Video Analyzer (TikTok + YouTube + Instagram)

Analyze videos from TikTok, YouTube, Instagram, Twitter, and others by URL, transcribing audio locally and answering questions about the content.

Security Scan

The skill's code, instructions, and requirements are consistent with a local video->audio->transcript analyzer that uses yt-dlp and a local Whisper model; it does not request unrelated credentials or hidden network exfiltration in the inspected files.

Purpose & Capability

Name and description (video download + local transcription) match the code and instructions: yt-dlp/ffmpeg for download, faster-whisper/whisper for local transcription, and local caching/saving of transcripts. Required tools and files are proportional to the stated purpose.

Instruction Scope

SKILL.md tightly prescribes runtime behavior (immediate acknowledgement messages, periodic progress messages, strict no-commentary rules) which is unusual but not inherently malicious. It instructs the agent to run local scripts, read/write only the skill's transcripts/staging folders, and not to send transcripts anywhere external. The periodic messaging requirement is a behavioral constraint an operator should be aware of.

Install Mechanism

There is no automatic installer in the registry entry (instruction-only + bundled scripts). README suggests standard package managers (brew/apt/pip) and git/gh clone. No downloads from untrusted shorteners or extraction of arbitrary archives were found. Note: model download from Hugging Face and pip installs are network operations but expected for this use case.

Credentials

The skill requires no environment variables or credentials. The only network interactions are: downloading video content via yt-dlp and (once) fetching the local transcription model from Hugging Face; both are consistent with the skill's purpose.

Persistence & Privilege

The skill stores transcripts under its own directory (~/.openclaw/skills/tiktok-analyzer/transcripts) and does not request system-wide privileges or set always:true. Saving files is limited to its own skill folder.

Assessment

This skill appears to do what it says: it downloads only audio (via yt-dlp), transcribes locally with faster-whisper/whisper, and caches transcripts in a local folder. Before installing, consider: 1) The first transcription run will download a ~150MB model from Hugging Face (network activity but expected). 2) You must install ffmpeg, yt-dlp, and the Python packages (pip may use --break-system-packages which can alter system Python envs). 3) The skill enforces frequent progress messages and a strict interaction style — make sure that behavior fits your UI/UX expectations. 4) Downloading videos may implicate copyright or privacy concerns depending on content. If you want extra assurance, review the repository origin (README suggests a GitHub repo) and run the Python scripts in a controlled environment first.

Video Analyzer Skill

Analyze any video by dropping a URL. Works with TikTok, YouTube, Instagram, Twitter/X, and 1000+ other sites. Transcribes the audio locally and answers any question about the content.

---

When to Use This Skill

Activate when the user:

• Shares a video URL (tiktok.com, youtube.com, instagram.com, twitter.com, x.com, etc.)
• Asks "what is this video about", "summarize this", "what are they teaching", "what's the hook", etc.
• Asks a question about a previously saved video

---

⚠️ CRITICAL RULES — READ BEFORE ANYTHING ELSE

Rule 1 — FIRST MESSAGE IS ALWAYS THIS, NO EXCEPTIONS:

Use the message tool to send this BEFORE running any exec command:

📡 Video received, analyzing...

This must be a message tool call, not your final reply text. Using the message tool sends it immediately to the user while you continue processing. If you put it in your reply text instead, the user won't see it until everything is done — which defeats the purpose entirely.

Do NOT reference conversation history, prior testing, or anything from the current session. Every URL is treated fresh.

Rule 2 — NEVER GO SILENT The user MUST receive a message every 30-60 seconds while processing. Silence = broken.

• After download step: send "📥 Downloaded! Transcribing now..."
• If anything takes more than 30 seconds: send "⏳ Still working..."

Rule 3 — NO PERSONAL COMMENTARY. EVER. Do NOT add ANY of the following:

• "This appears to be the video we already tested"
• "I recognize this URL" / "you've sent this before"
• "Heads up — this is the same link"
• Any footnote, parenthetical, or aside about the URL or prior usage

Just run the skill and give the answer. End with the save prompt. Nothing else. If the transcript is cached: say "📚 Found in your library!" then answer. That's it.

Rule 4 — First-run warning If the transcripts folder is empty (first ever run), warn upfront:

⚠️ First time running — downloading the AI model (~150MB). Takes 2-4 minutes once, never again.

---

Prerequisites Check

Before the first run, check if dependencies are installed:

which ffmpeg && python3 -c "import faster_whisper; print('ok')" && python3 -c "import yt_dlp; print('ok')"

If anything is missing, guide the user:

Mac/local:

brew install ffmpeg
pip3 install faster-whisper yt-dlp --break-system-packages

Linux/VPS:

apt install -y ffmpeg
pip install faster-whisper yt-dlp --break-system-packages

---

Flow

Step 1 — Acknowledge IMMEDIATELY (before anything else)

Send: 📡 Video received, analyzing...

Step 1b — First run warning

If this looks like the first time (no cached transcripts exist), warn the user:

⚠️ First time running — the AI transcription model needs to download (~150MB). This takes 2-4 minutes once and never again. Grab a coffee ☕

Step 2 — Download (step 1 of 2)

python3 ~/.openclaw/skills/tiktok-analyzer/transcribe.py --download-only "URL_HERE"

Returns JSON with status: "downloaded" and video_id. If from_cache: true + skip_transcribe: true → go straight to Step 3, skip Step 2b.

Step 2b — Send progress message (via message tool), then transcribe

Use the message tool to send: 📥 Downloaded! Transcribing now...

Then immediately run:

python3 ~/.openclaw/skills/tiktok-analyzer/transcribe.py --transcribe-only "VIDEO_ID"

Replace VIDEO_ID with the video_id from the previous step.

Returns JSON:

{
  "transcript": "full text here...",
  "language": "en",
  "video_id": "abc123",
  "from_cache": false
}

If from_cache: true (from Step 2) → say "📚 Found this in your library — instant answer!" and skip the wait messages.

If there's an "error" key → relay it cleanly (never show a Python stacktrace to the user).

Step 3 — Answer the question

Use the transcript to answer whatever they asked. If no specific question, provide:

• What it's about (1-2 sentences)
• Key points / what's being taught (bullet list)
• Tone / style (educational, entertainment, story, etc.)

Step 4 — Offer to save (MANDATORY if from_cache: false)

After giving the answer, ALWAYS ask this — do not skip it:

💾 Want to save this transcript so you can ask follow-up questions later without re-downloading? (yes/no)

Only skip this if from_cache: true (already saved).

If yes:

python3 ~/.openclaw/skills/tiktok-analyzer/save_transcript.py "VIDEO_ID" 'JSON_DATA'

Confirm: ✅ Saved to your video library!

---

Searching Saved Transcripts

When the user asks about something they've analyzed before:

1. List files in ~/.openclaw/skills/tiktok-analyzer/transcripts/
2. Read the relevant .json file(s)
3. Answer from the saved transcript

---

Error Handling

Error	Response
Private/removed video	"This video is private or has been removed. Try a different URL."
No ffmpeg	"You need ffmpeg. Run: brew install ffmpeg (Mac) or apt install ffmpeg (Linux)"
No faster-whisper	"Run: pip install faster-whisper yt-dlp then try again."
Timeout / very long video	"That one's taking a while — try a shorter clip or check your connection."

---

Demo Tips

• For demos: Use a video you've already analyzed (cache hit = instant response, looks great)
• First run: Always warn upfront about the 150MB model download
• Works on any platform yt-dlp supports — TikTok, YouTube, Instagram, Twitter, Reddit, Vimeo, and 1000+ more