tuna-agent 0.1.155 → 0.1.157

package/dist/daemon/analyze-video-handler.js

@@ -116,6 +116,85 @@ function run(cmd, args, opts = {}) {
         p.on('close', (code) => code === 0 ? resolve({ out, err }) : reject(new Error(`${cmd} exit ${code}: ${err.slice(0, 500)}`)));
     });
 }
+// Modern Chrome UA — TikTok/Douyin reject unknown user-agents.
+const SRC_UA = 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36';
+// Optional cookies for auth-gated sources (Facebook private/page videos,
+// some Douyin). YT_DLP_COOKIES = path to a Netscape cookies.txt; or
+// YT_DLP_COOKIES_FROM_BROWSER = a browser name yt-dlp can read cookies from.
+function cookieArgs() {
+    if (process.env.YT_DLP_COOKIES)
+        return ['--cookies', process.env.YT_DLP_COOKIES];
+    if (process.env.YT_DLP_COOKIES_FROM_BROWSER)
+        return ['--cookies-from-browser', process.env.YT_DLP_COOKIES_FROM_BROWSER];
+    return [];
+}
+function detectPlatform(url) {
+    let h = '';
+    try {
+        h = new URL(url).hostname.toLowerCase();
+    }
+    catch { /* malformed → 'other' */ }
+    if (/(^|\.)youtube\.com$|(^|\.)youtu\.be$/.test(h))
+        return 'youtube';
+    if (/(^|\.)tiktok\.com$/.test(h))
+        return 'tiktok';
+    if (/douyin\.com$|iesdouyin\.com$/.test(h))
+        return 'douyin';
+    if (/facebook\.com$|fb\.watch$|(^|\.)fb\.com$/.test(h))
+        return 'facebook';
+    return 'other';
+}
+// yt-dlp `%(title)s` is a clean title on YouTube but the full post caption
+// on TikTok/Douyin/Facebook (often 200+ chars, prefixed with engagement
+// stats). Strip the social-stats prefix, keep the first segment, and cap
+// to a sane title length so the cloned idea doesn't get a paragraph title.
+function cleanSourceTitle(raw) {
+    let t = (raw || '').trim().split('\n')[0].trim();
+    // Drop leading "1K views · 51 reactions | " / "230 likes, 12 comments - " noise.
+    t = t.replace(/^\s*(?:[\d.,]+\s*[KMB]?\s*(?:views?|likes?|reactions?|comments?|shares?|followers?)\b\s*[·,|–\-:]*\s*)+/i, '').trim();
+    // FB/TikTok pack "title | extra | hashtags" — keep the first real segment.
+    const seg = t.split(/\s*[|·]\s*/)[0].trim();
+    if (seg.length >= 8)
+        t = seg;
+    if (t.length > 90)
+        t = t.slice(0, 90).replace(/\s+\S*$/, '').trim() + '…';
+    return t.slice(0, 120);
+}
+// Download a source video across YouTube / TikTok / Douyin / Facebook.
+// yt-dlp supports all of them, but a single rigid `-f` that works for
+// YouTube fails on the others, so try a tolerant 720p-capped format then
+// fall back to letting yt-dlp pick. UA + optional cookies harden the
+// non-YouTube extractors (FB private/page + Douyin need cookies).
+async function downloadSourceVideo(url, dest) {
+    const platform = detectPlatform(url);
+    const common = [
+        '--no-playlist', '--no-warnings', '--retries', '3', '--fragment-retries', '3',
+        // dest ends in .mp4 — force merged/odd containers to mp4 so the file
+        // lands EXACTLY at `dest` (else fs.rename → ENOENT, e.g. YouTube shorts
+        // where bv*+ba merges to .mkv/.webm).
+        '--merge-output-format', 'mp4', '--remux-video', 'mp4',
+        '--user-agent', SRC_UA, ...cookieArgs(), '-o', dest, url,
+    ];
+    const attempts = [
+        ['-f', 'bv*[height<=720]+ba/b[height<=720]/best', ...common],
+        ['-f', 'best/mp4', ...common], // let yt-dlp choose (TikTok/Douyin/FB quirks)
+    ];
+    let lastErr;
+    for (let i = 0; i < attempts.length; i++) {
+        try {
+            await run(YT_DLP, attempts[i]);
+            return;
+        }
+        catch (e) {
+            lastErr = e;
+            console.warn(`[analyze_video] yt-dlp attempt ${i + 1}/${attempts.length} failed (${platform}): ${String(e?.message || e).slice(0, 220)}`);
+        }
+    }
+    const hint = (platform === 'facebook' || platform === 'douyin')
+        ? ' — FB private/page & some Douyin need cookies (set YT_DLP_COOKIES)'
+        : '';
+    throw new Error(`yt-dlp failed for ${platform} after ${attempts.length} attempts${hint}: ${String(lastErr?.message || lastErr).slice(0, 300)}`);
+}
 async function whisperTranscribe(audioPath) {
     if (!OPENAI_KEY)
         throw new Error('OPENAI_API_KEY not set');
@@ -450,7 +529,7 @@ export async function analyzeVideo(url, onProgress) {
             // analyze of the same URL never reads a half-written file.
             const dlTmp = path.join(CACHE_DIR, `${urlHash}.dl-${crypto.randomBytes(4).toString('hex')}.mp4`);
             try {
-                await run(YT_DLP, ['-f', 'best[height<=720]/best', '-o', dlTmp, '--no-playlist', '--quiet', url]);
+                await downloadSourceVideo(url, dlTmp);
                 await fs.rename(dlTmp, videoPath);
             }
             catch (e) {
@@ -462,8 +541,8 @@ export async function analyzeVideo(url, onProgress) {
         // clone idea gets a real name instead of "Clone: www.youtube.com".
         let source_title = '';
         try {
-            const t = await run(YT_DLP, ['--skip-download', '--no-warnings', '--no-playlist', '--print', '%(title)s', url]);
-            source_title = (t.out || '').trim().split('\n')[0].slice(0, 200);
+            const t = await run(YT_DLP, ['--skip-download', '--no-warnings', '--no-playlist', '--user-agent', SRC_UA, ...cookieArgs(), '--print', '%(title)s', url]);
+            source_title = cleanSourceTitle(t.out);
         }
         catch { /* title is best-effort — analysis still proceeds without it */ }
         progress('Đang tách audio...');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tuna-agent",
-  "version": "0.1.155",
+  "version": "0.1.157",
   "description": "Tuna Agent - Run AI coding tasks on your machine",
   "bin": {
     "tuna-agent": "dist/cli/index.js"