Communitygithub.com

nateherkai/hyperframes-student-kit

视频编辑流水线的第二个智能体。用于查找并删除人物出镜录音中的口误——口吃、重复词、假起始和重录音条。在完成静音剪切后使用,用于剪切错误、移除重复/口吃/填充重启、清理失误,或保留最佳片段。以审核为控制:它会为每个剪切提供上下文和理由以供批准,然后仅通过ffmpeg渲染批准的剪切。需要字级转录文本。

hyperframes-student-kit 是什么?

该工具是一个视频编辑AI智能体,作为自动化编辑流水线的第二步,专门处理人物出镜录音中的口误。其核心功能是检测并移除口吃(如“那个那个”)、重复词、假起始(未说完的短语重新开始)以及重录音条(重复录制同一句台词时保留最佳一条)。由于机械检测无法区分真正的口误与有意的强调重复(如“绝不,绝不”),因此该工具默认采用“审核门控”模式:它不会直接剪切,而是列出所有候选剪切,附带完整的上下文文本和推荐意见,让使用者(或上层AI)逐条审核批准。批准后的剪切会通过ffmpeg进行音视频同步的修剪与拼接。该工具需要前一步静音剪切输出的转录JSON文件,并生成新的转录JSON文件供后续的运动图形智能体使用。整体流程包括:候选检测、人工/智能审核、应用剪切三个步骤。

兼容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/nateherkai/hyperframes-student-kit/tree/HEAD/.agents/skills/cut-mistakes

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

剪切错误与重复(流水线智能体2)

自动化编辑的第二步,在cut-silences之后运行。它移除人工编辑在第二遍时会剪掉的内容:口吃("the the"、"I- I-")、立即重复的词语、假起始(未完成的短语重新开始)以及重录音条(重复录制的一行——保留干净的,丢掉失误的)。

判断什么是错误是判断而非公式——有意的重复("never, never")和修辞上的双重("the first piece is, is this…")在机械上看与口吃完全相同。因此,该智能体默认为审核门控:它以完整上下文展示候选;智能体(和用户)决定;仅渲染批准的剪切。

何时使用

  • "剪切错误"、"移除口吃/重复"、"清理失误"、"保留最佳片段"、"剪切假起始"
  • 作为主编辑工作流的第二阶段,在cut-silences的输出上运行。

三部分流程

1. 查找候选(机械部分)

node .agents/skills/cut-mistakes/scripts/find-cut-candidates.mjs \
  <silence-transcript.json> --out-dir video-projects/<slug>/assets

输入通常是智能体1的 <stem>.silence-transcript.json,因此剪切会落在已经静音的时间线上。写入:

  • <stem>.cut-candidates.json — 结构化候选(类型、置信度、提议的cut范围、removes/keeps文本、上下文、推荐)
  • <stem>.cut-candidates.md — 可读的提议

候选类型:stutter(立即词重复)、retake(通过Jaccard相似度检测到的重复/近似重复片段)、false_start(短且未完成的短语重新开始)。

2. 审核(门控 — 必需)

在上下文中阅读每个候选并决定保留或剪切。机械检测器无法区分有意的强调/修辞和真正的失误,因此不要盲目信任recommend——阅读removes+context。常见的误报:强调性重复("it's never, never hands-off")、系动词后接疑问("the piece is, is this…")、列举。

向用户展示候选项目,每项附有推荐;收集批准。将批准的剪切写入文件:

{ "cuts": [ { "start": 100.93, "end": 101.06, "reason": "I- false start" }, ... ] }

(start/end以秒为单位,位于输入转录/视频时间线上。您可以扩大候选范围——例如,对于重录音条,从失误片段的开始剪切到干净重录的开始。)

3. 应用批准的剪切

node .agents/skills/cut-mistakes/scripts/apply-cuts.mjs \
  <silence-transcript.json> --cuts approved-cuts.json \
  --video video-projects/<slug>/assets/edited-silenced.mp4 \
  --output video-projects/<slug>/assets/edited-clean.mp4 --apply

写入EDL、重新计时的 <stem>.mistakes-transcript.json(供运动图形智能体 + validate-beat-sync.mjs使用)、决策日志,并且如果使用 --apply,则写入剪切的视频(ffmpeg trim+concat,音视频同步)。去掉 --apply 则为干运行。

审核结果

使用共享审核工具目测剪切并抽查边界:

node scripts/build-edl-review.mjs <stem>.mistakes-edl.json \
  --original video-projects/<slug>/assets/edited-silenced.mp4 \
  --edited   video-projects/<slug>/assets/edited-clean.mp4 \
  --output   video-projects/<slug>/assets/mistakes-review.html
npx serve . -p 8080 -n

备注

  • 非常干净的录制可能产生很少或零个实际剪切——这是有效的结果;不要为了凑数而剪切自然的讲话。
  • 两个说话词之间的剪切是硬连接。对于口吃/假起始,它们通常很干净;对于更紧凑的音频,可以稍后添加20-30毫秒的淡入淡出。
  • 将 mistakes-transcript.json + 干净视频交给智能体3(运动图形/层级卡片)。

Individual skills in this repo

This repo contains 8 individual skills — each has its own dedicated page.

nateherkai/hyperframes-student-kit

Agent 1 of the video editing pipeline. Removes silences and dead air from a talking-head recording. Use when asked to cut silences, trim pauses, remove dead air / gaps, or tighten the pacing of a raw video, given a word-level transcript. Produces an edit list (EDL), a re-timed transcript for downstream agents, and optionally the cut video via ffmpeg. Does NOT cut mistakes, repeats, or false starts — that is the cut-mistakes agent.

nateherkai/hyperframes-student-kit

Edit a raw talking-head video through transcription, silence trimming, mistake review, visual storytelling, motion graphics, and verified HyperFrames rendering. Use for a complete edit; route a single requested operation directly to its specialist skill.

nateherkai/hyperframes-student-kit

GSAP animation reference for HyperFrames. Covers gsap.to(), from(), fromTo(), easing, stagger, defaults, timelines (gsap.timeline(), position parameter, labels, nesting, playback), and performance (transforms, will-change, quickTo). Use when writing GSAP animations in HyperFrames compositions.

nateherkai/hyperframes-student-kit

HyperFrames CLI tool — hyperframes init, lint, preview, render, transcribe, tts, doctor, browser, info, upgrade, compositions, docs, benchmark. Use when scaffolding a project, linting or validating compositions, previewing in the studio, rendering to video, transcribing audio, generating TTS, or troubleshooting the HyperFrames environment.

nateherkai/hyperframes-student-kit

Install and wire registry blocks and components into HyperFrames compositions. Use when running hyperframes add, installing a block or component, wiring an installed item into index.html, or working with hyperframes.json. Covers the add command, install locations, block sub-composition wiring, component snippet merging, and registry discovery.

nateherkai/hyperframes-student-kit

Plan, write, time, style, and validate motion-graphic beats for long-form talking-head HyperFrames videos. Use when Codex is asked to add or improve visual beats, glassmorphism lower thirds, crumpled-paper full-screen takeovers, transcript-synced callouts, pacing coverage, YouTube retention graphics, or render-ready overlays in an existing HyperFrames video project.

nateherkai/motion-showreel

Build 10–30s motion-design showreels in the style of Nate's Opus 5.5 CLAUDE/MOTION REEL — brand motif chapters, HUD, BPM-locked cuts, logo lockup.

nateherkai/website-to-hyperframes

Capture a website URL and produce a HyperFrames promo/product-tour video — site → finished marketing video outcome.

相关技能