bilibili-render-pdf

name	bilibili-render-pdf
description	Use when the user provides a Bilibili URL (BV number) and wants LaTeX course notes rendered as PDF. Triggers: Bilibili link, BV号, 'B站视频笔记', '整理B站课'. Falls back to Whisper when no CC subtitles.

Bilibili Render PDF

Turn a Bilibili video into a complete, compileable .tex note and a rendered PDF.

This skill extends the youtube-render-pdf workflow with Bilibili-specific adaptations.

Shared Rules

Read ../video-render-common/writing-and-figures.md for all writing rules, teaching content rules, figure handling, figure time provenance, and visualization guidelines. These rules are mandatory.

Bilibili vs YouTube: Key Differences

Aspect	Handling
Subtitle scarcity	CC subtitles → Whisper speech-to-text → visual-only mode
Login-gated HD	1080P+ requires cookies; use yt-dlp --cookies-from-browser chrome
Multi-part videos	Detect 分P videos, ask user which parts to process
URL formats	bilibili.com/video/BVxxxxxxx and b23.tv short links
Danmaku	Do not use danmaku as teaching content (too noisy)

Goal

Same as youtube-render-pdf — produce a professional Chinese lecture note with cover image, key frames, and synthesis section.

Source Acquisition

Supplementary Material Discovery

Same as youtube-render-pdf — search for official slides AND official lecture notes before downloading video.

Additional Bilibili-specific notes:

• Many Chinese courses host slides on their own course websites or Gitee/GitHub repos
• Some courses have companion textbooks or reading lists — check the video description
• Google Slides/Docs strategy: try /export/pdf URL pattern; fall back to video frames if export fails
• If slides are found, use slide screenshots as the primary figure source and ensure they appear in the final PDF rather than relying on video frames alone
• Official lecture notes are especially valuable for Bilibili courses where subtitles may be auto-generated and formula accuracy is lower

Metadata Inspection

1. Inspect video metadata first (title, chapters, duration, thumbnail, subtitle availability)
2. Detect multi-part (分P) videos — list all parts and ask user which to process

Subtitle Acquisition (Three-Level Fallback)

Priority 1: CC subtitles

yt-dlp --write-subs --sub-langs "zh-Hans,zh-CN,zh,ai-zh" --convert-subs srt \
  --skip-download -o "%(title)s.%(ext)s" "<URL>"

Priority 2: Whisper speech-to-text

yt-dlp -x --audio-format wav -o "audio.%(ext)s" "<URL>"
tools/scripts/transcribe_faster_whisper.py audio.wav \
  --out-prefix transcript.zh \
  --model large-v3 \
  --device cuda \
  --compute-type float16 \
  --language zh

Priority 3: Visual-only mode — skip subtitles, rely on dense frame sampling.

Video and Cover Download

1. Download original cover image (highest-res thumbnail)
2. Probe formats, choose highest downloadable resolution (1080P+ needs cookies)
3. Keep all artifacts local

Delivery

• the final .tex file
• the downloaded cover image
• any extracted or generated figure assets
• the compiled PDF
• the Whisper-generated SRT subtitle file, if speech-to-text was used
• evidence that the compiled PDF was visually inspected via rendered pages/contact sheet and corrected if needed

Rules

• Follow ai-course-notes/CLAUDE.md conventions for LaTeX structure, boxes, and figures
• Always compile PDF twice (xelatex two passes) to resolve references
• After the final compile, follow the PDF Visual QA workflow in ../video-render-common/writing-and-figures.md; do not mark the note complete until rendered pages have been inspected and layout/rendering issues fixed
• Prefer CC subtitles; fall back to Whisper only when unavailable
• For long videos, use the faster-whisper CUDA helper from the shared rules; do not silently run multi-hour CPU transcription
• For fixed-camera interviews with no slides/demos/whiteboard, use the cover only and do not repeat speaker-head frames in the body; generate concept diagrams/tables instead
• Skip non-teaching content (intros, sponsor segments, 一键三连, 关注投币)
• Use [H] float placement for all figures
• Set \noteauthors to "基于公开课程资料整理" or "基于 [Speaker Name] 授课内容整理" — never "XX & Codex" or similar
• Set \notedate to the video's publish date or course semester — never \today
• Always fill \videourl with the URL the user provided
• Keep \repourl{https://github.com/hqhq1025/ai-course-notes} unchanged from the template default
• Do NOT use TikZ for any visualization — it causes compilation timeouts. Use tables or pre-generated images instead

Asset

• assets/notes-template.tex: default LaTeX template to copy and fill