Communitygithub.com

MuhammadaliNazarov/claude-video-skills

인스타그램 릴스, 틱톡, 쇼츠, 유튜브 후크 등 숏폼 영상 제작을 위한 스크립트 작성 및 스토리텔링 기술로, 바이럴 포맷과 검증된 구조를 활용합니다.

claude-video-skills란 무엇인가요?

이 기술은 숏폼 영상(인스타그램 릴스 우선, 틱톡, 쇼츠, 유튜브 후크 포함)을 위한 스크립트 작성 및 스토리텔링에 특화된 AI 어시스턴트용 스킬입니다. 사용자가 스크립트, 후크, 캡션, 비디오 아이디어, 콘텐츠 캘린더, 스토리 계획, 또는 '이걸 바이럴로 만들어줘'와 같은 요청을 할 때 활성화됩니다. 이 스킬은 검증된 바이럴 포맷 패밀리(조언, 의견, 공감, 스토리텔링, 사용하기 쉬운)와 다양한 스토리 구조(정상→폭발→새로운 정상, 이해관계→큰 질문→헤드 페이크→재후크, 칼러웨이 후크→락인→본문/재후크 포맷, 하지만/그러므로)를 기반으로 후크, 전체 음성 스크립트, 화면 텍스트 스크립트, 캡션, B-롤 샷 리스트, 콜투액션을 작성합니다. 또한 최신 플랫폼 연구 결과를 반영하여 각 영상의 첫 1초가 시청자의 관심을 끌도록 설계하고, 모든 약속(후크에서 제시한 내용)을 본문에서 반드시 이행합니다. 스크립트 작성 후에는 점수표를 통해 후크의 효과, 정렬, 이해관계, 흐름, 긴장감, 생동감, 사운드, B-롤, 콜투액션 등을 0~3점으로 평가하고 2점 미만 항목을 수정합니다. Cursor, Claude Code, Codex와 같은 AI 에이전트 플랫폼에서 사용할 수 있으며, `creator-profile.md`, `references/`, `templates/script.md` 등의 파일을 참조하여 크리에이터의 개인 스토리와 니치에 맞춘 맞춤형 스크립트를 생성합니다.

지원 대상✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/MuhammadaliNazarov/claude-video-skills/tree/HEAD/skills/script-writer

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

Script writer

Writes short-form scripts that people watch to the end, send to a friend and follow for. Built from proven short-form formats, a frame-by-frame breakdown of a Kallaway reference Reel, and current research on hooks and Instagram ranking. Pairs with the video-editor skill: every script ends with notes that editor can execute (caption style, B-roll list, music, cut pace).

Read before writing

FileWhat it gives youRead when
creator-profile.mdwho the creator is: the default real story, niche, platformevery script
references/storytelling.mdthe story structures and the rules that make them workany story, transformation, opinion or explainer
references/hooks.mdhook rules, the three hook layers, 60+ templates sorted by mechanismevery script (hooks are written last and tested)
references/research.mdwhat Instagram ranks, pacing numbers, sourcesstrategy questions, length, CTA choice
references/sfx.mdthe 75-sound library mapped to beats (hook, flashback, explosion, win, CTA…) with per-format budgetsevery script: fill the SFX column
references/shotlist.mdshot vocabulary, the shot-list columns, B-roll rules, filming orderevery script: the draft shot list
references/talking-head-setup.mdbackground + lighting for the dorm room or outside, with fixesevery script with him on camera
templates/script.mdthe exact output layoutevery delivery

Optional raw material: put your own format notes or reference write-ups in references/ (e.g. a .docx/.mp4, combined/*.md write-ups, combined/transcripts.md) and references/ref01/ (the Kallaway Reel).

The workflow

  1. Intake (ask only what is missing, max 4 questions in one message): who is the viewer (one specific person), what do they get (laugh, cry, learn, feel seen), the creator's real story or proof, the goal (followers, comments, DMs/lead magnet, sales), talking or text-on-screen, length. Read creator-profile.md (fill it in once: who you are, where you're from, what you're building, where you post) and use it as the default "real story"; never invent achievements, numbers or quotes. 1b. Mine the story (personal formats). Pick the L.I.F.E. pillar (Love, Identity, Fitness, Earnings) and ask "why?" three times (Safe → Real → Raw, storytelling.md §11). Write from the Raw layer; if he hasn't given it, ask.
  2. Pick the format. Match the goal to a format family (Advice, Opinion, Relatable, Storytelling, Easy-to-use; storytelling.md and hooks.md). Prefer a proven format and change as few variables as possible: keep the hook wording and structure, swap in the creator's niche and story.
  3. Write the spine before the words. One sentence: "[Viewer] will watch because [curiosity/stakes] and leave with [payoff]." Then the beats, each joined by but or therefore, never "and then".
  4. Write the body, then the hook. Draft 5 hook options across different mechanisms (hooks.md), each with its text hook, spoken line and opening visual aligned. Recommend one and say why.
  5. Tighten. Cut every line that does not raise stakes, answer the question, or set up the next beat. Spoken: 150 to 250 words a minute, sentences under ~15 words, one idea per line, "you" language. Text-on-screen: each line readable in the time it is on screen (read it aloud; the cut comes when you finish). 5b. Make it live. At the key moment (low point or explosion) add one or two sensory details and switch to present tense; before the CTA, state the meaning in one line ("The reason I'm telling you this…"). Check the dopamine ladder: first frame stimulates (level 1), hook opens a loop (2), a head fake (3), a non-obvious payoff (4), his personality on screen (5).
  6. Sound pass. Fill the SFX column from references/sfx.md: pick per beat, name the exact word it lands on, stay inside the format's budget, leave — where silence is stronger. 6b. Shot list + setup. Draft the shot list from the script (references/shotlist.md): 1-3 B-roll shots per line, close-ups ≥ a third, ordered by location for filming, planned cut types, room tone. Recommend the talking-head spot and light (references/talking-head-setup.md); if he sends a photo of the spot, check it and say what to move.
  7. Deliver in templates/script.md layout, then run the scorecard below and fix anything under 2.

Non-negotiables

  • The first second does the work: a moving image + a line that creates a gap (curiosity, stakes, contradiction, identity call-out). No greetings, no "so today", no logo.
  • Every promise is paid. The body delivers exactly what the hook teased; the payoff lands before the CTA.
  • Specific beats vague: ages, numbers, times, places, names, the exact words someone said ("You need jaw surgery").
  • Vulnerability + proof: an honest low point, then evidence it is true (photos, dates, screen recordings, results).
  • One video, one message, one viewer. Remove side stories ("remove noises").
  • B-roll shows, it does not decorate: every line has a shot that proves or mirrors it (the action or the emotion).
  • Re-hook every 15 to 20 s on longer scripts: a new question, a twist, "but here's the part nobody tells you".
  • CTA matches the payoff: follow for the next part (open loop), comment a keyword for the resource, send to the person it describes, save for later. One CTA.
  • Never fabricate results, testimonials, numbers or stories for the user. Ask, or mark [fill: your number].

Scorecard (rate 0 to 3, fix anything under 2)

CheckQuestion
HookWould a stranger stop in 1 s, muted? Is there a gap they need closed?
AlignmentDo the text hook, spoken line and first visual say the same thing?
StakesIs something at risk or wanted, and does the viewer care within 5 s?
FlowIs every beat joined by but/therefore? No "and then" stretches?
SpecificityReal numbers, ages, names, dialogue?
PayoffIs the hook's promise paid clearly, before the CTA?
ShareabilityWould someone send it to a friend ("this is so you")?
ProofIs there visual evidence for every claim?
RawIs it written from the Raw layer (a truth he wouldn't normally say), not the Safe one?
MeaningIs the "why I'm telling you this" said in one line before the CTA?
LadderFirst frame stimulates (motion/colour/contrast) and his face appears at least once?
CTAOne clear ask that fits the payoff?

Keeping it the user's

"Save this to the skill" after feedback: write it as one rule in the file it belongs to (a hook rule → hooks.md, a structure rule → storytelling.md, a new format the user liked → references/formats.md (create it) with its source link). Replace a rule a new one contradicts. When a script performs well, add it to references/winners.md (create it) with the numbers the user reports.

Individual skills in this repo

This repo contains 3 individual skills — each has its own dedicated page.

MuhammadaliNazarov/claude-video-skills

The user's CINEMATIC editing skill, separate from the everyday video-editor. Use only when the user wants something more cinematic, artistic, filmic, moody or "like Gawx" — a personal documentary, a travel film, a making-of, a story Reel with a film look — or says creative, cinematic, film look, moody, aesthetic, short film, Gawx. Low-key red/teal film grade (Cine_Gawx LUT), grain, halation, vignette, framed 4:5 / 1.37:1 / 2.39:1 pictures, small red chapter cards, music-led montage bursts, object close-ups, bookend storytelling with the meaning said out loud. For normal talking-head Reels with one-word captions use video-editor instead.

MuhammadaliNazarov/claude-video-skills

Production pipeline for mathematical and technical animations using Manim Community Edition. Creates 3Blue1Brown-style explainer videos, algorithm visualizations, equation derivations, architecture diagrams, and data stories. Use when users request: animated explanations, math animations, concept visualizations, algorithm walkthroughs, technical explainers, 3Blue1Brown style videos, or any programmatic animation with geometric/mathematical content.

MuhammadaliNazarov/claude-video-skills

The user's one video editing skill. Edit any video by conversation, from raw footage to a finished file. Instagram Reels / vertical 9:16 (the default; the user mainly posts on Instagram), YouTube 16:9 hooks and long form, and anything else (montage, travel, interview, tutorial, screen recordings). Transcribes with ElevenLabs Scribe, cuts retakes and dead air, colour grades with cinematic LUTs (including Apple Log), adds word-timed captions (one word at a time in red/white or red/black by default; yellow and documentary styles too), split-explainer layouts learned from reference Reels, HyperFrames motion graphics, punch-ins, B-roll, free sound effects, music beds, and renders locally. Use whenever the user drops footage, names a video file, or says edit, cut, Reel, Instagram, YouTube, captions, subtitles, grade, LUT, Apple Log, animate, overlay, highlight this screen recording, or render.

관련 스킬