szhshp

[ENTRY POINT] Full NotebookLM pipeline — research → generate video → download → watermark → trim → generate meta files → cover → prepare upload → publish. Single entry point for all end-to-end video tasks.

Cross-platform video upload (login only when explicitly requested) — Bilibili, Douyin, Kuaishou, Weixin Video, YouTube

Research a topic with NotebookLM — create notebook, run deep research, auto-import sources, and summarize. Checks notebooklm installation, login, and skill installation first.

Create a NotebookLM notebook for a topic and generate a video overview in Simplified Chinese. Delegates installation and login checks to notebooklm-research skill.

Overview of the Panda Video Automation pipeline — from NotebookLM research to video upload. Run this skill to see the full workflow.

Download a NotebookLM video artifact to /input/video.mp4 and create a title.json with its title.

Generate minimalist academic-style cover images via Python/Pillow — black (#000000) solid background, centered bold white title (120px), thin subtitle (76px) below. ONLY text on black — no graphics, icons, dates, or logos. TRIGGER: activates when user mentions "simple cover", "minimalist cover", "academic cover", or "cover" without further qualifiers. For "detailed cover", "complex cover", "report-style cover", or "detailed cover"/"complex cover", do NOT use this skill — instead generate a detailed infographic with --style professional or --style scientific.

Prepare a downloaded NotebookLM video for upload: create the title JSON file and move the video to the correct path expected by the Bilibili/Douyin upload e2e scripts.

Multi-platform browser login and video upload via Playwright Test only (headed Chromium). Parallel uploads are allowed and encouraged: run one Playwright process per platform (e.g. separate terminals). Use when the user mentions publishing, uploading video, or logging in to Bilibili, Douyin, Kuaishou, Weixin Channels, or YouTube; or STEP3 / 多平台发布 / login:* / upload:*.

Prefer the hosting AI Agent to write narration into input.txt from crawl JSON (no external LLM). Fallback only: pnpm caption:env / generateVideoScript* (DeepSeek or Kimi) + optional WebVTT estimate. Triggers: 口播, input.txt, captions.vtt, LLM_PROVIDER, DEEPSEEK, MOONSHOT, KIMI.

Run Remotion renders from monorepo root: pnpm render:video, render:all / pipeline:tts-render, render:composition; sync-to-public, composition IDs, output/video paths, VIDEO_PUBLIC_DIR. Use for 渲染成片, output/video/video.mp4, remotion render, Cover-Still.

Atomic reference for three spider flows: spider:extract:file, spider:extract:url, spider:zhihu — inputs, outputs, env vars only.

Atomic reference for @panda-video-generator/tts-node: pnpm tts, cli.ts, processNarrationFile — Edge-TTS narration → audio.mp3 + audio.vtt; env vars TTS_*, EDGE_TTS_*, ffmpeg. Triggers: TTS, Edge-TTS, 口播音频, audio.mp3, public/tts.

Orchestrates repo pipeline from a single URL: branch Zhihu vs generic page → spider → DeepSeek caption when needed → Edge TTS → Remotion render to output/video/video.mp4. Use when the user gives a URL for 一条龙、网页转视频、爬取+口播+渲染、or automate spider→tts→render from link.

Cross-platform video upload — Bilibili, Douyin, Kuaishou, Weixin Video, YouTube