Clips
视频剪辑全链(download → transcribe → plan → cut → translate → karaoke)From its SKILL.md
npx -y skills add ChaosRealmsAI/NextFrame --skill clipsAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 24 stars24 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 8 commands, including `nf-guide clips download` and 7 more.
SKILL.md
2.0 KB, 677 tokens by cl100k_base, as published. Nobody here has run it
clips · 视频剪辑(必跑完 6 步 · 不允许中途 stop)
NextFrame 完整 clips pipeline。6 步全跑 · 每步 Bash 一次 tool · 一直走到 karaoke HTML。
🔴 硬约束(违反 = 任务失败)
- ❌ 不允许中途 FINAL 停 · translate 和 karaoke 必须跑完
- ❌ 不允许在 plan 步输出完 plan.json 就停
- ❌ 不允许说"现在我需要翻译..." 然后不调 tool
- ✅ 每步调完 tool 立刻下一步
- ✅ 只有 karaoke 产出 index.html 后才能 FINAL
推进状态机(6 步 · 严格顺序)
Bash("nf-guide clips download")→ 调nf-source download· 产 source.mp4Bash("nf-guide clips transcribe")→ 调whisperx --model base.en(小模型 · 避免 1GB 下载)· 产 sentences.json + words.jsonBash("nf-guide clips plan")→ 你挑 1 个亮点写 plan.json(只挑 1 个 · 省 translate 工作量)Bash("nf-guide clips cut")→ ffmpeg 切 · 产 clip_01.mp4 + cut_report.jsonBash("nf-guide clips translate")→ 翻译(一次 Bash heredoc 写完整个 translations.zh.json · 不要逐句译)· 产 clip_01.translations.zh.jsonBash("nf-guide clips karaoke")→ 产 index.html(终点)
终点验收
ls <episode>/clips/ 必须有:
- clip_01.mp4
- cut_report.json
- clip_01.translations.zh.json
- index.html(karaoke · 字级同步)
反例(上次 agent 踩的坑)
❌ 错:"现在我需要翻译这 4 个英文句子..."(输出纯文本后停)
✅ 对:直接 Bash("cat > <path>/clip_01.translations.zh.json <<EOF ... EOF") 一次写完
视频选择(agent 自主)
用户让你自主选一个短视频(< 60s · 新鲜 · 不要 me-at-the-zoo)· 用 yt-dlp 下。建议挑:TED 短演讲 / 新闻片段 / 搞笑短剧 / 科技演示。
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.