appariciojunior/video-editor-skills

video-editor-setup

Sets up an AI first-pass video editing workspace for DaVinci Resolve Studio with Claude Code - folders, tools, API keys handled safely, the Resolve template, and which small scripts to build first.

ソースを見る

使い方

  1. インストールコマンドをコピーし、自分の環境へ導入する前にソースを確認します。
  2. 必要なツール、認証情報、対応環境を原文で確認します。
  3. 入力と成果物を指定し、出力を資料と照合します。
npx skills add https://github.com/appariciojunior/video-editor-skills --skill video-editor-setup

開始前の確認

  • 導入にはモデル契約、API枠、ファイルへのアクセス許可は含まれません。
  • スター数や導入数は活動の指標であり、独立した品質・安全性試験ではありません。
リポジトリの原文

見出し、例、コード、表、リンク、参照画像を含む原文を表示しています。

Video editor: setup

You are going to be a first-pass editor. The creator films, you import, cut, mark, score and caption; they make the taste calls. Everything below was learned editing real YouTube vlogs and classes. Build the workspace with the user, step by step, and ask before installing anything.

What the user needs

  • DaVinci Resolve Studio (the free version has no external scripting). In Preferences > System > General set External scripting using: Local, restart Resolve.
  • ffmpeg / ffprobe, Python 3.9+, Node 18+ (only for motion graphics).
  • A speech-to-text API with word timestamps and verbatim mode (we used ElevenLabs Scribe: word-level timings, audio events like "(laughs)", auto language detection). Any service with per-word start/end works.
  • Optional: an Epidemic Sound account connected as an MCP server (music + SFX), Figma MCP (brand boards), Remotion (rendered overlays).

Folder layout (create it with them)

<workspace>/
  raw video/            the creator drops footage here: one file, or a folder of clips. READ-ONLY for you.
  export edited video/  the creator exports here. Never touch.
  brand/                BRAND.md (look + voice rules), brand.json (tokens), logo, sfx/
  system/
    config.json         all defaults (see templates/config.example.json in this package)
    keys.env            API keys. NEVER print, echo, cat or read it into the conversation.
    scripts/            the small scripts you write together (list below)
    looks/              exported grade stills (.drx) for log footage
    motion/             the motion-graphics kit (Remotion project)
    videos/<video>/     per-video work: transcript, decisions, cuts, music bed, captions, overlays
  CLAUDE.md             house rules + a "Learned facts" list you keep adding to (dated, one line each)

Hard rules to write into CLAUDE.md on day one

  1. Never modify, move or copy the raw footage. Resolve plays the camera files in place; you write only small JSON/text into system/videos/<v>/.
  2. In Resolve, only add. Import into bins, create new timelines (a rebuild makes "v2", never overwrites), add on free tracks. Never change the creator's template timeline, their own timelines or their media. Never switch, create or close projects.
  3. Transcripts stay verbatim. Re-transcribing costs money: ask first.
  4. Keys stay secret (see above). Load them from the file inside scripts only.
  5. Decisions are by word index, timing is by script. You decide which words stay; a script turns that into frame-exact cuts. Never hand-edit generated cut files.
  6. If a step fails twice, stop and report the exact error.

The scripts to build (in this order)

Write each as a small, boring CLI that prints a short summary. Keep them in system/scripts/.

ScriptDoes
doctorchecks ffmpeg, Python, the key file exists (without printing it), Resolve reachable, which project is open
ingest <v>probe each clip (fps, size, audio layout), read the camera's sidecar XML for picture profile, transcribe, detect silences, write a manifest
plan_timeline <v>keep-list + moments → frame-exact cuts, pauses, markers, a human-readable review file (see rough-cut + vlog-pacing)
resolve_build <v>import, copy the template timeline, lay the cut, zoom, grade, markers, audio normalize (see resolve-scripting)
settings <v>per-video overrides on top of config.json (set cut.max_pause=1.0)
score_mix <v>music cues + ducking envelope + scene SFX (see music-and-sound)
captions <v>SRT from the transcript, cleaned and corrected (see captions)
overlays <v>word→frame map of a timeline, render beats, place them (see motion-overlays)
clean <v>moves a finished video's work files to the Trash after the creator says yes

Ask the creator once, then save to CLAUDE.md

  • Camera, lens, picture profiles (log vs. baked), frame rate they shoot, microphones.
  • Their Resolve project: bins (we used 01Footage, 02Audio, 03Graphics, 04Timelines), the template timeline name, which audio track is voice (A1), music (A2), SFX (A3).
  • Languages they speak on camera.
  • Do they talk to "the editor" while recording ("editor, put a video here")? Those lines become markers.
  • Their brand (colours, fonts, padding, motion feel) for the motion kit.

Then hand over to edit-video.

同じリポジトリから

関連する Skills

すべての Skills
appariciojunior
コミュニティ

captions

Clean, readable burned-in style subtitles from the verbatim transcript - stutters and fillers removed, spoken slang written correctly, one short line at a time, names checked, every line read by Claude before import into DaVinci Resolve. Use for "subtitles", "captions", "legendas", or when the creator reports wrong words in the captions.

1時間の導入数
1
GitHub Stars
435
更新日
10月23日
appariciojunior
コミュニティ

motion-overlays

Brand motion graphics for creator videos - a Remotion component kit (key phrases, lists, lower thirds, journeys, big words…) timed to the creator's words from a beats sheet, rendered as transparent ProRes and placed on V2+ in DaVinci Resolve, with face-safe and caption-safe layouts. Use for "overlays", "motion graphics", "make it interesting", "graphics", or when overlays cover the face or fight the captions.

1時間の導入数
1
GitHub Stars
435
更新日
10月23日
appariciojunior
コミュニティ

resolve-scripting

Practical DaVinci Resolve Studio scripting knowledge for building edits from code - timelines from a template, frame-exact appends, zoom, grades, custom frame sizes, subtitles, renders, and the traps that silently fail or crash Resolve 21. Use when writing or debugging any script that talks to Resolve.

1時間の導入数
1
GitHub Stars
435
更新日
10月23日
appariciojunior
コミュニティ

rough-cut

Editorial rules for choosing what to keep in talking-head footage (retakes, false starts, spoken notes to the editor, fillers) and the keep-list / moments formats that a planner script turns into frame-exact cuts. Use when deciding a cut, revising one from feedback, or writing keep.json.

1時間の導入数
1
GitHub Stars
435
更新日
10月23日