stellavault 0.4.0 → 0.4.1

package/packages/core/dist/mcp/tools/generate-draft.d.ts

@@ -0,0 +1,34 @@
+import type { SearchEngine } from '../../search/index.js';
+export declare function createGenerateDraftTool(searchEngine: SearchEngine, vaultPath: string): {
+    name: string;
+    description: string;
+    inputSchema: {
+        type: "object";
+        properties: {
+            topic: {
+                type: string;
+                description: string;
+            };
+            format: {
+                type: string;
+                enum: string[];
+                description: string;
+            };
+            maxSources: {
+                type: string;
+                description: string;
+            };
+        };
+    };
+    handler: (args: {
+        topic?: string;
+        format?: string;
+        maxSources?: number;
+    }) => Promise<{
+        content: {
+            type: "text";
+            text: string;
+        }[];
+    }>;
+};
+//# sourceMappingURL=generate-draft.d.ts.map

package/packages/core/dist/mcp/tools/generate-draft.js

@@ -0,0 +1,120 @@
+// MCP Tool: generate-draft — Claude Code에서 직접 초안 생성
+// Claude가 vault 컨텍스트를 받아 직접 글을 쓰도록 구조화된 재료를 반환
+// 비용: $0 (이미 Claude Code 세션 안)
+import { scanRawDirectory, extractConcepts } from '../../intelligence/wiki-compiler.js';
+import { loadConfig, DEFAULT_FOLDERS } from '../../config.js';
+import { resolve } from 'node:path';
+import { existsSync } from 'node:fs';
+export function createGenerateDraftTool(searchEngine, vaultPath) {
+    return {
+        name: 'generate-draft',
+        description: 'Gather knowledge from your vault to write a draft. Returns structured context (notes, concepts, excerpts) so you can compose a blog post, report, or outline. Use this when the user asks you to "write a draft", "blog post from my notes", or "summarize my knowledge about X".',
+        inputSchema: {
+            type: 'object',
+            properties: {
+                topic: {
+                    type: 'string',
+                    description: 'Topic or keyword to focus on. Leave empty for all knowledge.',
+                },
+                format: {
+                    type: 'string',
+                    enum: ['blog', 'report', 'outline'],
+                    description: 'Desired output format (default: blog)',
+                },
+                maxSources: {
+                    type: 'number',
+                    description: 'Maximum number of source documents to include (default: 10)',
+                },
+            },
+        },
+        handler: async (args) => {
+            const { topic, format = 'blog', maxSources = 10 } = args;
+            const config = loadConfig();
+            const folders = config.folders ?? DEFAULT_FOLDERS;
+            // 1. 토픽 기반 검색 (있으면)
+            let searchResults = [];
+            if (topic) {
+                const results = await searchEngine.search({ query: topic, limit: maxSources });
+                searchResults = results.map(r => ({
+                    title: r.document?.title ?? 'Untitled',
+                    content: r.chunk?.content ?? '',
+                    filePath: r.document?.filePath ?? '',
+                    score: r.score ?? 0,
+                }));
+            }
+            // 2. vault 전체 스캔 (검색 결과가 부족하면)
+            const allDocs = [];
+            for (const dir of [folders.fleeting, folders.literature, folders.permanent, folders.wiki]) {
+                const fullDir = resolve(vaultPath, dir);
+                if (existsSync(fullDir)) {
+                    const docs = scanRawDirectory(fullDir);
+                    allDocs.push(...docs);
+                }
+            }
+            // 토픽 필터
+            const relevantDocs = topic
+                ? allDocs.filter(d => d.title.toLowerCase().includes(topic.toLowerCase()) ||
+                    d.tags.some(t => t.toLowerCase().includes(topic.toLowerCase())) ||
+                    d.content.toLowerCase().includes(topic.toLowerCase()))
+                : allDocs;
+            // 3. 개념 추출
+            const concepts = extractConcepts(relevantDocs.length > 0 ? relevantDocs : allDocs);
+            const topConcepts = [...concepts.entries()]
+                .sort((a, b) => b[1].length - a[1].length)
+                .slice(0, 8)
+                .map(([name, docs]) => ({ name, documentCount: docs.length }));
+            // 4. 발췌 준비
+            const excerpts = (searchResults.length > 0 ? searchResults : relevantDocs.slice(0, maxSources))
+                .map(doc => {
+                const body = doc.content
+                    .replace(/^---[\s\S]*?---\n?/, '') // frontmatter 제거
+                    .replace(/^#+\s+.+\n/m, '') // 첫 heading 제거
+                    .trim()
+                    .slice(0, 500);
+                return {
+                    title: doc.title,
+                    excerpt: body,
+                    filePath: doc.filePath,
+                };
+            })
+                .filter(e => e.excerpt.length > 20);
+            // 5. 구조화된 컨텍스트 반환
+            const context = {
+                topic: topic ?? 'All Knowledge',
+                format,
+                totalDocuments: relevantDocs.length || allDocs.length,
+                concepts: topConcepts,
+                sources: excerpts,
+                instruction: `Use the sources above to write a ${format}. Rules:
+- Only use information from the provided sources
+- Cite sources using [[title]] wikilink format
+- Write in the same language as the source material
+- Add your analysis connecting ideas across sources
+- Format: ${format === 'blog' ? 'Engaging blog post with intro, body sections, conclusion' : format === 'report' ? 'Formal report with executive summary, sections, conclusion' : 'Hierarchical outline with numbered sections'}`,
+            };
+            const text = `# Draft Context: ${context.topic}
+
+## Format: ${format}
+## Sources: ${context.totalDocuments} documents, ${excerpts.length} excerpts
+
+## Key Concepts
+${topConcepts.map(c => `- **${c.name}** (${c.documentCount} documents)`).join('\n')}
+
+## Source Excerpts
+
+${excerpts.map(e => `### ${e.title}
+> ${e.excerpt}
+`).join('\n')}
+
+## Instructions
+${context.instruction}
+
+---
+Please write the ${format} draft based on the context above. Save the result to the vault's _drafts/ folder.`;
+            return {
+                content: [{ type: 'text', text }],
+            };
+        },
+    };
+}
+//# sourceMappingURL=generate-draft.js.map

package/packages/core/package.json

@@ -13,6 +13,14 @@
     "./intelligence/youtube-extractor": {
       "import": "./dist/intelligence/youtube-extractor.js",
       "types": "./dist/intelligence/youtube-extractor.d.ts"
+    },
+    "./intelligence/file-extractors": {
+      "import": "./dist/intelligence/file-extractors.js",
+      "types": "./dist/intelligence/file-extractors.d.ts"
+    },
+    "./intelligence/draft-generator": {
+      "import": "./dist/intelligence/draft-generator.js",
+      "types": "./dist/intelligence/draft-generator.d.ts"
     }
   },
   "scripts": {
@@ -30,7 +38,9 @@
     "vitest": "^3.0.0"
   },
   "dependencies": {
+    "@anthropic-ai/sdk": "^0.82.0",
     "@modelcontextprotocol/sdk": "^1.28.0",
+    "@types/multer": "^2.1.0",
     "@xenova/transformers": "^2.17.2",
     "b4a": "^1.8.0",
     "better-sqlite3": "^12.8.0",
@@ -39,6 +49,11 @@
     "express": "^5.2.1",
     "gray-matter": "^4.0.3",
     "hyperswarm": "^4.17.0",
-    "sqlite-vec": "^0.1.7"
+    "mammoth": "^1.12.0",
+    "multer": "^2.1.1",
+    "officeparser": "^6.0.7",
+    "sqlite-vec": "^0.1.7",
+    "unpdf": "^1.4.0",
+    "xlsx": "^0.18.5"
   }
 }