-Output schema:
-{
-  "title": string,
-  "description": string,
-  "children": [{ "title": string, "description": string, "item_type": "feature"|"task"|"bug"|"idea"|"note", "priority": "high"|"medium"|"low", "children": [...] }],
-  "remove_children": boolean
-}
-
-Rules:
-- Output MUST be a JSON object
-- No markdown fences, no explanation, no text before or after the JSON
-- Keep titles concise (under 50 chars)
-- Write in Korean
-- description should be detailed and actionable
-- "children" is OPTIONAL: include it only if the user asks to add/restructure sub-items
-- "remove_children" is OPTIONAL: set true only if the user explicitly asks to remove existing children before adding new ones
-- If the user just wants to modify the title/description, omit "children"
-- If the user asks to break down, detail, or expand the item, provide "children" array with sub-items`;
-
-  const context = brainstorm?.content ? `\n\n브레인스토밍 원문:\n${brainstorm.content}` : '';
-
-  const prompt = `${systemPrompt}\n\n현재 항목:
-제목: ${item.title}
-설명: ${item.description}
-유형: ${item.item_type}
-우선순위: ${item.priority}${childrenInfo}${context}
-
-사용자 요청: ${userMessage}`;
-
-  const resultText = await runClaude(prompt);
-
-  const cleaned = resultText.replace(/```(?:json)?\s*/g, '').replace(/```\s*/g, '').trim();
-  const jsonMatch = cleaned.match(/\{[\s\S]*\}/);
-  if (!jsonMatch) {
-    throw new Error('AI did not return valid JSON');
-  }
-
-  const parsed = JSON.parse(jsonMatch[0]) as RefineResult;
-
-  // Update the item itself
-  updateItem(item.id, {
-    title: parsed.title,
-    description: parsed.description,
-  });
-
-  // Handle children changes
-  if (parsed.children && parsed.children.length > 0) {
-    // If remove_children is set, delete existing children first
-    if (parsed.remove_children && existingChildren.length > 0) {
-      for (const child of existingChildren) {
-        deleteItem(child.id);
-      }
-    }
-
-    // Add new children
-    addChildItems(item.project_id, item.id, parsed.children.map(c => ({
-      parent_id: item.id,
-      title: c.title,
-      description: c.description,
-      item_type: c.item_type,
-      priority: c.priority,
-      children: c.children?.map(mapChild) ?? undefined,
-    })));
-  }
-
-  // Return updated tree
-  const tree = getItemTree(item.project_id);
-
-  return { title: parsed.title, description: parsed.description, tree };
-}
-
-function mapChild(c: RefineChild): {
-  parent_id: null;
-  title: string;
-  description: string;
-  item_type: RefineChild['item_type'];
-  priority: RefineChild['priority'];
-  children?: ReturnType<typeof mapChild>[];
-} {
-  return {
-    parent_id: null,
-    title: c.title,
-    description: c.description,
-    item_type: c.item_type,
-    priority: c.priority,
-    children: c.children?.map(mapChild),
-  };
-}

package/src/lib/ai/structurer.ts

@@ -1,403 +0,0 @@
-import { runStructure, runStructureWithQuestions, runAnalysis, runClaude, extractJson, type IStructuredItem, type OnTextChunk, type OnRawEvent } from './client';
-import { replaceItems, appendItems, getItemTree } from '../db/queries/items';
-import { getRecentConversations, addMessage } from '../db/queries/conversations';
-import {
-  getProjectContextSummary,
-  getProjectContextsBySubProject,
-  buildSubProjectSummary,
-} from '../db/queries/context';
-import { resolveMemos, createMemosFromQuestions } from '../db/queries/memos';
-import type { IItemTree, IMemo, IConversation } from '@/types';
-
-const AI_CONTEXT_LIMIT = 150_000; // 150KB - threshold for chunking
-const PHASE3_CONTEXT_LIMIT = 300_000; // 300KB - Phase 3 final structuring (hub doc can be large)
-const AI_CHUNK_LIMIT = 80_000; // 80KB - max context per AI call
-
-export async function structureBrainstorm(
-  projectId: string,
-  brainstormId: string,
-  content: string,
-): Promise<IItemTree[]> {
-  if (!content.trim()) {
-    return [];
-  }
-
-  const projectContext = getProjectContextSummary(projectId) || undefined;
-  const structured = await runStructure(content, projectContext);
-
-  const dbItems = mapToDbFormat(structured);
-
-  return replaceItems(projectId, brainstormId, dbItems);
-}
-
-export async function structureWithChat(
-  projectId: string,
-  brainstormId: string,
-  content: string,
-): Promise<{ items: IItemTree[]; memos: IMemo[]; message: IConversation | null }> {
-  // Always use single mode for auto-structuring (triggered by brainstorming edits).
-  // Multi-agent analysis is only used via streaming endpoint (structureWithChatDirect).
-  const projectContext = getProjectContextSummary(projectId) || undefined;
-  const safeContext = projectContext ? truncateContext(projectContext, AI_CONTEXT_LIMIT) : undefined;
-  return structureSingle(projectId, brainstormId, content, safeContext);
-}
-
-/**
- * Streaming structure with direct SSE callback.
- * Instead of async generator, takes a `send` callback to emit SSE events directly.
- * This avoids timing issues with generator yield + async queue.
- */
-export async function structureWithChatDirect(
-  projectId: string,
-  brainstormId: string,
-  content: string,
-  send: (event: string, data: unknown) => void | Promise<void>,
-): Promise<void> {
-  const projectContext = getProjectContextSummary(projectId) || undefined;
-  const contextSize = projectContext?.length || 0;
-
-  await send('status', { message: '컨텍스트 크기 확인 중...' });
-
-  const onText: OnTextChunk = (text) => {
-    send('ai_text', { text });
-  };
-
-  const onRawEvent: OnRawEvent = (event) => {
-    send('ai_event', event);
-  };
-
-  if (contextSize <= AI_CONTEXT_LIMIT) {
-    await send('status', { message: 'AI 구조화 중...', mode: 'single' });
-
-    const history = getRecentConversations(projectId, 20);
-    const historyForAi = history.map(h => ({ role: h.role, content: h.content }));
-    const safeContext = projectContext ? truncateContext(projectContext, AI_CONTEXT_LIMIT) : undefined;
-
-    const existingItems = getItemTree(projectId);
-    const existingContext = existingItems.length > 0
-      ? serializeExistingItems(existingItems)
-      : undefined;
-
-    const result = await runStructureWithQuestions(content, historyForAi, safeContext, onText, onRawEvent, existingContext);
-
-    const dbItems = mapToDbFormat(result.items as IStructuredItem[]);
-    const tree = replaceItems(projectId, brainstormId, dbItems);
-    resolveMemos(projectId);
-
-    let aiMessage: IConversation | null = null;
-    let memos: IMemo[] = [];
-    if (result.questions.length > 0) {
-      const messageContent = result.questions
-        .map((q, i) => `${i + 1}. ${q.question}`)
-        .join('\n');
-      aiMessage = addMessage(projectId, 'assistant', messageContent);
-      memos = createMemosFromQuestions(projectId, aiMessage.id, result.questions);
-    }
-
-    await send('done', { items: tree, memos, message: aiMessage });
-    return;
-  }
-
-  // ============================================================
-  // 3-Phase Multi-Agent Analysis
-  // ============================================================
-  const subProjects = getProjectContextsBySubProject(projectId);
-  const brainstormContext = content.trim()
-    ? `\n\n사용자의 브레인스토밍 메모:\n${content}`
-    : '';
-
-  // Send phase list to frontend
-  await send('phase_list', {
-    phases: [
-      { name: '전체 아키텍처 분석', status: 'pending' },
-      { name: `서브 프로젝트 병렬 분석 (${subProjects.length}개)`, status: 'pending' },
-      { name: '최종 구조화', status: 'pending' },
-    ],
-  });
-
-  // Send sub-project list
-  await send('chunk_list', {
-    chunks: subProjects.map((c, i) => ({
-      name: c.name,
-      index: i + 1,
-      fileCount: c.contexts.length,
-    })),
-    total: subProjects.length,
-  });
-
-  // ----------------------------------------------------------
-  // Phase 1: Build hub document from docs/configs
-  // ----------------------------------------------------------
-  await send('phase_update', { index: 0, status: 'active' });
-  await send('status', { message: 'Phase 1: 문서/설정 기반 아키텍처 분석 중...' });
-
-  // Collect root files + all docs/configs for overview
-  const rootSub = subProjects.find(s => s.name === '(root)');
-  const rootContext = rootSub ? truncateContext(buildSubProjectSummary(rootSub), AI_CHUNK_LIMIT) : '';
-
-  // Also collect file listing for all sub-projects
-  const projectFileTree = subProjects
-    .map(s => `[${s.name}] (${s.contexts.length}개 파일)\n${s.contexts.map(c => `  ${c.file_path}`).join('\n')}`)
-    .join('\n\n');
-
-  const phase1Prompt = `당신은 소프트웨어 프로젝트 아키텍트입니다. 아래 프로젝트의 문서와 설정 파일을 분석하여 "프로젝트 개요 문서"를 작성하세요.
-
-이 문서는 다른 AI 에이전트들이 각 서브 프로젝트를 분석할 때 참조하는 중추 문서가 됩니다.
-
-다음 내용을 포함해주세요:
-1. 프로젝트 전체 목적과 구조
-2. 기술 스택 (프레임워크, 언어, 주요 라이브러리)
-3. 서브 프로젝트 간의 관계와 의존성
-4. 아키텍처 패턴 (모노레포, 마이크로서비스, 등)
-5. 주요 컨벤션과 규칙
-6. 배포/인프라 구조 (파악 가능한 경우)
-
-한국어로 작성하세요. Markdown 형식으로 작성하세요.
-${brainstormContext}
-
-=== 프로젝트 파일 트리 ===
-${projectFileTree}
-
-=== 루트 문서/설정 파일 ===
-${rootContext}`;
-
-  let hubDocument = '';
-  try {
-    hubDocument = await runAnalysis(phase1Prompt, onText, onRawEvent);
-  } catch (err) {
-    const errMsg = err instanceof Error ? err.message : String(err);
-    hubDocument = `# 프로젝트 개요 (자동 생성 실패)\n오류: ${errMsg.slice(0, 300)}\n\n## 서브 프로젝트 목록\n${subProjects.map(s => `- ${s.name}`).join('\n')}`;
-  }
-
-  await send('phase_update', { index: 0, status: 'done' });
-  await send('hub_document', { content: hubDocument });
-
-  // ----------------------------------------------------------
-  // Phase 2: Parallel sub-project analysis
-  // ----------------------------------------------------------
-  await send('phase_update', { index: 1, status: 'active' });
-  await send('status', { message: 'Phase 2: 서브 프로젝트 병렬 분석 중...' });
-
-  const CONCURRENCY = 2;
-  const subAnalyses = new Map<string, string>();
-  const nonRootSubs = subProjects.filter(s => s.name !== '(root)');
-
-  // Process in batches of CONCURRENCY
-  for (let batch = 0; batch < nonRootSubs.length; batch += CONCURRENCY) {
-    const batchSubs = nonRootSubs.slice(batch, batch + CONCURRENCY);
-
-    const batchPromises = batchSubs.map(async (sub, batchIdx) => {
-      const globalIdx = batch + batchIdx;
-      const subIdx = subProjects.indexOf(sub);
-
-      await send('structuring_sub', {
-        subProject: sub.name,
-        current: globalIdx + 1,
-        total: nonRootSubs.length,
-        files: sub.contexts.map(c => c.file_path),
-      });
-
-      const subContext = truncateContext(buildSubProjectSummary(sub), AI_CHUNK_LIMIT);
-      const phase2Prompt = `당신은 소프트웨어 분석가입니다. 아래 "프로젝트 개요 문서"를 참조하여 서브 프로젝트 "${sub.name}"의 소스코드를 분석하세요.
-
-분석 결과를 다음 형식으로 작성하세요:
-1. **역할**: 이 서브 프로젝트가 전체 시스템에서 하는 역할
-2. **주요 기능**: 구현된 핵심 기능 목록 (각 기능의 구현 상태 포함)
-3. **기술 스택**: 사용 중인 기술/라이브러리
-4. **구현 상태**: done/in_progress/pending 판단 근거
-5. **TODO/개선점**: 코드에서 발견된 TODO, 미구현 부분, 개선 가능 사항
-6. **다른 서브 프로젝트와의 관계**: 의존하거나 의존받는 프로젝트
-
-한국어로 작성하세요. Markdown 형식으로 작성하세요.
-
-=== 프로젝트 개요 문서 (중추) ===
-${truncateContext(hubDocument, 30_000)}
-
-=== ${sub.name} 소스코드 ===
-${subContext}`;
-
-      try {
-        // Each sub gets its own text stream tagged with sub name
-        const subOnText: OnTextChunk = (text) => {
-          send('ai_text', { text, subProject: sub.name });
-        };
-
-        const analysis = await runAnalysis(phase2Prompt, subOnText);
-        subAnalyses.set(sub.name, analysis);
-
-        await send('structuring_sub_done', {
-          subProject: sub.name,
-          current: globalIdx + 1,
-          total: nonRootSubs.length,
-          itemCount: 0,
-        });
-      } catch (err) {
-        const errMsg = err instanceof Error ? err.message : String(err);
-        console.error(`[structurer] Phase 2 "${sub.name}" failed:`, errMsg);
-        subAnalyses.set(sub.name, `# ${sub.name} (분석 실패)\n오류: ${errMsg.slice(0, 300)}`);
-
-        await send('structuring_sub_done', {
-          subProject: sub.name,
-          current: globalIdx + 1,
-          total: nonRootSubs.length,
-          error: errMsg.slice(0, 200),
-        });
-      }
-    });
-
-    await Promise.all(batchPromises);
-  }
-
-  // Build complete hub document with all analyses
-  const completeDocument = `${hubDocument}\n\n---\n\n# 서브 프로젝트 상세 분석\n\n${
-    Array.from(subAnalyses.entries())
-      .map(([name, analysis]) => `## ${name}\n\n${analysis}`)
-      .join('\n\n---\n\n')
-  }`;
-
-  await send('phase_update', { index: 1, status: 'done' });
-  await send('hub_document', { content: completeDocument });
-
-  // ----------------------------------------------------------
-  // Phase 3: Final structuring from complete hub document
-  // ----------------------------------------------------------
-  await send('phase_update', { index: 2, status: 'active' });
-  await send('status', { message: 'Phase 3: 중추 문서 기반 최종 구조화 중...' });
-  await send('ai_text_reset', {});
-
-  const phase3Prompt = `You are a JSON-only structuring machine. You NEVER respond with text, explanations, or conversation.
-You ALWAYS output ONLY a raw JSON array, nothing else.
-
-Your job: convert the comprehensive project analysis document below into a structured JSON tree.
-
-Schema per item:
-{ "title": string, "description": string, "item_type": "feature"|"task"|"bug"|"idea"|"note", "priority": "high"|"medium"|"low", "status": "pending"|"in_progress"|"done", "children": [same schema] }
-
-Rules:
-- Output MUST start with [ and end with ]
-- No markdown fences, no explanation, no text before or after the JSON
-- Top-level items should be sub-projects (one per analyzed project)
-- Each top-level item should have children representing features/tasks/bugs found in that sub-project
-- Keep titles concise (under 50 chars)
-- Judge status based on the analysis:
-  - "done": fully implemented as described
-  - "in_progress": partially implemented or has TODOs
-  - "pending": not yet started or only planned
-- Prioritize items that have TODOs or are in_progress as "high" priority
-- Include bugs, improvements, and missing features mentioned in the analysis
-${brainstormContext}
-
-=== 프로젝트 분석 문서 ===
-${truncateContext(completeDocument, PHASE3_CONTEXT_LIMIT)}`;
-
-  try {
-    const resultText = await runClaude(phase3Prompt, onText, onRawEvent);
-    const json = extractJson(resultText, 'array');
-    const structured = JSON.parse(json) as IStructuredItem[];
-
-    const dbItems = mapToDbFormat(structured);
-    const tree = appendItems(projectId, brainstormId, dbItems);
-    resolveMemos(projectId);
-
-    const summaryMsg = addMessage(
-      projectId,
-      'assistant',
-      `3단계 분석 완료: ${nonRootSubs.length}개 서브 프로젝트를 병렬 분석 후 구조화했습니다.`,
-    );
-
-    await send('phase_update', { index: 2, status: 'done' });
-    await send('done', { items: tree, memos: [], message: summaryMsg });
-  } catch (err) {
-    const errMsg = err instanceof Error ? err.message : String(err);
-    console.error('[structurer] Phase 3 failed:', errMsg);
-
-    await send('phase_update', { index: 2, status: 'error' });
-    await send('error', { error: `Phase 3 구조화 실패: ${errMsg.slice(0, 300)}` });
-  }
-}
-
-// ============================================================
-// Internal helpers
-// ============================================================
-
-async function structureSingle(
-  projectId: string,
-  brainstormId: string,
-  content: string,
-  projectContext?: string,
-): Promise<{ items: IItemTree[]; memos: IMemo[]; message: IConversation | null }> {
-  const history = getRecentConversations(projectId, 20);
-  const historyForAi = history.map(h => ({
-    role: h.role,
-    content: h.content,
-  }));
-
-  const existingItems = getItemTree(projectId);
-  const existingContext = existingItems.length > 0
-    ? serializeExistingItems(existingItems)
-    : undefined;
-
-  const safeContext = projectContext ? truncateContext(projectContext, AI_CONTEXT_LIMIT) : undefined;
-  const result = await runStructureWithQuestions(content, historyForAi, safeContext, undefined, undefined, existingContext);
-
-  const dbItems = mapToDbFormat(result.items as IStructuredItem[]);
-  const tree = replaceItems(projectId, brainstormId, dbItems);
-
-  resolveMemos(projectId);
-
-  let aiMessage: IConversation | null = null;
-  let memos: IMemo[] = [];
-
-  if (result.questions.length > 0) {
-    const messageContent = result.questions
-      .map((q, i) => `${i + 1}. ${q.question}`)
-      .join('\n');
-
-    aiMessage = addMessage(projectId, 'assistant', messageContent);
-    memos = createMemosFromQuestions(projectId, aiMessage.id, result.questions);
-  }
-
-  return { items: tree, memos, message: aiMessage };
-}
-
-function serializeExistingItems(items: IItemTree[], depth = 0): string {
-  if (items.length === 0) return '';
-  const lines: string[] = [];
-  for (const item of items) {
-    const indent = '  '.repeat(depth);
-    lines.push(`${indent}- [${item.item_type}/${item.priority}] ${item.title}: ${item.description || ''}`);
-    if (item.children && item.children.length > 0) {
-      lines.push(serializeExistingItems(item.children, depth + 1));
-    }
-  }
-  return lines.join('\n');
-}
-
-function truncateContext(context: string, limit: number): string {
-  if (context.length <= limit) return context;
-
-  const fileSections = context.split(/(?=--- .+ ---\n)/);
-  let result = '';
-
-  for (const section of fileSections) {
-    if (result.length + section.length > limit) {
-      result += `\n\n--- (${fileSections.length - result.split('---').length / 2}개 파일 생략됨, 컨텍스트 크기 제한) ---\n`;
-      break;
-    }
-    result += section;
-  }
-
-  return result;
-}
-
-function mapToDbFormat(items: IStructuredItem[]): Parameters<typeof replaceItems>[2] {
-  return items.map((item) => ({
-    parent_id: null,
-    title: item.title,
-    description: item.description,
-    item_type: item.item_type,
-    priority: item.priority,
-    status: item.status,
-    children: item.children ? mapToDbFormat(item.children) : undefined,
-  }));
-}