Bilibili Video Notes skill
Skill turns an educational Bilibili video into a DOCX note with screenshots: subtitles, frames, OCR and frame scoring.
- 54 ★ on the repository
- Last commit
Extended access
Why this level
- The skill uses your Bilibili cookie and a model key, both held in local files.
- Lecture frames are sent to an external multimodal model.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/bilibili-video-notesDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill uses your Bilibili cookie and a model key, both held in local files.
- Lecture frames are sent to an external multimodal model.
This is third-party code. Review the repository files before installing.
What it does
Bilibili Video Notes builds a structured DOCX note from an educational video. The chain is: download the video and AI subtitles, extract a frame every ten seconds, drop duplicates by OCR and perceptual hash, score frames with a multimodal model, read the content of selected frames and assemble a document from subtitles and screenshots. 129 frames shrink to 20 to 40. The model is set in `.env` through `VISION_API_KEY`, `VISION_BASE_URL` and `VISION_MODEL`, and any OpenAI-compatible API with image support works, including local vLLM. The skill targets several agents, among them Hermes and Claude Code, and temporary files are cleaned up automatically.
Who it is for. Students and self-learners watching lectures on Bilibili who want a note with key slides.
Good fit when
- An hour-long lecture with formulas becomes a note with screenshots of the needed slides.
- Duplicate frames should be removed, keeping the best one per topic.
- A multimodal model is connected through your own OpenAI-compatible endpoint.
- The agent is handed an installation in one message with commands.
Not a fit when
- The video is not from Bilibili, since the skill works only with that site.
- You have no key for a multimodal model, which frame scoring requires.
- You need rights to the material and the lecture is protected by its author.
Example request
Make a note of this Bilibili lecture from the link: take the subtitles, pick the most useful frames and assemble a DOCX with screenshots.Limitations
It needs Python 3.9 or newer, `ffmpeg` on the system PATH and a multimodal model key, and calls are billed by the provider. Video access uses your `SESSDATA` cookie from the browser, stored in a local `bilibili_cookies.txt` file, so never publish that file. Downloading and noting other people's lectures is subject to site rules and copyright, and responsibility lies with the user. The documentation is in Chinese. GitHub metadata shows no license file, although a badge says MIT.
How to disable. Delete the skill folder and the local `.env` and `bilibili_cookies.txt` files. Rotate the `SESSDATA` cookie if needed by signing out of Bilibili in the browser.
FAQ
What is needed to run it?
Python 3.9 or newer, `ffmpeg`, a multimodal model key in `.env` and the `SESSDATA` cookie in `bilibili_cookies.txt`.
Which models work?
Any OpenAI-compatible API with image support: OpenAI, Sophnet or local vLLM. Set address and model in `VISION_BASE_URL` and `VISION_MODEL`.
How many frames remain?
From 129 frames, 20 to 40 remain after OCR and perceptual hashing, and model scoring picks the best one per point.
Where is the cookie stored?
In a local `bilibili_cookies.txt` file next to the skill. Do not commit it or show it to others.
Does it work with other agents?
Yes, the README names Hermes, Claude Code and others. You can hand the installation to the agent in one message from the one-click section.
Discussion
No questions or reviews yet.
Related
SkillsTurns a lecture recording into slide-illustrated notes and an HTML page where video, transcript and summary stay in sync.
Four skills for Codex: YouTube and Bilibili videos into LaTeX PDF notes, SRT refinement and tensor formula diagrams.
Learn from materials
learn-from-materials
A skill that turns books, PDFs, slides and documents into a traceable knowledge base and an interactive learning page with sources, maps and self-check
16 ready skills for Cursor, Claude Code and Codex: theses, PDF and Office, draw.io diagrams, project documentation
Turns a technical book or a folder of documents into a structured skill with chapters, a glossary and a cheatsheet
A skill that turns tender documents into a Word technical bid: scoring-based, with industry guides and a 14-step pipeline
README in short
The README, in Chinese, covers installation, features, `.env` and cookie setup, the full workflow, commands, FAQ and folder structure.