video-breakdown
Skill4 starsChanged 5 months agoMITView on GitHub
Break down any video (YouTube, Loom, Instagram, TikTok, local file) frame by frame. Downloads the video, rips scene-change frames + a timestamped transcript locally, then Claude reads the visuals AND the words together to produce a content teardown — hooks, structure, on-screen visuals, why it works. Built in-house, no third-party code, only yt-dlp + ffmpeg. Trigger on "/video-breakdown", "break down this video", "analyze this reel/video", "teardown <url>", "what's working in this video".
1 other project uses a version of this.
Adopted by
Most recently taken up by
- RahmanBuildsAI/ravyn-reel-kittheir own version, first seen Oct 25, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- RahmanBuildsAI/ravyn-reel-kit★ 9same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- ai-writing-qaSkill
- build-captureSkill
- build-modeSkill
- fitfoSkill
- flow-mapSkill
- generate-system-mapSkill
- imessageSkill
- leanSkill
- meta-promptSkill
- no-bullshitSkill
- plain-englishSkill
- pressure-testSkill
- problem-logSkill
- reality-checkSkill
- swipeSkill
- viral-hook-creatorSkill
- working-with-claudeSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 11
No video model. The skill splits the video into the two things Claude reads natively — images (frames) and text (transcript) — with …
- Very specific: line 37
Read the frames. Use times.txt to know when each frame happens, so visuals align to the script.
- Distinctive: line 44
If asked: graphics/motion prompts, script template, hook variations for a niche
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video Breakdown
Reverse-engineer any video the way you actually watch one: visuals + audio at the same time. Built for competitor teardowns and ripping apart trending content in your niche.
How it works
No video model. The skill splits the video into the two things Claude reads natively — images (frames) and text (transcript) — with timestamps lined up, so Claude knows what's on screen the moment something is said.
yt-dlppulls the video + auto-captions from almost any site (YouTube, Loom, IG, TikTok) or a local file.ffmpegrips frames at every scene change (falls back to interval sampling for static talking-heads), capped at ~80.- Captions become a clean timestamped
transcript.txt(no captions = on-screen text in frames still carries it). - Claude reads the frames + transcript and writes the breakdown.
Run it
bash ~/.claude/skills/video-breakdown/scripts/grab.sh "<url-or-path>" [slug] [--start MM:SS] [--end MM:SS] [--every N]slug— folder name (auto-derived if omitted)--start / --end— only analyze a window--every N— seconds between frames when falling back to interval sampling (default 3)
Output lands in ~/claude-watch/library/<slug>/:
frames/frame_*.jpg— the visualsframes/times.txt— ffmpeg pts_time per frame (maps frame number → timestamp)transcript.txt—[MM:SS] lineformatpacing.txt— cut-rhythm summary (cut count, first-cut time, avg gap, slow-open/static/fast flags). Heuristic from ffmpeg scene detection, not a motion read.meta.txt— title, source, frame count
Then (Claude's job)
- Read
transcript.txtfully. - Read the frames. Use
times.txtto know when each frame happens, so visuals align to the script. - Produce the teardown. Default structure (adapt to what you asked for):
- TLDR — one line on why this video works
- Hook — first 0–3s, exact words + what's on screen
- Structure — beat-by-beat with timestamps (acts, retention turns, CTA)
- Visual playbook — cuts, text overlays, b-roll, motion, pacing
- Why it's trending / what to steal — the transferable moves
- If asked: graphics/motion prompts, script template, hook variations for a niche
Notes
- Caption-less sources (often IG/TikTok, local files): grab.sh falls back to local whisper.cpp (
whisper-cli+~/.cache/whisper-cpp/ggml-small.en.bin, no API key, on-device, wired in 2026-07-05). If whisper-cli or the model is missing, transcript says NO_CAPTIONS. Lean on frames + on-screen text, say so rather than guessing audio. - Everything runs locally. No content leaves the machine except the original yt-dlp fetch.
- Frame folder is wiped + rebuilt on each run for the same slug.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.