video-transcript
Skill1 starsChanged 31 days agoMITView on GitHub
视频文案提取与总结(抖音 / B 站)。用户发来抖音分享链接(v.douyin.com)或 B 站视频链接(bilibili.com/video/BVxxx、b23.tv),并要求提取文案、转录、分析视频内容、或问这个视频讲了什么时使用。抖音走本地 Traffic Compass 服务,B 站走 scripts/bili_asr.py(音频直链 + 阿里云百炼实时识别),之后生成结构化总结并存成 Markdown。首次安装、服务起不来见 references/setup.md;提取报错见 references/fallback.md;平台与链路细节见 references/platforms.md。
17 other projects use a version of this.
Adopted by
Most recently taken up by
- spongetimblin/claude-skill-librarytheir own version, first seen Nov 8, 2026
- thatjuan/agent-skillstheir own version, first seen Nov 8, 2026
- notque/vexjoy-agenttheir own version, first seen Nov 7, 2026
- Ntmib/jarvis-architecttheir own version, first seen Nov 1, 2026
- guipi888/workbuddy-video-transcripttheir own version, first seen Oct 25, 2026
- juliasadler/muse-skills-cntheir own version, first seen Oct 17, 2026
Other versions17 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 17
- ZeroPointRepo/youtube-skills★ 972same name
- notque/claude-code-toolkit★ 448same name
- notque/vexjoy-agent★ 448same name
- Backtthefuture/huangshu★ 274same name
- artemchuikin/youtube-skills★ 273same name
- bozhouDev/video-skills-toolkit★ 147same name
- Ntmib/jarvis-architect★ 4same name
- shalomb/agent-skills★ 3same name
- kuiilabs/claude-skills★ 2same name
- guipi888/workbuddy-video-transcript★ 1same name
7 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
Files
视频文案提取与总结
收到视频链接时:先判平台 → 选链路 → 提取 → 通读 → 结构化总结 → 存 Markdown → 交付。
触发条件
用户消息同时满足:
- 含抖音链接(
v.douyin.com/douyin.com/video/xxx)或 B 站链接(bilibili.com/video/BVxxx、b23.tv/xxx) - 表达提取 / 分析意图之一:「提取文案」「转成文字」「帮我分析这个视频」「这个视频讲了什么」「做个读书笔记」
用户要求安装、配置、排查本功能时同样处理——先读 references/setup.md。
平台矩阵(先看这张表)
| 平台 | 链路 | 入口 |
| 抖音 | Playwright 抓直链 + 百炼批量 ASR(需要登录 Cookie) | bash scripts/douyin_extract.sh "<分享链接>" |
| B 站 | 音频直链下载 + 转码切段 + 百炼实时 ASR 并发 | python scripts/bili_asr.py "BVxxx" -o <目录> |
| 其他(YouTube 等) | 未支持 | 直接告诉用户不支持,别硬试 |
两条链路的细节、限制和真实耗时见 references/platforms.md。
配置(已落默认值,一般不用改)
| 配置项 | 默认值 | 覆盖方式 |
| Traffic Compass 项目 | ~/Documents/Playground/traffic-compass | TC_PROJECT |
| 服务地址 | http://127.0.0.1:8900 | TC_ADDR |
| 百炼密钥 | webapp/_data/apikeys.json 的 bailian 字段 | BAILIAN_KEY |
| 运行 Python | <项目>/.venv/bin/python(要有 dashscope≥1.27) | 直接换解释器 |
| 输出目录 | 当前工作目录 | -o 参数 |
工作流
Step 1 · 探活服务(抖音链路必需)
bash scripts/ensure_service.sh返回 up: 或 ready: 才算通;启动失败 时看日志 ~/Library/Logs/traffic-compass.log。
B 站链路不依赖这个服务,但转码需要 ffmpeg/ffprobe。
Step 2 · 提取文案
抖音:
bash scripts/douyin_extract.sh "用户消息里的完整分享内容"取回 JSON 里的 data.text。耗时通常 30–120 秒,脚本已设 300 秒超时。
B 站:
<项目>/.venv/bin/python scripts/bili_asr.py "BV1xxxxxxxxx" -o "<输出目录>"- 输出
B站_<UP主>_<标题>_transcript.md,正文是带[mm:ss]时间轴的逐句文案,另有字/秒统计。 - 默认 600 秒一段、4 路并发。96 分钟的视频实测 3–5 分钟出全稿(每段先推送、再等结果,远快于实时)。
- 重复提取同一视频会命中
--work下的缓存(segs/、asr/),不重复花钱;缓存默认在<输出目录>/.bili_work/<BV>/。 - 只想验证链路或省钱:
--segments 0:1(只跑第一段)。 - 日志里出现
[err]/[empty],或末尾[warn] 字/秒偏低,说明有段落失败——按references/fallback.md处理后再写总结,别拿残缺稿当全文。
Step 3 · 通读全文
总结前必须通读全部文案。 长文分块读,禁止只看开头就下判断。B 站脚本输出的 Markdown 已按段分好,可按 ### 第 N 段 逐段读。
Step 4 · 生成结构化总结
拆解时区分两件事,尤其在"带读/讲解类"视频里:
- 原作者的观点(原文写了什么)
- 讲解者的观点(UP 主/解读者自己加的判断、吐槽、案例)
两者混着写会让总结失真,必要时在文中标注归属。
总结包含:
- 标题:提炼核心议题
- 视频信息:平台、作者、时长、系列与否(是否"第 X 期"、是否延续上期)
- 核心观点:1–2 句主旨
- 观点详细拆解:按逻辑顺序,每节含子观点标题 + 完整论述链 + 关键数据/案例(保留人名、数字、场景、走向,别只写"举了个例子")+ 原文金句(引用块,原样保留)
- 金句摘录:5–10 条最有传播力的原话,各附一句语境说明
- 关键结论:视频最后的结论或立场;系列视频注明"本期覆盖范围"与"下期讲什么"
长视频(>60 分钟)额外要求:先出一份时间轴章节地图(哪段时间讲什么),再按章节拆解。否则 2 万字的稿子会 80% 篇幅堆在前 20 分钟。
Step 5 · 保存 Markdown
文件名:YYYY-MM-DD_作者名或关键词_标题摘要.md
(例:2026-10-31_胡泊Hubo_置身钉内_大厂如何落地AI产品一.md)
# [标题]
**来源:抖音 / B 站**
**提取日期:YYYY-MM-DD**
**原链接:[原视频链接]**
---
## 核心观点
## 观点详细拆解
### 1. [子观点一]
[论述 + 论据/案例]
> [原文金句]
## 金句摘录
## 关键结论
---
<!-- 分隔线以下为原文 -->
[原文:不修改、不删减、不润色;同音错字可轻量修正,不改原意]Step 6 · 交付
stage_files 交付文件。对话里只给提炼和判断,不要把几万字原文倒进聊天。
什么情况读 references
- 首次使用 / 安装 / 配置 / 服务起不来 →
references/setup.md - 提取报错、cookie 失效、ASR 为空、直链抓不到 →
references/fallback.md - 想知道某平台具体怎么走、能处理多长、有哪些已知坑 →
references/platforms.md
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

