elanous 0.2.0 → 0.2.5

package/skills/omni-digest/src/fetch.ts

@@ -0,0 +1,237 @@
+/** Content fetchers: Web (firecrawl → Jina fallback), GitHub (gh CLI), Local files */
+
+import { execSync } from 'node:child_process';
+import { readFileSync, existsSync, unlinkSync } from 'node:fs';
+import { join, dirname } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { tmpdir } from 'node:os';
+import { runOcr } from './ocr.js';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+
+/* ── Web: summarize CLI / firecrawl / Jina Reader ── */
+
+/** Check if summarize CLI is available */
+function hasSummarize(): boolean {
+  try { execSync('which summarize', { encoding: 'utf-8', timeout: 5000, stdio: 'pipe' }); return true; } catch { return false; }
+}
+
+/** Check if firecrawl CLI is available */
+function hasFirecrawl(): boolean {
+  try { execSync('firecrawl --version', { encoding: 'utf-8', timeout: 5000, stdio: 'pipe' }); return true; } catch { return false; }
+}
+
+/** Scrape via firecrawl CLI */
+function fetchWithFirecrawl(url: string): { title: string; content: string } | null {
+  console.log('  firecrawl scrape...');
+  const tmpFile = join(tmpdir(), `omni-digest-fc-${Date.now()}.md`);
+  try {
+    execSync(
+      `firecrawl scrape "${url}" --only-main-content -f markdown -o "${tmpFile}"`,
+      { encoding: 'utf-8', timeout: 60_000, stdio: 'pipe' },
+    );
+    if (!existsSync(tmpFile)) return null;
+    const text = readFileSync(tmpFile, 'utf-8').trim();
+    if (text.length < 30) return null;
+
+    let title = 'Web Article';
+    const hm = text.match(/^#\s+(.+)$/m);
+    if (hm) title = hm[1].trim();
+    else { const tm = text.match(/^Title:\s*(.+)$/mi); if (tm) title = tm[1].trim(); }
+
+    console.log(`  firecrawl 성공 (${text.length.toLocaleString()}자)`);
+    return { title, content: text };
+  } catch (e: any) {
+    console.log(`  firecrawl 실패: ${e.message?.split('\n')[0]}`);
+    return null;
+  } finally {
+    try { unlinkSync(tmpFile); } catch {}
+  }
+}
+
+/** Scrape via Jina Reader (fallback) */
+async function fetchWithJina(url: string): Promise<{ title: string; content: string }> {
+  console.log('  Jina Reader 폴백...');
+  const res = await fetch(`https://r.jina.ai/${url}`, { headers: { Accept: 'text/markdown', 'X-Return-Format': 'markdown' } });
+  if (!res.ok) throw new Error(`Jina Reader error: ${res.status}`);
+  const text = await res.text();
+  if (!text || text.trim().length < 30) throw new Error('웹 콘텐츠가 너무 짧음');
+  let title = 'Web Article';
+  const hm = text.match(/^#\s+(.+)$/m);
+  if (hm) title = hm[1].trim();
+  else { const tm = text.match(/^Title:\s*(.+)$/mi); if (tm) title = tm[1].trim(); }
+  return { title, content: text.trim() };
+}
+
+/** Extract content via summarize CLI (--extract --format md) */
+function fetchWithSummarizeCli(url: string): { title: string; content: string } | null {
+  console.log('  summarize CLI 추출...');
+  try {
+    const text = execSync(
+      `summarize "${url}" --extract --format md --lang auto`,
+      { encoding: 'utf-8', timeout: 120_000, stdio: ['pipe', 'pipe', 'pipe'] },
+    ).trim();
+    if (!text || text.length < 30) return null;
+
+    let title = 'Web Article';
+    const hm = text.match(/^#\s+(.+)$/m);
+    if (hm) title = hm[1].trim();
+    else { const tm = text.match(/^Title:\s*(.+)$/mi); if (tm) title = tm[1].trim(); }
+
+    console.log(`  summarize CLI 성공 (${text.length.toLocaleString()}자)`);
+    return { title, content: text };
+  } catch (e: any) {
+    console.log(`  summarize CLI 실패: ${e.message?.split('\n')[0]}`);
+    return null;
+  }
+}
+
+/** summarize CLI end-to-end: 추출 + LLM 요약까지 한번에 (Grok 스킵) */
+export function summarizeFullWithCli(url: string, length: string = 'xl'): { title: string; content: string } | null {
+  console.log(`  summarize CLI end-to-end (--length ${length})...`);
+  try {
+    const text = execSync(
+      `summarize "${url}" --length ${length} --lang auto --model auto`,
+      { encoding: 'utf-8', timeout: 180_000, stdio: ['pipe', 'pipe', 'pipe'] },
+    ).trim();
+    if (!text || text.length < 50) return null;
+
+    let title = 'Summary';
+    const hm = text.match(/^#\s+(.+)$/m);
+    if (hm) title = hm[1].trim();
+
+    console.log(`  summarize CLI end-to-end 성공 (${text.length.toLocaleString()}자)`);
+    return { title, content: text };
+  } catch (e: any) {
+    console.log(`  summarize CLI end-to-end 실패: ${e.message?.split('\n')[0]}`);
+    return null;
+  }
+}
+
+/**
+ * Fetch web content.
+ * preferSummarize=true ("sum" 키워드): summarize CLI → firecrawl → Jina
+ * preferSummarize=false (기본):        firecrawl → summarize CLI → Jina
+ */
+export async function fetchWebContent(url: string, preferSummarize = false): Promise<{ title: string; content: string }> {
+  if (preferSummarize && hasSummarize()) {
+    const result = fetchWithSummarizeCli(url);
+    if (result) return result;
+    // fallback to firecrawl → Jina
+  }
+
+  if (hasFirecrawl()) {
+    const result = fetchWithFirecrawl(url);
+    if (result) return result;
+  }
+
+  // Default fallback: try summarize CLI before Jina if not already tried
+  if (!preferSummarize && hasSummarize()) {
+    const result = fetchWithSummarizeCli(url);
+    if (result) return result;
+  }
+
+  return fetchWithJina(url);
+}
+
+/* ── GitHub: gh CLI ── */
+
+export function fetchGitHubRepo(owner: string, repo: string): { title: string; content: string } {
+  console.log(`  GitHub 레포 ${owner}/${repo}...`);
+  const meta = JSON.parse(execSync(`gh repo view ${owner}/${repo} --json name,description,primaryLanguage,stargazerCount,forkCount,updatedAt`, { encoding: 'utf-8', timeout: 30_000 }));
+  let readme = ''; try { readme = execSync(`gh api repos/${owner}/${repo}/readme --header "Accept: application/vnd.github.raw" 2>/dev/null`, { encoding: 'utf-8', timeout: 15_000 }); } catch {}
+  let commits = ''; try { commits = execSync(`gh api repos/${owner}/${repo}/commits?per_page=10 --jq '.[] | "- \\(.commit.message | split("\\n") | .[0]) (\\(.commit.author.date | .[0:10]))"'`, { encoding: 'utf-8', timeout: 15_000 }); } catch {}
+  const content = [`# ${meta.name}`, meta.description ? `> ${meta.description}` : '', '', `- Language: ${meta.primaryLanguage?.name || 'N/A'}`, `- Stars: ${meta.stargazerCount || 0} | Forks: ${meta.forkCount || 0}`, commits ? `## Recent Commits\n${commits}` : '', readme ? `## README\n${readme}` : ''].filter(Boolean).join('\n');
+  return { title: `${owner}/${repo}`, content };
+}
+
+export function fetchGitHubPR(owner: string, repo: string, number: string): { title: string; content: string } {
+  console.log(`  GitHub PR #${number}...`);
+  const data = JSON.parse(execSync(`gh pr view ${number} --repo ${owner}/${repo} --json title,body,state,author,createdAt,changedFiles,additions,deletions,files`, { encoding: 'utf-8', timeout: 30_000 }));
+  const files = (data.files || []).map((f: any) => `- ${f.path} (+${f.additions}/-${f.deletions})`).join('\n');
+  const content = [`# PR #${number}: ${data.title}`, `- Author: ${data.author?.login} | State: ${data.state} | +${data.additions}/-${data.deletions}`, data.body ? `## Description\n${data.body}` : '', files ? `## Changed Files\n${files}` : ''].filter(Boolean).join('\n');
+  return { title: `PR #${number}: ${data.title}`, content };
+}
+
+export function fetchGitHubIssue(owner: string, repo: string, number: string): { title: string; content: string } {
+  console.log(`  GitHub Issue #${number}...`);
+  const data = JSON.parse(execSync(`gh issue view ${number} --repo ${owner}/${repo} --json title,body,state,author,createdAt,labels,comments`, { encoding: 'utf-8', timeout: 30_000 }));
+  const labels = (data.labels || []).map((l: any) => l.name).join(', ');
+  const comments = (data.comments || []).slice(0, 10).map((c: any) => `### ${c.author?.login} (${c.createdAt?.slice(0, 10)})\n${c.body}`).join('\n\n');
+  const content = [`# Issue #${number}: ${data.title}`, `- Author: ${data.author?.login} | State: ${data.state}`, labels ? `- Labels: ${labels}` : '', data.body ? `## Description\n${data.body}` : '', comments ? `## Comments\n${comments}` : ''].filter(Boolean).join('\n');
+  return { title: `Issue #${number}: ${data.title}`, content };
+}
+
+export function fetchGitHubCommit(owner: string, repo: string, sha: string): { title: string; content: string } {
+  console.log(`  GitHub Commit ${sha.slice(0, 7)}...`);
+  const data = JSON.parse(execSync(`gh api repos/${owner}/${repo}/commits/${sha}`, { encoding: 'utf-8', timeout: 30_000 }));
+  const files = (data.files || []).map((f: any) => `- ${f.filename} (+${f.additions}/-${f.deletions})`).join('\n');
+  const content = [`# Commit ${sha.slice(0, 7)}`, `- Author: ${data.commit?.author?.name} | Date: ${data.commit?.author?.date?.slice(0, 10)}`, `## Message\n${data.commit?.message}`, files ? `## Changed Files\n${files}` : ''].filter(Boolean).join('\n');
+  return { title: `Commit: ${data.commit?.message?.split('\n')[0] || sha.slice(0, 7)}`, content };
+}
+
+/* ── omni-crawl 컨텍스트 보강 ── */
+
+const OMNI_CRAWL_SCRIPT = join(__dirname, '..', '..', 'omni-crawl', 'scripts', 'main.ts');
+
+/** omni-crawl을 호출하여 키워드 기반 배경 정보를 수집 */
+export function enrichWithOmniCrawl(query: string): string | null {
+  const scriptPath = OMNI_CRAWL_SCRIPT;
+  if (!existsSync(scriptPath)) {
+    console.log('  omni-crawl 스킬을 찾을 수 없음, 건너뜀');
+    return null;
+  }
+  console.log(`  omni-crawl 컨텍스트 보강: "${query}"...`);
+  try {
+    const output = execSync(
+      `npx tsx "${scriptPath}" "${query}" --no-save --print`,
+      { encoding: 'utf-8', timeout: 120_000, stdio: ['pipe', 'pipe', 'pipe'], cwd: join(__dirname, '..', '..', 'omni-crawl') },
+    );
+    const startMarker = '---BEGIN_OMNI_CRAWL_MARKDOWN---';
+    const endMarker = '---END_OMNI_CRAWL_MARKDOWN---';
+    const si = output.indexOf(startMarker);
+    const ei = output.indexOf(endMarker);
+    const crawled = (si >= 0 && ei > si) ? output.slice(si + startMarker.length, ei).trim() : output.trim();
+    if (crawled.length < 50) return null;
+    console.log(`  omni-crawl 보강 완료 (${crawled.length.toLocaleString()}자)`);
+    return crawled;
+  } catch (e: any) {
+    console.log(`  omni-crawl 보강 실패: ${e.message?.split('\n')[0]}`);
+    return null;
+  }
+}
+
+/* ── Local file ── */
+
+export async function fetchLocalFile(filePath: string): Promise<{ title: string; content: string }> {
+  console.log(`  로컬 파일 읽기... ${filePath}`);
+  if (!existsSync(filePath)) throw new Error(`파일 없음: ${filePath}`);
+  const ext = filePath.toLowerCase().split('.').pop() || '';
+  const baseName = filePath.split('/').pop() || filePath;
+
+  if (['txt', 'md', 'csv', 'json', 'xml', 'html', 'log'].includes(ext)) return { title: baseName, content: readFileSync(filePath, 'utf-8') };
+
+  if (ext === 'pdf') {
+    try { const t = execSync(`pdftotext "${filePath}" -`, { encoding: 'utf-8', timeout: 30_000 }); if (t.trim().length > 50) return { title: baseName, content: t }; } catch {}
+    console.log('  pdftotext 실패, OCR 시도...');
+    const ocr = await runOcr(filePath);
+    if (ocr) return { title: baseName, content: ocr };
+    return { title: baseName, content: `[PDF 텍스트 추출 실패. Claude Read 도구로 직접 읽어주세요: ${filePath}]` };
+  }
+
+  if (['png', 'jpg', 'jpeg', 'gif', 'bmp', 'tiff', 'webp'].includes(ext)) {
+    console.log('  이미지 OCR 처리...');
+    const ocr = await runOcr(filePath);
+    if (ocr) return { title: baseName, content: ocr };
+    return { title: baseName, content: `[이미지 OCR 실패. Claude Read 도구로 직접 읽어주세요: ${filePath}]` };
+  }
+
+  if (ext === 'docx' || ext === 'doc') {
+    try { const t = execSync(`textutil -convert txt -stdout "${filePath}"`, { encoding: 'utf-8', timeout: 15_000 }); if (t.trim().length > 30) return { title: baseName, content: t }; } catch {}
+    return { title: baseName, content: `[DOCX 추출 실패: ${filePath}]` };
+  }
+
+  try { return { title: baseName, content: readFileSync(filePath, 'utf-8') }; } catch {}
+  return { title: baseName, content: `[파일을 읽을 수 없습니다: ${filePath}]` };
+}

package/skills/omni-digest/src/media.ts

@@ -0,0 +1,180 @@
+/**
+ * 첨부 영상 다운로드 + STT 전사
+ *
+ * 엔진 우선순위:
+ *   1. 로컬 whisper CLI (무료·오프라인, `whisper` 설치 시)
+ *   2. youtube-master 의 transcribeChunks (ElevenLabs / OpenAI / Gemini 폴백 체인)
+ *
+ * X 는 media.variants 로 MP4 직링크를 주므로 yt-dlp 없이 curl 로 바로 받는다.
+ */
+
+import { execFile as execFileCb, execFileSync } from 'node:child_process';
+import { mkdir, readdir, readFile, rm } from 'node:fs/promises';
+import { existsSync } from 'node:fs';
+import { tmpdir } from 'node:os';
+import { join } from 'node:path';
+import { promisify } from 'node:util';
+import { env } from './env.js';
+import type { TweetMedia } from './types.js';
+
+const execFile = promisify(execFileCb);
+
+/** STT 대상으로 삼을 최대 재생 길이 (기본 20분) */
+const MAX_DURATION_MS = Number(env('OMNI_DIGEST_MEDIA_MAX_MS', String(20 * 60 * 1000)));
+/** 오디오 분할 단위 (초) — 긴 영상에서 API 업로드 한도 회피 */
+const CHUNK_SECONDS = 600;
+
+function has(bin: string): boolean {
+  try {
+    execFileSync('which', [bin], { stdio: 'ignore' });
+    return true;
+  } catch { return false; }
+}
+
+async function run(bin: string, args: string[]): Promise<string> {
+  const { stdout } = await execFile(bin, args, { maxBuffer: 32 * 1024 * 1024 });
+  return stdout;
+}
+
+/** MP4 직링크를 내려받는다. */
+export async function downloadVideo(url: string, outdir: string): Promise<string> {
+  await mkdir(outdir, { recursive: true });
+  const out = join(outdir, 'media.mp4');
+  await run('curl', ['-sL', '--fail', '-o', out, url]);
+  if (!existsSync(out)) throw new Error('영상 다운로드 실패');
+  return out;
+}
+
+/** 영상에서 16kHz 모노 mp3 를 추출한다 (STT 엔진 공통 입력 규격). */
+export async function extractAudio(videoPath: string, outdir: string): Promise<string> {
+  if (!has('ffmpeg')) throw new Error('ffmpeg 가 필요합니다 (brew install ffmpeg)');
+  const out = join(outdir, 'audio.mp3');
+  await run('ffmpeg', [
+    '-hide_banner', '-loglevel', 'error', '-y',
+    '-i', videoPath,
+    '-vn', '-ar', '16000', '-ac', '1', '-c:a', 'libmp3lame', '-q:a', '4',
+    out,
+  ]);
+  if (!existsSync(out)) throw new Error('오디오 추출 실패');
+  return out;
+}
+
+/** 로컬 whisper CLI 전사. 미설치면 null 을 반환해 폴백을 유도한다. */
+async function transcribeLocalWhisper(audioPath: string, outdir: string, lang: string): Promise<string | null> {
+  if (!has('whisper')) return null;
+  const model = env('OMNI_DIGEST_WHISPER_MODEL', 'small');
+  try {
+    await run('whisper', [
+      audioPath,
+      '--model', model,
+      '--language', lang,
+      '--output_format', 'txt',
+      '--output_dir', outdir,
+    ]);
+    const txt = (await readdir(outdir)).find((f: string) => f.endsWith('.txt'));
+    if (!txt) return null;
+    const text = (await readFile(join(outdir, txt), 'utf-8')).trim();
+    return text || null;
+  } catch {
+    return null;
+  }
+}
+
+/** youtube-master 의 STT 체인(ElevenLabs → OpenAI → Gemini)으로 폴백. */
+async function transcribeViaYoutubeMaster(audioPath: string, outdir: string): Promise<string | null> {
+  const skillRoot = env('YOUTUBE_MASTER_ROOT', join(process.env.HOME || '', '.claude/skills/youtube-master'));
+  const script = join(skillRoot, 'scripts/transcribe-local.ts');
+  if (!existsSync(script)) return null;
+
+  // transcribe-local.ts 는 chunk_%03d.mp3 가 든 디렉터리를 입력으로 받는다
+  const chunksDir = join(outdir, 'chunks');
+  await mkdir(chunksDir, { recursive: true });
+  await run('ffmpeg', [
+    '-hide_banner', '-loglevel', 'error', '-y',
+    '-i', audioPath,
+    '-f', 'segment', '-segment_time', String(CHUNK_SECONDS),
+    '-c', 'copy', join(chunksDir, 'chunk_%03d.mp3'),
+  ]);
+
+  const outFile = join(outdir, 'transcript.txt');
+  try {
+    await run('npx', ['tsx', script, chunksDir, outFile]);
+    const text = (await readFile(outFile, 'utf-8')).trim();
+    return text || null;
+  } catch {
+    return null;
+  }
+}
+
+export interface MediaTranscript {
+  transcript: string;
+  engine: 'local-whisper' | 'youtube-master-stt';
+  durationMs?: number;
+}
+
+/**
+ * 트윗 첨부 영상을 받아 전사한다.
+ * 영상이 없거나 너무 길거나 모든 엔진이 실패하면 null (요약은 그대로 진행).
+ */
+/** 오디오 스트림이 있나 — X 데모 영상은 무음이 많고, 그때 ffmpeg -vn 은 «Invalid argument» 로 죽는다. 못 재면 있다고 본다. */
+async function hasAudioStream(videoPath: string): Promise<boolean> {
+  try {
+    const { stdout } = await execFile('ffprobe', ['-v', 'error', '-select_streams', 'a', '-show_entries', 'stream=index', '-of', 'csv=p=0', videoPath]);
+    return stdout.trim().length > 0;
+  } catch { return true; }
+}
+
+export async function transcribeTweetMedia(
+  media: TweetMedia[],
+  opts: { lang?: string } = {},
+): Promise<MediaTranscript | null> {
+  const video = media.find(m => (m.type === 'video' || m.type === 'animated_gif') && (m.sttMp4Url || m.bestMp4Url));
+  // 전사는 오디오만 쓰므로 최저 화질 변형을 받는다 (최고 화질은 수백 MB가 될 수 있음)
+  const srcUrl = video?.sttMp4Url || video?.bestMp4Url;
+  if (!video || !srcUrl) return null;
+
+  if (video.durationMs && video.durationMs > MAX_DURATION_MS) {
+    console.log(`  영상 ${Math.round(video.durationMs / 1000)}s — 상한(${Math.round(MAX_DURATION_MS / 1000)}s) 초과로 전사 생략`);
+    return null;
+  }
+
+  const workdir = join(tmpdir(), `omni-digest-media-${video.mediaKey}`);
+  try {
+    console.log(`  영상 다운로드 (${video.durationMs ? Math.round(video.durationMs / 1000) + 's' : '길이 미상'}, STT용 저화질 변형)...`);
+    const videoPath = await downloadVideo(srcUrl, workdir);
+    if (!(await hasAudioStream(videoPath))) {
+      console.log('  소리 트랙 없음(무음 영상) — 전사 생략, 화면은 시각 흡수가 본다');
+      return null;
+    }
+    const audioPath = await extractAudio(videoPath, workdir);
+
+    const lang = opts.lang || env('OMNI_DIGEST_STT_LANG', 'en');
+    console.log(`  STT 전사 중 (lang=${lang})...`);
+
+    const local = await transcribeLocalWhisper(audioPath, workdir, lang);
+    if (local) {
+      console.log(`  전사 완료 (local-whisper, ${local.length}자)`);
+      return { transcript: local, engine: 'local-whisper', durationMs: video.durationMs };
+    }
+
+    const remote = await transcribeViaYoutubeMaster(audioPath, workdir);
+    if (remote) {
+      console.log(`  전사 완료 (youtube-master STT, ${remote.length}자)`);
+      return { transcript: remote, engine: 'youtube-master-stt', durationMs: video.durationMs };
+    }
+
+    console.log('  전사 실패 — 모든 STT 엔진 사용 불가. 텍스트만으로 요약합니다.');
+    return null;
+  } catch (e: any) {
+    console.log(`  미디어 처리 실패 (${e.message}) — 텍스트만으로 요약합니다.`);
+    return null;
+  } finally {
+    await rm(workdir, { recursive: true, force: true }).catch(() => {});
+  }
+}
+
+/** 전사문을 Obsidian 부록용 블록으로 만든다. */
+export function transcriptAppendix(t: MediaTranscript): string {
+  const dur = t.durationMs ? ` (${Math.round(t.durationMs / 1000)}초)` : '';
+  return `# 영상 전사${dur}\n> 엔진: ${t.engine}\n\n${t.transcript}`;
+}

package/skills/omni-digest/src/obsidian.ts

@@ -0,0 +1,69 @@
+import { mkdir, writeFile } from 'node:fs/promises';
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { env } from './env.js';
+import { classifyDomain, stagingEnabled, stagingSubdir } from './domain-staging.js';
+import { safeName, dateStamp, nowISO, nowFull, escapeQuotes } from './util.js';
+import type { SourceType } from './types.js';
+
+const SUBDIR: Record<string, string> = {
+  'x-post': '00. Inbox/03. X Summary', 'x-article': '00. Inbox/03. X Summary',
+  youtube: '00. Inbox/02. Youtube Summary', web: '00. Inbox/04. Web Digest',
+  'github-repo': '00. Inbox/06. GitHub Digest', 'github-pr': '00. Inbox/06. GitHub Digest',
+  'github-issue': '00. Inbox/06. GitHub Digest', 'github-commit': '00. Inbox/06. GitHub Digest',
+  'local-file': '00. Inbox/07. Document Digest',
+};
+
+export function getSaveDir(sourceType: SourceType): string | null {
+  const root = env('OBSIDIAN_VAULT_ROOT');
+  if (!root) return null;
+  const sub = env('OMNI_DIGEST_SAVE_SUBDIR') || SUBDIR[sourceType] || '00. Inbox/08. OmniDigest';
+  return join(root, ...sub.split('/'));
+}
+
+export interface SaveOpts {
+  title: string; sourceType: SourceType; sourceUrl?: string;
+  summaryBody: string; genre: string; keywords: string[];
+  saveDir: string;
+  /** X post-specific extra frontmatter */
+  extraFrontmatter?: string;
+  /** Append after body */
+  appendix?: string;
+}
+
+export async function saveMarkdown(opts: SaveOpts): Promise<string> {
+  // 도메인 스테이징 라우팅 — title/genre/keywords/sourceType 을 domain 분류 → 00. Inbox/_staging/<domain>/
+  //   (토글 OBSIDIAN_STAGING_BY_DOMAIN·단일지점이라 4개 소스 호출부 전부 커버). off 면 opts.saveDir(레거시).
+  const domain = classifyDomain([opts.title, opts.genre, ...opts.keywords, opts.sourceType]);
+  const root = env('OBSIDIAN_VAULT_ROOT');
+  const saveDir = stagingEnabled() && root ? join(root, ...stagingSubdir(domain).split('/')) : opts.saveDir;
+  await mkdir(saveDir, { recursive: true });
+  const base = `${dateStamp()}_${safeName(opts.title)}`;
+  let fp = join(saveDir, `${base}.md`); let seq = 1;
+  while (existsSync(fp)) { fp = join(saveDir, `${base}_${seq}.md`); seq++; }
+
+  const genreTags = opts.genre.split('#').map(g => g.trim().replace(/\s+/g, '_')).filter(Boolean);
+  const kwTags = opts.keywords.map(k => k.replace(/\s+/g, '_'));
+  const srcTag = opts.sourceType.replace(/-/g, '_');
+  const allTags = ['OmniDigest', srcTag, 'AI요약', ...genreTags, ...kwTags];
+
+  const fm = [
+    '---',
+    `title: "${escapeQuotes(opts.title)}"`,
+    `created: ${nowISO()}`,
+    `category: "${domain}"`,
+    'tags:', ...allTags.map(t => `  - "${t}"`),
+    `source_type: "${opts.sourceType}"`,
+    opts.sourceUrl ? `source_url: "${opts.sourceUrl}"` : '',
+    opts.extraFrontmatter || '',
+    `genre: "${opts.genre}"`,
+    `keywords: "${opts.keywords.join(', ')}"`,
+    'ai_generated: true',
+    `generated_at: "${nowFull()}"`,
+    '---',
+  ].filter(Boolean).join('\n');
+
+  const content = [fm, '', opts.summaryBody, '', '---', '', opts.appendix || '', opts.sourceUrl ? `> 원문: ${opts.sourceUrl}` : ''].filter(l => l !== undefined).join('\n');
+  await writeFile(fp, content, 'utf-8');
+  return fp;
+}

package/skills/omni-digest/src/ocr.ts

@@ -0,0 +1,149 @@
+/** Upstage Document OCR — absorbed from photo-intake-ocr (Python → TypeScript) */
+
+import { readFileSync, existsSync } from 'node:fs';
+import { execSync } from 'node:child_process';
+import { resolve } from 'node:path';
+import { env } from './env.js';
+
+const API_URL = 'https://api.upstage.ai/v1/document-digitization';
+
+/** Resolve Upstage API key: env → cache file → zshrc */
+function getUpstageApiKey(): string {
+  let key = env('UPSTAGE_API_KEY');
+  if (key) return key;
+
+  const cachePath = resolve(process.env.HOME || '~', '.cache', 'upstage_api_key');
+  if (existsSync(cachePath)) {
+    key = readFileSync(cachePath, 'utf-8').trim();
+    if (key) return key;
+  }
+
+  try {
+    key = execSync('zsh -lc "source ~/.zshrc >/dev/null 2>&1; printenv UPSTAGE_API_KEY"', { encoding: 'utf-8', timeout: 5000 }).trim();
+    if (key) return key;
+  } catch { /* ignore */ }
+
+  throw new Error('UPSTAGE_API_KEY를 찾을 수 없습니다. .env, ~/.cache/upstage_api_key, 또는 ~/.zshrc에 설정하세요.');
+}
+
+/** Guess MIME type from extension */
+function guessMime(filePath: string): string {
+  const ext = filePath.toLowerCase().split('.').pop() || '';
+  const map: Record<string, string> = {
+    png: 'image/png', jpg: 'image/jpeg', jpeg: 'image/jpeg', gif: 'image/gif',
+    bmp: 'image/bmp', tiff: 'image/tiff', webp: 'image/webp',
+    pdf: 'application/pdf',
+  };
+  return map[ext] || 'application/octet-stream';
+}
+
+export interface OcrResult {
+  text: string;
+  json?: any;
+}
+
+/**
+ * Run Upstage Document OCR on a local file.
+ * Returns extracted text (and optionally full JSON).
+ */
+export async function upstageOcr(filePath: string, opts?: { textOnly?: boolean; model?: string }): Promise<OcrResult> {
+  const apiKey = getUpstageApiKey();
+  const model = opts?.model || 'ocr';
+  const fileName = filePath.split('/').pop() || 'file';
+  const mimeType = guessMime(filePath);
+  const fileBytes = readFileSync(filePath);
+
+  const boundary = '----OmniDigestUpstageOCRBoundary';
+  const bodyParts = [
+    `--${boundary}\r\nContent-Disposition: form-data; name="model"\r\n\r\n${model}\r\n`,
+    `--${boundary}\r\nContent-Disposition: form-data; name="document"; filename="${fileName}"\r\nContent-Type: ${mimeType}\r\n\r\n`,
+  ];
+
+  const textEncoder = new TextEncoder();
+  const parts: Uint8Array[] = [
+    textEncoder.encode(bodyParts[0]),
+    textEncoder.encode(bodyParts[1]),
+    fileBytes,
+    textEncoder.encode(`\r\n--${boundary}--\r\n`),
+  ];
+
+  const totalLen = parts.reduce((s, p) => s + p.length, 0);
+  const body = new Uint8Array(totalLen);
+  let offset = 0;
+  for (const p of parts) { body.set(p, offset); offset += p.length; }
+
+  const res = await fetch(API_URL, {
+    method: 'POST',
+    headers: {
+      Authorization: `Bearer ${apiKey}`,
+      'Content-Type': `multipart/form-data; boundary=${boundary}`,
+    },
+    body,
+  });
+
+  if (!res.ok) {
+    const errText = await res.text().catch(() => '');
+    throw new Error(`Upstage OCR error: ${res.status} ${errText}`);
+  }
+
+  const payload = await res.json();
+  const text = payload.text || '';
+
+  if (opts?.textOnly) return { text };
+  return { text, json: payload };
+}
+
+/**
+ * Fallback OCR via OCR.space API (for when Upstage is unavailable).
+ */
+export async function ocrSpaceFallback(filePath: string): Promise<string> {
+  const apiKey = env('OCR_API_KEY');
+  if (!apiKey) throw new Error('OCR_API_KEY 없음');
+
+  const fileBytes = readFileSync(filePath);
+  const fileName = filePath.split('/').pop() || 'file';
+
+  const formData = new FormData();
+  formData.append('file', new Blob([fileBytes]), fileName);
+  formData.append('language', 'kor');
+  formData.append('isOverlayRequired', 'false');
+  formData.append('OCREngine', '2');
+
+  const res = await fetch('https://api.ocr.space/parse/image', {
+    method: 'POST',
+    headers: { apikey: apiKey },
+    body: formData,
+  });
+  if (!res.ok) throw new Error(`OCR.space error: ${res.status}`);
+  const data = await res.json();
+  return data.ParsedResults?.map((r: any) => r.ParsedText).join('\n') || '';
+}
+
+/**
+ * Run OCR with automatic fallback: Upstage → OCR.space
+ */
+export async function runOcr(filePath: string): Promise<string> {
+  // Try Upstage first
+  try {
+    const result = await upstageOcr(filePath, { textOnly: true });
+    if (result.text.trim().length > 10) {
+      console.log(`  Upstage OCR 성공 (${result.text.length.toLocaleString()}자)`);
+      return result.text;
+    }
+  } catch (e: any) {
+    console.log(`  Upstage OCR 실패: ${e.message}`);
+  }
+
+  // Try OCR.space fallback
+  try {
+    const text = await ocrSpaceFallback(filePath);
+    if (text.trim().length > 10) {
+      console.log(`  OCR.space 폴백 성공 (${text.length.toLocaleString()}자)`);
+      return text;
+    }
+  } catch (e: any) {
+    console.log(`  OCR.space 폴백 실패: ${e.message}`);
+  }
+
+  return '';
+}