GitHub radar
reelbench-skills: Shot-by-Shot Video Analysis for AI Agents
Wesley Wang's Claude Code and Codex skills turn a finished video into an interactive shot-by-shot report — I review how that helps a content business copy what works.
reelbench-skills is a small set of skills for Claude Code and Codex from developer Wesley Wang. The first skill breaks a finished video into a shot table: duration, shot size, camera movement, what is on screen and the editing rhythm, with cut points measured by ffmpeg and only the judgement calls left to the model. The second one renders a video side by side with its shot information, highlighting each shot as it plays. The output is a single offline HTML report with a built-in player, a searchable shot table and keyframes. The author notes it has no npm dependencies and needs no API key. Parts of the repository description are in Chinese — ask your agent to summarize the README in English if anything is unclear.
Why a vibe-coder should care
For anyone running a content factory this is a cheap way to learn from videos that already perform: break down a competitor's best reel, see how short the shots are and where the cuts fall, and brief your own editor or generation pipeline with that rhythm. It runs on your agent's existing subscription, so there is no separate tool to pay for. You can also rebuild the idea for your niche — for example, a report that compares your last ten videos with the top ten in your topic.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install the video analysis skills from https://github.com/eternityspring/reelbench-skills — follow the README for Claude Code, then break my video down shot by shot and open the report for me.
Обычный ноутбук: нужны только Node.js и ffmpeg, агент поставит их сам.
Open on GitHub▌ More finds