메뉴
The routine for making a film end to end in this repo - brief, direction, build, verify, score. Phases have pass-gates. Use when starting, reviewing, or shipping any video composition.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/ytrofr/claude-remotion-editor --skill video-playbook명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
The law for making a film in this repo - what makes a shot cinematic rather than a slide, and a linter that enforces it. Use before authoring any composition, when a render "doesn't feel right", or when reviewing motion work.
React reducer state management with multi-undo, coalesced drag updates, and gesture-first dispatch. Use when building interactive editors, implementing undo/redo, or managing complex UI state.
Choose between Remotion, HyperFrames, or hybrid for new video work in this project. Use when asked "should I use Remotion or HyperFrames for X", "which framework for ___", before starting any new composition, or when evaluating migration of existing scenes. Covers the dual-stack decision matrix derived from the 2026-04-22 evaluation session.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | video-playbook |
| description | The routine for making a film end to end in this repo - brief, direction, build, verify, score. Phases have pass-gates. Use when starting, reviewing, or shipping any video composition. |
| user-invocable | true |
| allowed-tools | Read, Write, Edit, Grep, Glob, Bash |
The routine lives in routines.json (house schema: phases, each with a gate; steps
typed MACHINE / AGENT / PRINCIPLE). This file is the router.
1 MEASURE THE BRIEF gate: duration x script arithmetic done, conflict surfaced
2 DECLARE THE DIRECTION gate: one big idea + restraint statement written down FIRST
3 BUILD gate: doctrine linter clean
4 LOOK gate: range-rendered and actually watched, not stills alone
5 MEASURE THE ARTIFACT gate: probes run, negative controls fired
6 SCORE gate: blind, order-swapped, baseline present
Phase 2, the restraint statement. Before any code: what is the one big idea, and what will you deliberately NOT do? A direction without a stated restraint converges on the same shell every time, and "add more effects" is the default failure mode of fixing "it feels flat". Ask the grid-fallback question: if this were reduced to a plain centred layout, what idea would break? If the answer is "not much", the direction is not ready.
Phase 4, looking at it. Stills at mid-segment frames are clean by construction.
Flicker, mount flashes, transition seams and camera judder live between waypoints and
are invisible to a still. Range-render (--frames=A-B) anything with a camera move.
| what happened | the rule it became |
|---|---|
| A brief fixed 15s AND supplied a script measuring 302s | Measure the script before accepting the duration (phase 1) |
| 3 of 5 defects in one arm were visible ONLY in the render | Phase 4 is not optional |
A cutsheet guard emitted NaN: while its test matched nan: | Every gate needs a mutation that makes it fire |
| Full render broke, scoped render clean, 5 sessions of code edits | Same code + different output ⇒ vary the ENVIRONMENT (rule 57) |
scale: 0 eaten by || 1, read by the operator as "flickering" | ?? for every numeric default |
| feTurbulence at 1920x1080 cost 1.68 s/frame | Price the grade; grain renders at 1/3 res and scales up |
video-doctrine (the law + linter) · framework-selection (Remotion vs HyperFrames) ·
scripts/bench-video/ (probes + scoring) · video-storyboard (planning artifacts)