videocut-clip
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chinese<!--
input: 删除任务 TodoList（口误+静音）
output: 剪辑后视频、字幕文件
pos: 执行 skill，用户确认删除任务后调用
架构守护者：一旦我被修改，请同步更新：
1. ../README.md 的 Skill 清单
2. /CLAUDE.md 路由表
-->
<!--
input: Deletion Task TodoList (verbal slips + silent segments)
output: Edited video, subtitle file
pos: Execute the skill, called after the user confirms the deletion task
Architecture Guardian: Once I am modified, please update synchronously:
1. Skill list in ../README.md
2. Routing table in /CLAUDE.md
-->
剪辑
Editing
执行删除 → 重新审查 → 循环直到零口误 → 生成字幕
Execute deletion → Re-review → Loop until no verbal slips → Generate subtitles
快速使用
Quick Start
用户: 确认，执行剪辑
用户: 全删
用户: 保留静音3和5，其他都删User: Confirm, execute editing
User: Delete all
User: Keep silent segments 3 and 5, delete the rest前置条件
Prerequisites
需要先执行 生成删除任务 TodoList
/videocut:剪口播You need to first execute to generate the Deletion Task TodoList
/videocut:edit-voiceover流程
Workflow
1. 读取用户确认的删除任务
↓
2. 计算保留时间段
↓
3. 生成 FFmpeg filter_complex
↓
4. 执行剪辑
↓
5. 重新转录 + 审查 ←───┐
↓ │
有口误? ──是─────────┘
↓ 否
6. 生成字幕（SRT）
↓
7. 完成1. Read the user's confirmed deletion tasks
↓
2. Calculate the reserved time segments
↓
3. Generate FFmpeg filter_complex
↓
4. Execute editing
↓
5. Re-transcribe + Review ←───┐
↓ │
Any verbal slips? ──Yes─────────┘
↓ No
6. Generate subtitles (SRT)
↓
7. Complete进度 TodoList
Progress TodoList
启动时创建：
- [ ] 确认删除任务
- [ ] 执行 FFmpeg 剪辑
- [ ] 重新转录审查
- [ ] 生成字幕循环时更新版本号（v2→v3→...）
Created at startup:
- [ ] Confirm deletion tasks
- [ ] Execute FFmpeg editing
- [ ] Re-transcribe and review
- [ ] Generate subtitlesUpdate the version number during loops (v2→v3→...)
一、读取删除任务（时间戳驱动）
I. Read Deletion Tasks (Timestamp-driven)
从 输出的 TodoList 读取。直接使用时间戳，不要搜索文本：
/videocut:剪口播口误（N处）：
- [x] 1. `(start-end)` 删"错误文本" → 保留"正确文本" ← 勾选=删除
语气词（N处）：
- [x] 1. `(前字end-后字start)` 删"嗯" ← 勾选=删除
静音（N处）：
- [x] 1. `(start-end)` 静音Xs ← 勾选=删除
- [ ] 2. `(start-end)` 静音Xs ← 未勾选=保留Read from the TodoList output by . Use timestamps directly, do not search for text:
/videocut:edit-voiceoverVerbal slips (N instances):
- [x] 1. `(start-end)` Delete "incorrect text" → Keep "correct text" ← Checked = Delete
Fillers (N instances):
- [x] 1. `(previous word end - next word start)` Delete "um" ← Checked = Delete
Silent segments (N instances):
- [x] 1. `(start-end)` Delete silent segment of Xs ← Checked = Delete
- [ ] 2. `(start-end)` Delete silent segment of Xs ← Unchecked = Keep⚠️ 关键规则
⚠️ Key Rules
- 直接用时间戳：从 解析，不要搜索文本
(start-end) - 不要重新搜索：审查稿已经计算好精确时间戳
- 勾选 = 删除，未勾选 = 保留
- Use timestamps directly: Parse from , do not search for text
(start-end) - Do not re-search: The review draft has already calculated precise timestamps
- Checked = Delete, Unchecked = Keep
二、FFmpeg 命令
II. FFmpeg Command
bash
ffmpeg -y -i input.mp4 \
-filter_complex_script filter.txt \
-map "[outv]" -map "[outa]" \
-c:v libx264 -crf 18 -c:a aac \
output.mp4bash
ffmpeg -y -i input.mp4 \
-filter_complex_script filter.txt \
-map "[outv]" -map "[outa]" \
-c:v libx264 -crf 18 -c:a aac \
output.mp4filter.txt 格式
filter.txt Format
[0:v]trim=start=0:end=1.36,setpts=PTS-STARTPTS[v0];
[0:a]atrim=start=0:end=1.36,asetpts=PTS-STARTPTS[a0];
[0:v]trim=start=2.54:end=10.5,setpts=PTS-STARTPTS[v1];
...
[v0][a0][v1][a1]...concat=n=N:v=1:a=1[outv][outa][0:v]trim=start=0:end=1.36,setpts=PTS-STARTPTS[v0];
[0:a]atrim=start=0:end=1.36,asetpts=PTS-STARTPTS[a0];
[0:v]trim=start=2.54:end=10.5,setpts=PTS-STARTPTS[v1];
...
[v0][a0][v1][a1]...concat=n=N:v=1:a=1[outv][outa]三、重新转录审查
III. Re-transcribe and Review
剪辑后必须：
- 用 FunASR 重新转录
- 检查是否还有口误
- 有 → 回到 重新识别
/videocut:剪口播 - 无 → 生成字幕
After editing, you must:
- Re-transcribe using FunASR
- Check for any remaining verbal slips
- If yes → Return to to re-identify
/videocut:edit-voiceover - If no → Generate subtitles
四、输出文件
IV. Output Files
01-xxx-v2.mp4 # 剪辑后视频
01-xxx-v2_transcript.json # 重新转录（验证用）
01-xxx-v2.srt # 字幕文件版本递增：v1→v2→v3...
01-xxx-v2.mp4 # Edited video
01-xxx-v2_transcript.json # Re-transcription (for verification)
01-xxx-v2.srt # Subtitle fileVersion increment: v1→v2→v3...
五、反馈记录
V. Feedback Records
2026-01-15
2026-01-15
- 语气词删除边界不精确：删语气词时把前面的字也删了
- 原因：直接用语气词的时间戳删除
- 正确：从前一字 end 到后一字 start
- 语气词 + 静音要一起删：要删整段 (A.end - B.start)
A [静音] 语气词 B - 教训：删除语气词时，边界是 到
前一字.end后一字.start
- Inaccurate boundary for filler deletion: Deleted the preceding word along with the filler when deleting fillers
- Cause: Directly used the filler's timestamp for deletion
- Correct approach: From the end of the preceding word to the start of the next word
- Delete fillers + silent segments together: For , delete the entire segment (A.end - B.start)
A [silent segment] filler B - Lesson: When deleting fillers, the boundary is from to
preceding word.endnext word.start
2026-01-14
2026-01-14
- 口误文字没删干净，只删了静音段
- 教训：直接从 TodoList 读取时间戳，不要重新查找
- "拉满新"删成了"会的时候"：搜索"拉满新"时间戳跨度7秒（含6秒静音），把"拉满"也删了
- 教训：对于"删前面保后面"的口误，只删差异部分
- "AI就是AI"出现两次AI：只删了"就是"，没删第一个"AI"
- 教训：替换型口误必须删完整的第一个版本
- 系统性解决：时间戳驱动，审查稿直接标注 ，剪辑脚本不再搜索文本
(start-end)
- Verbal slip text was not completely deleted, only the silent segments were deleted
- Lesson: Read timestamps directly from the TodoList, do not re-search
- "拉满新" was deleted into "会的时候": The timestamp span for searching "拉满新" was 7 seconds (including 6 seconds of silence), which also deleted "拉满"
- Lesson: For verbal slips where "delete the front part and keep the back part", only delete the differing segment
- "AI就是AI" had two instances of "AI": Only deleted "就是", not the first "AI"
- Lesson: For replacement-type verbal slips, the entire first version must be deleted
- Systematic solution: Timestamp-driven, the review draft directly marks , and the editing script no longer searches for text
(start-end)