メニュー
Use when a video project needs reusable media metadata, word-level transcription, objective speech analysis, or evidence-backed semantic understanding before optional editing skills run.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Add word-timed captions to an Open Recut program. Use this skill to map the canonical transcript through timeline.json, review a maintained style on source-backed pixels, render a local transparent HyperFrames PNG sequence, and register it as an overlay contribution for the shared delivery render.
Use when an understood video project needs selective transcript-timed titles, lower-thirds, statistics, lists, quotes, chapter cards, or calls to action authored as HyperFrames HTML graphics.
End-to-end turn an unedited long-form talking-head / vlog / podcast video into a compact "first cut" (rough cut). Use when asked to edit/剪辑 a raw YouTube (or local) video into a tighter version: download, word-level transcribe, diagnose bad-edit spots (slow intro, fillers, dead air, tangents, rambling), decide cuts in a JSON edit plan (each kept/dropped span with in/out + a one-line reason), render with ffmpeg, then self-check the result (re-transcribe + frame/black/ silence checks). Triggers: "剪成第一版", "rough cut", "first cut", "压缩时长", "把这条原片剪短", "cut down this video".
Use when a completed cut-as-code delivery needs a strict side-by-side review against its original source, including cuts, varispeed, color grade, and graphics.
Select complete short-form moments from a verified Open Recut main delivery, extract approved horizontal derivatives, and optionally render reviewed 9:16 versions. Uses the shared program transcript/timeline, hash-bound human or delegated-agent decisions, and Project Protocol V1 output locations.
Use when footage needs color correction, white-balance or exposure repair, Log-to-Rec.709 conversion, named creative looks, skin-tone review, or a portable .cube LUT for a cut-as-code delivery.
| name | video-understand |
| description | Use when a video project needs reusable media metadata, word-level transcription, objective speech analysis, or evidence-backed semantic understanding before optional editing skills run. |
Build the shared evidence layer once. Keep observations in source time and leave editorial decisions to downstream skills.
This skill is a prerequisite for /video-cut, /video-to-shorts,
/video-add-captions, and /video-add-content-cards. Run it first so those skills
consume the same validated evidence and timeline.
Require ffmpeg/ffprobe, Python, and faster-whisper for transcription. Check them before processing media.
Initialize a project from the original source:
python scripts/init_project.py path/to/source.mp4 path/to/my-video-project
This creates input/, review/00-video-understanding/, final/, the minimal
machine-facing work/ tree, an identity timeline, project.json, media facts, and
START-HERE.md. It does not create folders for unselected edit operations.
Probe again only when the source needs an explicit metadata refresh:
python scripts/probe.py input/original-video.mp4 work/understand/media.json
Extract 16 kHz mono audio and transcribe it:
ffmpeg -y -i input/original-video.mp4 -ac 1 -ar 16000 work/cache/audio16k.wav
python scripts/transcribe.py work/cache/audio16k.wav work/understand/transcript medium `
--lang auto --cache-dir work/cache/faster-whisper
Use --lang auto for unknown or mixed-language speech. Never infer the spoken language
from the language of the user's prompt. Pass a fixed language such as --lang zh only
when the audio itself or explicit user metadata establishes it. Keep model downloads in
the project-local work/cache/faster-whisper/ cache. Faster-whisper may emit an
occasional point-timed word with equal start/end values; the shared timeline mapper
preserves it as a 1 ms interval so captions and derivatives do not silently lose text.
Generate objective metrics and semantic candidates:
python scripts/analyze.py work/understand/transcript.json work/understand/analysis.json
Read the source, transcript, and analysis. Author work/understand/understanding.json with factual summaries, source-time ranges, confidence, and transcript evidence. Do not prescribe cuts, cards, or looks.
Validate before downstream use:
python scripts/validate.py understanding work/understand/understanding.json work/understand/transcript.json
Create only these useful review artifacts under review/00-video-understanding/:
video-summary.md, transcript.srt, and contact-sheet.jpg. Verify metadata,
timestamps, evidence references, and visible frames. Do not substitute PNG or ad hoc
filenames for the protocol names.
Mark the understanding operation check.status as pass only after the review artifacts
and semantic evidence validate. Operation lifecycle status and check result are separate;
never write verified into check.status.
project.json.All durable machine files live in work/understand/. Treat work/cache/ as disposable.