Maintainer of this skill?
by Win-Hao
Agent skill: turn one topic into a 1–3 minute knowledge explainer video — fact-checked article, scene-by-scene copy, freely licensed images, three looks to pick from, a beat-synced storyboard, the opening first. Rendered with HyperFrames.
# Add to your Claude Code skills
git clone https://github.com/Win-Hao/knowledge-videoGuides for using ai agents skills like knowledge-video.
See how knowledge-video compares with popular alternatives.
knowledge-video is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by Win-Hao. Agent skill: turn one topic into a 1–3 minute knowledge explainer video — fact-checked article, scene-by-scene copy, freely licensed images, three looks to pick from, a beat-synced storyboard, the opening first. Rendered with HyperFrames. It has 50 GitHub stars.
knowledge-video's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/Win-Hao/knowledge-video" and add it to your Claude Code skills directory (see the Installation section above).
knowledge-video is primarily written in Python. It is open-source under Win-Hao on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh knowledge-video against similar tools.
No comments yet. Be the first to share your thoughts!
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
An agent skill for the kind of knowledge video that took off under Douyin's #Vibe知识大赏 tag: animation drawn in code, short lines of on-screen text over music, two or three minutes on one idea. The agent doesn't generate video; it researches, writes, and builds a HyperFrames project that renders to MP4.
The skill stays deliberately light. It holds only a few hard rules (every fact sourced, every asset licensed, no AI-drawn faces of real people, frames that render the same every time), a path that works (research → copy → images → look and storyboard → render, showing you each step by default so a wrong direction is caught before a long render), and small tools for music edits, beat timing, sound effects and self-checks. How the video is written, looks and moves is left to the agent; what earlier videos taught sits in references/ as examples, not rules. The steps below are the path as it was first tested.
| Step | What the agent does | What you check |
|---|---|---|
| ① Research | decides what shape the topic has (timeline, chain of causes, shallow to deep, side by side, Q&A, story), searches the web, writes a sourced article with more candidate points than needed; cuts anything without a reliable source; offers two or three options for how many points to cover | how many points (your pick), what was cut, what is second-hand |
| ② Copy | rewrites it as scene-by-scene screens, ≤ 20 characters each; the first screen is the question itself, then a story, an everyday comparison, one twist, an ending about ordinary people | the opening question, the length |
| ③ Images | either lists real images per scene (old photo, paper, object, diagram) with URL, which part to capture, license and author, public domain / CC0 first; or writes layered prompts (one background + separate subjects, one shared style prefix, no text, no real faces) for your own image generator | the licenses or the prompts; you capture or generate the images |
| ④ Look + storyboard | three clearly different looks as stills of the first screen, with a recommendation; then measures the music's tempo and bars (scripts/beats.py) and writes every shot to a bar and beat, with the reveal and the twist on the music's changes |
pick a look; review the storyboard |
| ⑤ Review | applies your changes, re-aligns to the beats, says which hits moved | — |
| ⑥ Render | renders the opening (~10 s), checks sampled frames for clipped or tiny text, wrong years and off-beat moves, fixes them, then the full video after your OK | the opening |
Two runs, both in an empty folder with Claude Code and Claude Opus 5.5:
references/example-ai-history.md (Chinese).No skill installed? Send the prompt below to an agent and replace 【】 with your topic. Before each step it fills the tags into a complete prompt for that step, shows you, then follows it (handing the filled prompt to a subagent where it can). The six original prompts from the first test are in the git history.
<task>Help me turn 【your topic】 into a knowledge video, along the path research → copy → images → look and storyboard → render. Start by writing a brief, prompts/brief.md; before each step, fill in that step's prompt, save it as prompts/<step>.md, show me, then follow it. Hand large self-contained steps (research, finding images, writing the composition and rendering) to a subagent with the brief and the step's prompt, and check what it produces.</task>
<context>The brief has one context tag: the topic, audience, takeaway, length, orientation, music, what I like and dislike, what earlier steps settled, where the files are; update it after each step.
Each step's prompt has four tags: role (which role does this step and what it cares about most, e.g. knowledge editor, screenwriter, art director, storyboard artist, motion designer, picked per step); task (what it produces and where it goes); style (recommended approaches with reasons; recommendations, not rules; follow me where I've said what I want); check (what to verify, item by item, before the step is done). Leave out style where it isn't needed.
Infer what I haven't said from the topic, and ask me when unsure.</context>
<check>Every fact sourced, and anything unsourced left out; images, music, sound effects and fonts licensed, credits in credits.md; no AI-drawn faces of real people; watch the whole video yourself before handing it over.</check>
Put a piece of music in an empty folder, open your agent there, and say what you want to explain:
I want to make a knowledge video about why the sky is blue. The music is in the folder.
Paste into any coding agent:
Install the knowledge-video skill from https://github.com/Win-Hao/knowledge-video
Or the skills CLI:
npx skills add Win-Hao/knowledge-video -g
Claude Code plugin (tracks this repo, updates on release):
/plugin marketplace add Win-Hao/knowledge-video
/plugin install knowledge-video@knowledge-video
Manual (copies the files, pins the current version):
git clone https://github.com/Win-Hao/knowledge-video.git
cp -R knowledge-video/skills/* ~/.claude/skills/ # Claude Code
credits.md for this)skills/knowledge-video/
SKILL.md the path, the hard rules, fill-then-follow prompts
scripts/beats.py tempo, beats, bar lines and loudness changes (numpy + FFmpeg)
scripts/music_loops.py bars that repeat or cut without an audible seam
scripts/cut_music.py rebuild the music from a bar list, beat grid exact
scripts/timing.py lay every screen of the copy onto the beat grid
scripts/sfx.py download Kenney's CC0 foley, mix sound-effect cues
scripts/qa.py contact sheets of every screen and key beats
references/ examples and lessons from tested videos (not rules)
docs/hero.gif
MIT. The photos in docs/hero.gif come from Wikimedia Commons: IBM 704 (NASA, public domain), blue sky (TheUltimateGrass, CC0).