chubbyskillsSAFE

From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.

链接 / 本地文档 → Markdown 原文与附件 → 本地搜索 → 带出处的资料包 → Agent 整理选题

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: bilibili-transcribe
description: >
  哔哩哔哩视频 → 下载 → 转录 → 存为 Markdown 的完整工作流。
  支持 BV 号和完整链接,无需登录。
metadata:
  category: "media"
  triggers: "[\"用户发送 B 站视频链接\", \"把这个 B 站视频转成文字\", \"帮我转录这个 bilibili\"]"
  version: "1.0.0"
  tags: "[\"media\", \"audio\", \"video\", \"transcription\", \"bilibili\"]"
---

# B 站视频转录 Skill

**字幕优先**:先尝试抓取官方/自动字幕(秒级、免 GPU、免 funasr);抓不到再下载音频用 SenseVoice-Small 转录,存为 Markdown 文件。加 `--no-subtitle` 可强制走音频转录。

## 环境要求

```bash
# Python 3.9+
python -m venv .venv
source .venv/bin/activate

# 依赖
pip install funasr modelscope torch torchaudio

# 系统依赖
# macOS: brew install ffmpeg yt-dlp
# Ubuntu: sudo apt install ffmpeg && pip install yt-dlp
```

## 使用方法

```bash
python scripts/transcribe.py "https://www.bilibili.com/video/BV1rrQGBeEen/"
python scripts/transcribe.py "BV1rrQGBeEen"
python scripts/transcribe.py "BV1rrQGBeEen" --no-subtitle   # 强制音频转录

# 批量处理:每行一个 B站 URL 或 BV 号
python scripts/batch_transcribe.py urls.txt -o ./output
```

## 流程

### Step 1: yt-dlp 下载音频

B 站有反爬机制,需要加 User-Agent 和 Referer:

```bash
yt-dlp --extract-audio --audio-format mp3 --audio-quality 128K \
  --user-agent "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) ..." \
  --referer "https://www.bilibili.com" \
  "https://www.bilibili.com/video/BV1rrQGBeEen/"
```

### Step 2: SenseVoice-Small 转录

```python
from funasr import AutoModel
from funasr.utils.postprocess_utils import rich_transcription_postprocess

model = AutoModel(
    model="iic/SenseVoiceSmall",
    trust_remote_code=True,
    vad_model="fsmn-vad",
    vad_kwargs={"max_single_segment_time": 30000},
    device="cpu",
)

result = model.generate(input=audio_path, language="zh", use_itn=True, batch_size_s=60)
```

### Step 3: 生成 Markdown

自动创建带 frontmatter 的 Markdown 文件。

### 批量模式

`scripts/batch_transcribe.py` 会逐条调用单篇转录脚本,默认单条失败后继续处理下一条;需要严格模式时加 `--stop-on-error`。

## 性能数据

| 视频时长 | 转录耗时 |
|------|------|
| 5 min | ~10s |
| 10 min | ~20s |
| 30 min | ~60s |

## 已知限制

- B 站 4K/1080P60 需要大会员,普通用户可下载 720P 以下
- SenseVoice-Small 模型首次下载约 893MB
- 不支持说话人分离
- 合集视频会下载第一个分 P

## 参考项目

- [yt-dlp/yt-dlp](https://github.com/yt-dlp/yt-dlp) - 视频下载工具
- [FunAudioLLM/SenseVoice](https://github.com/FunAudioLLM/SenseVoice) - 语音识别模型

## ⚖️ 合规声明

仅供**个人学习与研究**使用。请遵守目标平台的服务条款(ToS)与 robots 规则,控制请求频率,不要用于批量抓取、商用爬取或侵犯他人权益的场景。下载内容的版权归原作者所有。

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

declared (5 observation(s))

Network

declared (6 observation(s))

Shell

none-observed

Dependencies

not all pinned

Secrets in source

none-found

Findings (7)

urls = [self.audio.as_uri(), "ftp://public.example/a.mp3", "https://user:[email protected]/audio.mp3", "http://127.0.0.1/a.mp3", "http://[::1]/audio.mp3", "http://169.254.169.254/audio.mp3", "https:

urls = [self.audio.as_uri(), "ftp://public.example/a.mp3", "https://user:[email protected]/audio.mp3", "http://127.0.0.1/a.mp3", "http://[::1]/audio.mp3", "http://169.254.169.254/audio.mp3", "https:

funasr, modelscope, torch, torchaudio

funasr, modelscope, torch, torchaudio

funasr, modelscope, torch, torchaudio

PyYAML, ruff

funasr, transformers, modelscope, torch, torchaudio, beautifulsoup4, markitdown, pymupdf

Gates applied: no_behavioural_pass.