bilibili-up-batch

haizhouyuan's avatarfrom haizhouyuan

批量下载 B 站 UP 最近 N 条视频到 imports，并用 video_pipeline 生成证据包（适合 worker panes 并行跑）。

⭐0stars🔀0forks📁View on GitHub🕐Updated Dec 17, 2025⬇️1downloads
[Web Research]

When & Why to Use This Skill

The Bilibili UP Batch Download & Analysis skill is a high-performance automation tool designed for researchers and content analysts. It streamlines the process of bulk-downloading recent videos from specific Bilibili creators and passing them through a sophisticated processing pipeline. By leveraging OCR and transcription, it transforms raw video data into structured evidence packages and searchable digests, making it an essential asset for digital archiving, competitive intelligence, and large-scale social media research.

Use Cases

  • Competitive Intelligence: Automatically track and analyze the latest content strategies and themes from specific Bilibili creators to stay ahead in market trends.
  • Academic & Social Research: Efficiently gather large-scale video datasets and transcriptions for linguistic, cultural, or sociological studies of Chinese digital media.
  • Digital Archiving & Evidence Gathering: Securely download and document video content with OCR-backed metadata for legal, investigative, or historical preservation purposes.
  • Automated Content Summarization: Convert long-form video content into concise, searchable text summaries and evidence files, significantly reducing manual review time.
namebilibili-up-batch
description批量下载 B 站 UP 最近 N 条视频到 imports，并用 video_pipeline 生成证据包（适合 worker panes 并行跑）。

触发条件

当用户给出一个 B 站 UP 主页（space.bilibili.com/<mid> 或 b23.tv 短链），并要求“拉取最近 N 条视频并批处理”为研究资产时使用。

目标

  1. 下载：把最新 N 条视频下载到 imports/content/videos/bilibili/<mid>/
  2. 分析：对每个视频运行 video_pipeline，产出 state/video-analyses/<analysis_id>/evidence.json 与 evidence_compact.md
  3. （可选）后续由 digest-content 读取 evidence_compact.md 生成 digest 并归档到 topic

依赖

  • ffmpeg（系统已安装即可）
  • yt-dlp（建议安装到仓库 venv：.venv/bin/python -m pip install yt-dlp）

硬约束

  • 不在 git 里提交下载的视频文件（imports/ 默认不入 git）
  • video_pipeline 只处理本地文件；下载是独立步骤
  • 不写入任何 cookie/token 到仓库（如需登录态下载，只允许通过环境变量/本地配置注入）

SOP（推荐用 worker panes）

  1. Controller pane（主控）先设置接收通知

    • 运行：scripts/tmux_set_controller_pane.sh

  2. Worker pane：批量下载 + 分析

    • 对单个 UP（最近 30 条）：
      • python3 scripts/bilibili_up_batch.py --up '<space_or_b23_url>' --limit 30 --download --analyze --enable-ocr
    • 若你要节省时间（先跑转写，稍后再 OCR）：
      • python3 scripts/bilibili_up_batch.py --up '<space_or_b23_url>' --limit 30 --download --analyze

  3. Worker 完成后通知 controller

    • scripts/tmux_notify_controller_done.sh --topic <topic_id或占位> --record <state/runs/...json> --status done

  4. 生成 digest（可选但推荐）

    • 对每个 state/video-analyses/<analysis_id>/evidence_compact.md 使用 digest-content 生成 digest；
    • 如需归档到 topic：再用 topic-ingest 更新 sources.md/timeline.md。