mattthomas7641

Newsletter Video Pipeline — Data skill for Claude Code

Data community

ByteSize News (Newsletter-to-Video Pipeline): a decision model screens the TLDR newsletter stories, Claude writes 40-second explainers, a locally cloned voice reads them, HyperFrames renders animated.

How to install Newsletter Video Pipeline

This entry records only its repository, not the path inside it, so there is no exact command to give. Open mattthomas7641/newsletter-video-pipeline and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Newsletter Video Pipeline does

Alternatives in Data

HTML Anything — ✨ The agentic HTML editor — your local AI agent writes the HTML, you ship it 8.5k ★
Clinical Decision Support — Evidence-based clinical decision support tools 6.4k ★
Memex — Open-source, local-first AI journal app for iOS and Android 697 ★

README

ByteSize News (Newsletter-to-Video Pipeline)

An unattended daily pipeline that turns the [TLDR](https://tldr.tech) newsletters into 40–45 second vertical (1080×1920) tech-news explainers and uploads them to YouTube Shorts.

tldr.tech (4 editions) ──▶ scrape ──▶ Jev screens ──▶ Claude writes ──▶ voice each line ──▶ HyperFrames ──▶ MP4s ──▶ YouTube
   tech · ai ·              ~40–50     in scope? how     N scripts:        (cloned voice via   render                  Shorts
   marketing · founders     stories    newsworthy?       hook, beats,      Voicebox, or Kokoro)
                                       (optional)        cards, outro

Each video covers one big story or 2–3 related ones. It opens with a spoken hook shown as big kinetic text, then plays as a comic-style explainer over full-screen gameplay footage: a topic badge, one punchy card per spoken beat that slides, drops, zooms or flips in (numbers get a big stamped stat), word-by-word captions with the spoken word highlighted, and a narrator character at the bottom of the frame. An outro card asks viewers to comment.

Runs itself: a macOS launchd job writes, voices, renders and uploads the day's videos at 7:00 AM, logging every stage. With a locally cloned voice, 10 videos take about 85 minutes on a laptop.
Two-stage story selection: with a TypeSafe API key, Jev screens every candidate for scope and newsworthiness first, so Claude only sees the shortlist (see Story selection with Jev).
Claude as head writer, with a fallback: with an API key, Claude Sonnet 5.5 picks the most interesting tech stories from the candidates, combines related ones, and writes each script plus its post title, caption and hashtags (prompt: WRITER_SYSTEM in script_writer.py). Without a key, a rule-based writer groups TLDR's top stories into videos.
Designed in HyperFrames: the look lives in an HTML composition (newsletter_video/hf_template/) you can