bilibili-video-summary

name	bilibili-video-summary
description	Summarize Bilibili (B站) videos using opencli bilibili subtitle + summary + video commands. Extracts subtitles, official AI summary, and video metadata, then synthesizes into structured summaries, blog posts, or threads.
trigger	B站视频|bilibili视频|b23.tv|BV[a-zA-Z0-9]+|总结.*视频|视频.*总结|这个视频.*看看|bilibili.*总结|B站.*总结
version	1.2.0
tags	["bilibili","video","summary","opencli","subtitle","transcript"]

Bilibili Video Summary

Summarize any B站 video by extracting subtitles, official AI summary, and metadata via opencli bilibili.

Prerequisites

• opencli installed: npm install -g @jackwener/opencli
• If install fails with EACCES: npm config set prefix ~/.npm-global && npm install -g @jackwener/opencli
• Chrome running with B站 logged in
• opencli Browser Bridge Chrome extension installed

Workflow

1. Resolve short link → BV号

If user shares a b23.tv short link, resolve it first:

curl -sI "https://b23.tv/XXXXX" | grep -i location | grep -oP 'BV[a-zA-Z0-9]+'

For direct bilibili.com/video/BV... links, extract the BV号 directly.

2. Parallel fetch — all three data sources at once

These three commands are independent and SHOULD be run in parallel (3 simultaneous terminal calls):

# Video metadata (title, author, duration, stats)
opencli bilibili video <bvid> -f json

# Official B站 AI summary (chapter outline + timestamps)
opencli bilibili summary <bvid> -f json

# CC subtitles (full transcript)
opencli bilibili subtitle <bvid> -f json

Running them in parallel saves 2 round-trips compared to sequential execution.

3. Synthesize

Priority order for synthesis:

1. summary → Use as the structural skeleton. It provides chapter-level outlines with timestamps — the highest-quality, most condensed source.
2. subtitle → Fill in details, key quotes, specific data points, and "金句" (memorable one-liners).
3. video → Header metadata: title, author, date, views/likes/favorites stats.

Output Format (Chinese)

## 📺 标题 — UP主

**BV号** | 时长 | 发布日期 | 播放量 · 点赞数 · 收藏数

---

### 🔑 核心观点（一句话提炼）

### 分段内容（按 summary 时间戳结构化展开，subtitle 补充细节）

### 💡 金句（从 subtitle 提取的 memorable quotes）

---

总结：（一句话收尾）

For long videos (20min+ with 600+ subtitle entries), synthesize by chapter rather than line-by-line.

Fallback Paths

A) All opencli commands timeout (Chrome not running / bridge not connected)

This is the most common failure mode — all three parallel commands return [Command timed out after 30s]. Do not retry; switch immediately to the browser fallback:

1. Use browser_navigate to the B站 video page directly: https://www.bilibili.com/video/<bvid>
2. The page snapshot includes: video title, description (often contains project links and structured content), tags, UP主 name, and basic stats (plays, likes, danmaku)
3. Supplement with web_search for the BV号 to find related discussions/blog posts that may contain detailed summaries
4. For "Github一周热点" type videos, the description often lists all projects with GitHub links — extract these and use web_extract on each GitHub repo for details

Fallback Paths

If subtitle returns nothing (no CC subtitles):

1. Use summary alone — the AI summary is often sufficient for an overview
2. Use video metadata (description + tags) for a lighter summary
3. Report to user that the video lacks captions

Command Reference

Command	What it gets
opencli bilibili video <bvid> -f json	Title, author, duration, stats, description
opencli bilibili summary <bvid>	Official AI summary with chapter outline + timestamps
opencli bilibili subtitle <bvid>	CC subtitles (manual or auto-generated)

All commands require Chrome running with B站 login.

Whisper 语音转文字（无字幕降级）

当视频无任何字幕时，用 Whisper 本地语音转文字。适用于少数无字幕视频的兜底方案。

GPU 加速（WSL）

WSL 2 的 GPU 通过 /usr/lib/wsl/lib/ 访问：

uv venv --python 3.12 .venv-cuda
uv pip install --python .venv-cuda/bin/python \
  --index-url https://download.pytorch.org/whl/cu128 torch \
  -i https://pypi.tuna.tsinghua.edu.cn/simple openai-whisper

GPU 显存	模型	速度
≥12GB	medium	~0.45x 实时
6-12GB	small	~0.21x 实时（推荐）
<6GB	base	—

⚠️ WSL CUDA 下 medium 比 small 慢（4070 实测: 95s vs 45s），选 small。 ⚠️ 不要强设 --language zh：让 Whisper 自动检测语言。

B站 412 反爬

yt-dlp 访问 B站返回 HTTP 412 时加：

--add-header "Origin:https://www.bilibili.com" --add-header "Referer:https://www.bilibili.com/"

依赖

⚠️ Python 3.14 不兼容 openai-whisper（缺少 pkg_resources）。必须用 Python 3.12 venv。

工具	用途
yt-dlp	下载视频/字幕
ffmpeg	音频处理
torch + openai-whisper	语音转文字
opencc	繁转简

详见：

• references/whisper-wsl-cuda-setup.md — WSL CUDA 完整配置
• references/python314-whisper-compat.md — Python 3.14 兼容问题
• references/bilibili-412-fix.md — B站 412 反爬修复

Pitfalls

• No subtitles: Many B站 videos lack CC subtitles. Fall back to summary + video metadata.
• Browser not running / all commands timeout: If all three opencli bilibili commands return [Command timed out], Chrome or the Browser Bridge extension is not running. Do not retry — immediately switch to the browser fallback path.
• Not logged in: Some content (大会员, age-restricted) requires B站 login in Chrome.
• Summary unavailable: B站 AI summary is not available for all videos — it's a relatively new feature.
• Rate limiting: Avoid rapid-fire calls. Space commands a few seconds apart.
• npm prefix: If npm install -g fails with EACCES, run npm config set prefix ~/.npm-global first. The global prefix may drift to /usr/lib after system npm updates.

References

• references/opencli-bilibili-commands.md — Full opencli bilibili -h output snapshot