diff --git a/tiktok2bsky.py b/tiktok2bsky.py
index b6ad65d..66d6e8b 100644
--- a/tiktok2bsky.py
+++ b/tiktok2bsky.py
@@ -44,7 +44,6 @@ _STEALTH_SYNC = None   # will hold the stealth_sync callable if v1.x is present
 try:
     from playwright_stealth import stealth_sync as _stealth_sync_import
     _STEALTH_SYNC = _stealth_sync_import
-    logging.getLogger(__name__).debug("playwright-stealth v1.x detected (stealth_sync)")
 except ImportError:
     # v2.x is installed but its API is too unstable to use reliably —
     # browser launch args provide equivalent protection for our use case
@@ -494,7 +493,7 @@ def compress_video(
             )
             return False
 
-        trim_to    = min(duration, max_duration)
+        trim_to     = min(duration, max_duration)
         target_bits = max_size_bytes * 8 * 0.85
         total_kbps  = int(target_bits / trim_to / 1000)
         audio_kbps  = 96
@@ -553,24 +552,18 @@ def compress_video(
 # ─────────────────────────────────────────────────────────────────────────────
 #  yt-dlp helpers
 # ─────────────────────────────────────────────────────────────────────────────
-def get_best_impersonation_target() -> str | None:
+def get_best_impersonation_target():
     """
     Ask yt-dlp directly which impersonation targets are actually available
-    in the current environment. This is the only reliable method —
-    curl_cffi's BrowserType enum values change between versions and do not
-    map 1:1 to yt-dlp's target names.
+    in the current environment. Returns the best ImpersonateTarget object,
+    or None if none are available.
 
-    Returns the best available target string, or None if none are available.
+    This is the only reliable method — curl_cffi's BrowserType enum values
+    change between versions and do not map 1:1 to yt-dlp's target names.
     """
     try:
         import yt_dlp
-        # yt-dlp exposes available impersonation targets via
-        # ImpersonateTarget.supported_targets() in newer builds,
-        # or via YoutubeDL._impersonate_target_key in older ones.
-        # The safest cross-version approach is to instantiate a YoutubeDL
-        # object with quiet=True and inspect _impersonate_targets.
         with yt_dlp.YoutubeDL({"quiet": True, "no_warnings": True}) as ydl:
-            # _impersonate_targets is a dict of {ImpersonateTarget: handler}
             targets = getattr(ydl, "_impersonate_targets", None)
             if not targets:
                 logging.warning(
@@ -578,11 +571,8 @@ def get_best_impersonation_target() -> str | None:
                 )
                 return None
 
-            # Convert to string representations and pick the best one
-            preferred = ["chrome", "safari", "firefox", "edge"]
             available_strs = []
             for t in targets.keys():
-                # ImpersonateTarget has .client and optionally .version
                 client  = getattr(t, "client", None) or str(t)
                 version = getattr(t, "version", None)
                 label   = f"{client}-{version}" if version else str(client)
@@ -593,7 +583,7 @@ def get_best_impersonation_target() -> str | None:
                 f"{[s for s, _ in available_strs]}"
             )
 
-            # Pick highest-versioned chrome first, then others
+            # Prefer highest-versioned chrome, then anything else
             chrome_targets = sorted(
                 [(s, t) for s, t in available_strs if "chrome" in s],
                 key=lambda x: x[0],
@@ -602,9 +592,8 @@ def get_best_impersonation_target() -> str | None:
             if chrome_targets:
                 best_label, best_target = chrome_targets[0]
                 logging.info(f"🎭 Selected impersonation target: {best_label}")
-                return best_target   # return the actual ImpersonateTarget object
+                return best_target
 
-            # Fallback to any available target
             best_label, best_target = available_strs[0]
             logging.info(f"🎭 Selected impersonation target (fallback): {best_label}")
             return best_target
@@ -617,11 +606,76 @@ def get_best_impersonation_target() -> str | None:
     return None
 
 
+def fetch_video_metadata_ytdlp(
+    url: str,
+    netscape_cookies_path: str = None,
+) -> dict:
+    """
+    Fetch metadata (title, description, timestamp, uploader) for a single
+    TikTok video URL using yt-dlp without downloading the video file.
+
+    TikTok captions (the text the creator wrote) live in the 'description'
+    field of yt-dlp's info dict. 'title' is a shorter auto-generated label.
+
+    Returns a dict with keys: description, title, timestamp, uploader.
+    All values default to empty string / None on failure.
+    """
+    import yt_dlp
+
+    impersonate = get_best_impersonation_target()
+
+    ydl_opts = {
+        "quiet":         True,
+        "no_warnings":   True,
+        "skip_download": True,
+    }
+    if netscape_cookies_path and os.path.exists(netscape_cookies_path):
+        ydl_opts["cookiefile"] = netscape_cookies_path
+    if impersonate is not None:
+        ydl_opts["impersonate"] = impersonate
+
+    try:
+        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+            info = ydl.extract_info(url, download=False)
+
+        if not info:
+            return {}
+
+        raw_desc  = (info.get("description") or "").strip()
+        raw_title = (info.get("title") or "").strip()
+
+        # Prefer description (full caption with hashtags) over title
+        description = raw_desc or raw_title
+
+        logging.info(
+            f"📝 Fetched metadata for {url}: "
+            f"description={description[:80]!r}"
+            f"{'...' if len(description) > 80 else ''}"
+        )
+
+        return {
+            "description": description,
+            "title":       raw_title,
+            "timestamp":   info.get("timestamp"),
+            "uploader":    info.get("uploader") or info.get("channel") or "",
+        }
+
+    except Exception as e:
+        logging.warning(
+            f"⚠️ Could not fetch metadata for {url}: {type(e).__name__}: {e}"
+        )
+        return {}
+
+
 def download_video_ytdlp(
     url: str,
     output_path: str,
     netscape_cookies_path: str = None,
 ) -> bool:
+    """
+    Download a TikTok video using yt-dlp with browser impersonation.
+    Accepts a Netscape-format cookie file path (not JSON).
+    """
     impersonate = get_best_impersonation_target()
 
     ydl_opts = {
@@ -631,10 +685,8 @@ def download_video_ytdlp(
         "no_warnings":         False,
         "merge_output_format": "mp4",
     }
-
     if netscape_cookies_path and os.path.exists(netscape_cookies_path):
         ydl_opts["cookiefile"] = netscape_cookies_path
-
     if impersonate is not None:
         ydl_opts["impersonate"] = impersonate
 
@@ -665,7 +717,9 @@ def download_video(
     netscape_cookies_path: str = None,
 ) -> bool:
     logging.info(f"⬇️  Downloading: {url}")
-    return download_video_ytdlp(url, output_path, netscape_cookies_path=netscape_cookies_path)
+    return download_video_ytdlp(
+        url, output_path, netscape_cookies_path=netscape_cookies_path
+    )
 
 
 # ─────────────────────────────────────────────────────────────────────────────
@@ -692,14 +746,12 @@ def upload_video_to_bluesky(
 
         except Exception as e:
             err_detail = f"{type(e).__name__}: {e}"
-
             if attempt >= BSKY_UPLOAD_MAX_RETRIES:
                 logging.error(
                     f"❌ Blob upload failed after {BSKY_UPLOAD_MAX_RETRIES} attempts: "
                     f"{err_detail}"
                 )
                 return None
-
             logging.warning(
                 f"⚠️ Blob upload attempt {attempt}/{BSKY_UPLOAD_MAX_RETRIES} "
                 f"failed: {err_detail}. Retrying in {delay:.1f}s..."
@@ -734,6 +786,43 @@ def post_video_to_bluesky(
         return False
 
 
+# ─────────────────────────────────────────────────────────────────────────────
+#  Caption builder
+# ─────────────────────────────────────────────────────────────────────────────
+def build_caption(video_info: dict, tiktok_handle: str, max_len: int = 290) -> str:
+    """
+    Build a Bluesky post caption from video metadata.
+
+    Format:
+        <description>
+        <tiktok_url>
+
+    If description + URL exceeds 290 chars, the description is trimmed at
+    the last whitespace boundary before the limit to avoid cutting mid-word
+    or mid-hashtag.
+    """
+    desc = (video_info.get("description") or "").strip()
+    url  = video_info.get("url", "").strip()
+
+    if not desc:
+        # No caption available — just post the URL
+        return url
+
+    # Reserve space for newline + URL
+    url_block = f"\n{url}"
+    max_desc  = max_len - len(url_block)
+
+    if len(desc) > max_desc:
+        trimmed = desc[:max_desc - 1]
+        cut     = trimmed.rfind(" ")
+        # Only use word boundary if it doesn't cut off too much
+        if cut > max_desc // 2:
+            trimmed = trimmed[:cut]
+        desc = trimmed + "…"
+
+    return f"{desc}{url_block}"
+
+
 # ─────────────────────────────────────────────────────────────────────────────
 #  TikTok scraping — Playwright
 # ─────────────────────────────────────────────────────────────────────────────
@@ -801,9 +890,10 @@ def _run_playwright_scrape_loop(page, profile_url: str, limit: int) -> list[dict
                                 else f"https://www.tiktok.com{link}"
                             )
                             videos.append({
-                                "video_id":  video_id,
-                                "url":       full_url,
-                                "timestamp": None,
+                                "video_id":    video_id,
+                                "url":         full_url,
+                                "timestamp":   None,
+                                "description": "",
                             })
                 except Exception:
                     pass
@@ -864,7 +954,6 @@ def scrape_tiktok_profile_playwright(
             ),
             viewport={"width": 1280, "height": 900},
             locale="es-ES",
-            # Mask automation signals at the context level
             extra_http_headers={
                 "Accept-Language": "es-ES,es;q=0.9,en;q=0.8",
             },
@@ -873,7 +962,7 @@ def scrape_tiktok_profile_playwright(
         inject_cookies_into_context(context, cookies)
         page = context.new_page()
 
-        # Apply stealth v1.x if available; skip v2.x entirely
+        # Apply stealth v1.x if available; skip v2.x entirely (unstable API)
         if _STEALTH_SYNC is not None:
             try:
                 _STEALTH_SYNC(page)
@@ -928,6 +1017,9 @@ def scrape_tiktok_profile_ytdlp(
     """
     Fallback: use yt-dlp to extract the video list from a TikTok profile.
     Accepts a Netscape-format cookie file path (not JSON).
+
+    Note: flat playlist extraction gives us basic metadata (title, timestamp)
+    but not the full description — that is fetched per-video in process_videos().
     """
     import yt_dlp
 
@@ -970,10 +1062,13 @@ def scrape_tiktok_profile_ytdlp(
                     url = f"https://www.tiktok.com/@{handle}/video/{vid_id}"
                     vid_match = re.search(r"/video/(\d+)", url)
             if vid_match:
+                # description from flat extraction is usually just the title —
+                # the full caption is fetched per-video in process_videos()
                 videos.append({
-                    "video_id":  vid_match.group(1),
-                    "url":       url,
-                    "timestamp": entry.get("timestamp"),
+                    "video_id":    vid_match.group(1),
+                    "url":         url,
+                    "timestamp":   entry.get("timestamp"),
+                    "description": (entry.get("description") or "").strip(),
                 })
 
         logging.info(f"✅ yt-dlp fallback produced {len(videos)} usable videos.")
@@ -984,21 +1079,6 @@ def scrape_tiktok_profile_ytdlp(
         return []
 
 
-# ─────────────────────────────────────────────────────────────────────────────
-#  Caption builder
-# ─────────────────────────────────────────────────────────────────────────────
-def build_caption(video_info: dict, tiktok_handle: str, max_len: int = 290) -> str:
-    desc = (video_info.get("description") or "").strip()
-    url  = video_info.get("url", "")
-    if desc:
-        url_len  = len(url) + 1
-        max_desc = max_len - url_len
-        if len(desc) > max_desc:
-            desc = desc[: max_desc - 1] + "…"
-        return f"{desc}\n{url}"
-    return url
-
-
 # ─────────────────────────────────────────────────────────────────────────────
 #  Main processing loop
 # ─────────────────────────────────────────────────────────────────────────────
@@ -1012,6 +1092,14 @@ def process_videos(
     max_age_days: int,
     video_max_size_bytes: int,
 ) -> int:
+    """
+    For each new video:
+      0. Fetch full metadata (description/caption) via yt-dlp
+      1. Download the video file
+      2. Compress to fit within the PDS size limit
+      3. Upload blob to Bluesky
+      4. Create the post with caption + URL
+    """
     posted_count = 0
     now = arrow.utcnow()
 
@@ -1023,6 +1111,7 @@ def process_videos(
             logging.info(f"⏭️  Already posted: {video_id}")
             continue
 
+        # Age filter (only when timestamp is available)
         ts = video.get("timestamp")
         if ts:
             try:
@@ -1039,6 +1128,19 @@ def process_videos(
 
         logging.info(f"🎬 Processing video {video_id}: {video_url}")
 
+        # ── 0. Fetch full metadata if description not already populated ───
+        if not video.get("description"):
+            logging.info(f"🔍 Fetching metadata for {video_id}...")
+            meta = fetch_video_metadata_ytdlp(
+                video_url,
+                netscape_cookies_path=netscape_cookies_path,
+            )
+            if meta:
+                video["description"] = meta.get("description", "")
+                # Backfill timestamp if we didn't have one from scraping
+                if not video.get("timestamp") and meta.get("timestamp"):
+                    video["timestamp"] = meta["timestamp"]
+
         with tempfile.TemporaryDirectory() as tmpdir:
             raw_path  = os.path.join(tmpdir, f"{video_id}_raw.mp4")
             comp_path = os.path.join(tmpdir, f"{video_id}.mp4")
@@ -1053,7 +1155,9 @@ def process_videos(
                 continue
 
             # 2. Compress
-            ok = compress_video(raw_path, comp_path, max_size_bytes=video_max_size_bytes)
+            ok = compress_video(
+                raw_path, comp_path, max_size_bytes=video_max_size_bytes
+            )
             if not ok:
                 logging.error(f"❌ Compression failed for {video_id}. Skipping.")
                 continue
@@ -1066,6 +1170,7 @@ def process_videos(
 
             # 4. Post
             caption = build_caption(video, tiktok_handle)
+            logging.info(f"📝 Caption preview: {caption[:120]!r}")
             ok = post_video_to_bluesky(client, blob, caption, langs, video_id)
             if ok:
                 mark_as_posted(video_id, state, meta={"url": video_url})
@@ -1122,7 +1227,9 @@ def main():
     logging.info("=" * 60)
 
     state  = load_state()
-    client = connect_bluesky(args.bsky_handle, args.bsky_app_password, args.bsky_base_url)
+    client = connect_bluesky(
+        args.bsky_handle, args.bsky_app_password, args.bsky_base_url
+    )
 
     # Convert JSON cookies → Netscape format once for all yt-dlp calls
     netscape_cookies_path = convert_json_cookies_to_netscape(args.cookies_path)
@@ -1178,6 +1285,7 @@ def main():
         logging.info("=" * 60)
 
     finally:
+        # Always clean up the temporary Netscape cookie file
         if netscape_cookies_path and os.path.exists(netscape_cookies_path):
             try:
                 os.remove(netscape_cookies_path)
@@ -1185,7 +1293,9 @@ def main():
                     f"🧹 Removed temporary Netscape cookie file: {netscape_cookies_path}"
                 )
             except Exception as e:
-                logging.warning(f"⚠️ Could not remove Netscape cookie file: {e}")
+                logging.warning(
+                    f"⚠️ Could not remove Netscape cookie file: {e}"
+                )
 
 
 if __name__ == "__main__":