本文へ移動
cccskills
無料GitHub で公開

kai-motion-spot

Code-rendered motion video for a product or a client, built the way a motion studio would do it. Every frame comes from an HTML seek(t) page rendered by Playwright at 1080p60 with real motion blur, using real product UI and screenshots with a privacy blur, and an ElevenLabs hype score cut on bar lines with its drop verified on the reveal. Three formats — a 20-30 s hype launch spot, a polished feature walkthrough from real screenshots, and a 19 s pain-point video with a muted site cut and burned-in text. Delivered with baked posters, H.264 + VP9 web encodes at 1280 px (1.5-3 MB), a site-embed kit (muted autoplay or click-to-play), 9:16 social re-layouts and share copy, plus a service playbook for selling it. Use when "make a launch video", "make a commercial / promo / ad spot", "motion design video", "product walkthrough video from screenshots", "pain point videos for the site", "videos for the homepage", "sell video production", "price a video package", or a polished brand spot is wanted rather than a talking-head edit or a Remotion template batch (use /kai-video-production for those).

インストール方法を見る

含まれるファイル(1)

  • SKILL.md6.9 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Kai root note: knowledge/, harness/, and scripts/ paths in this skill live in the Kai install, not the user's project. Resolve them against the first ancestor directory of this SKILL.md that contains a knowledge/ folder (the Kai plugin root, ~/.claude/kai, or the kai-cmo-harness repo). MARKETING.md, memory/, and any output files live in the current project. If a referenced scripts/ command is not available in this install, say so, skip it, and continue with the file-based guidance — never fabricate its output.

/kai-motion-spot — product video that looks like a studio made it, from code

Objective

A delivered video package the approver signs off: a hype launch spot, a walkthrough set built from real screenshots, or a pain-point set. Each video has a 16:9 master, a real 9:16 re-layout, a baked poster, web encodes, an embed snippet and share copy. Everything on screen comes from one HTML page where every style is a pure function of time, and the score is cut so its drop lands on the reveal. When the request is about selling this, the objective is a service offer built from the playbook, with the price left to the founder.

Done when

Work type social-post for anything published (floor E5/C2/O3, harness/eco-floors.yaml). A client delivery is done at E3, when the approver signs off the final package.

  • E3: the approver chose a score variant and signed off the final. Publishing is a separate, per-post approval.
  • C2:
    • analyze.py --verify passes on every master (drop within 30 ms of the picture), loudness is -14 LUFS, and SFX are within 15 ms of the grid.
    • qa.py sheets were reviewed for the whole video at phone size in both formats, against every row of the watch-back table.
    • Privacy sheets were read with no personal data legible, and site cuts were reviewed muted.
    • ffprobe matches the meta, frame 0 is the baked poster, and web encodes are 1.5-3 MB with faststart.
  • O3 (published cuts): reach, engagement rate, profile clicks and link clicks at 7 days, against a baseline recorded before publishing.

Constraints

  • Keys come from the environment only: ELEVENLABS_API_KEY, and UPLOAD_POST_API_KEY for publishing. Never print them or write them into the project.
  • Claims: show what is live, plus roadmap features the client already claims publicly, and list those roadmap claims in the report. Never invent features, stats, customers or testimonials. Use the official logo files only; never retype a logo.
  • Real material:
    • The newest UI only, in tight readable crops, never a whole spinning page, never demo cruft.
    • Screenshots go through assets.py blur, which downsamples before blurring; a plain Gaussian can leave text readable.
  • Grid: 120 BPM, with cuts on downbeats and the drop on the reveal. Any line meant to be read stays settled for at least 0.3 s per word. Stats stack and stay.
  • Music is hype for launches, whatever the brand looks like. Calm or ambient only when asked.
  • Motion taste: one idea per shot, one accent, masked reveals, one camera language. The effects banned in the method (shake, RGB split, flares, particles, glitches, bouncy easing) stay banned.
  • Formats:
    • 9:16 is its own re-layout, never a pillarbox, with key content above the bottom ~300 px.
    • Pain-point site cuts are muted, and the burned-in text carries the story.
    • Walkthroughs are click-to-play; short loops are muted autoplay.
  • Render fixes stay on: the warm-up pass, the two-rAF paint wait, every hard cut in cuts, and fast moves in fast.
  • Publishing: posting follows harness/references/social-automation-rules.md and the platform posting rules. upload_post.py stays a dry run until the account owner approves that exact post.
  • Selling: the price is a founder decision; cite market reference points only with a URL. Internal costs (music credits, render time, operator time) never appear on a customer-facing proposal, page or email.

Context

NeedLoad / run
Formats, page contract, score, render fixes, taste rules, watch-back QA tableharness/references/motion-spot-method.md
Deliverables menu, client inputs, turnaround, QA checklist, approval, cost basis (INTERNAL)harness/references/motion-spot-service-playbook.md
Engine (run from workspace/video/<slug>/)scripts/motion_spot/: render.py, probe.py, qa.py, poster.py, web_encode.py, analyze.py, build_score.py, sfx.py, music.py (--dry-run), assets.py, walkthrough/walkthrough.py, upload_post.py
Starting pointsreference_spot.html + .meta.json, examples/walkthrough.example.json, examples/cues.example.json, examples/blur-boxes.example.json, examples/upload-spec.example.json
Product, claims, voiceMARKETING.md (project root)
Posting rulesharness/references/social-automation-rules.md, harness/references/*-organic-posting-rules.md

Requirements: Python 3.10+ with playwright numpy scipy pillow (then playwright install chromium), and ffmpeg with libx264 and libvpx-vp9. Smoke test: DUR=2 MASTER=none PAGE=index.html python "$KAI/render.py".

Output: workspace/video/<slug>/:

  • the page and meta
  • assets/, audio/
  • out/: masters, posters, share-copy.txt, render-log.txt, web/ with the encodes and snippets, and site-embed.md for sets
  • qa/: the QA sheets

Escalate when

  • The product, the promise or which features are live cannot be established from MARKETING.md or the client's materials.
  • A screen can only be shown with real personal data that the blur cannot fully cover.
  • A claim, stat or testimonial the client wants has no source.
  • analyze.py --verify keeps failing after a snap. Offer another track rather than shipping an off-beat drop.
  • Anything would be posted publicly, or a price would be quoted to a customer.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

ad-copy

無料

Write platform-compliant ad copy for Meta, Google, TikTok, LinkedIn, Pinterest, Snapchat, Amazon, or X. Loads TOS rules automatically. Shows char counts and format preview.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

ad-render

無料

Render video ads using Remotion. Scaffolds a project from brand tokens + creative brief, generates scene compositions for 4 archetypes, renders to MP4 in multiple formats.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

Research competitor ads using Meta Ad Library, Google Ads Transparency Center, and TikTok Creative Center. Analyze creative patterns, hooks, offers, and landing pages.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

checklist

無料

Surface the right marketing checklist for any task. Covers content, SEO, ads, email, press, landing pages, launches, and technical audits. 17+ checklists indexed by task type.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

Generate a strategic content brief from format + site + keyword. Auto-resolves persona, pulls GSC data, generates hooks, angles, and competitor weakness via LLM.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

Run the full quality gate on any content file. Score against 28 rules, apply policy, show detailed scorecard with fix suggestions. Works on any markdown file.

日本語の概要は準備中です。原文の説明を表示しています。

cgallic/kai-cmo-harness562026年10月3日 更新

cgallic のスキルをすべて見る

このスキルの問題を報告する