linkedin-launch-video — Claude Code skill: LinkedIn launch video

What linkedin-launch-video does

The skill walks from brief to finished MP4. It looks for product and audience context first in .agents/product-marketing-context.md or .claude/product-marketing-context.md, then in the Cogny MCP context tree, and only falls back to WebFetch on the URL when both are empty. After that it asks about the reader's role and seniority, the costly problem they face, the proof point, who speaks on screen, and the conversion goal. Vague answers on role, proof or goal stop the script step — the skill pushes back instead of guessing.

Brand capture is its own step. You supply a site URL, a repo path holding tailwind.config.* or CSS variables, an existing brand-kit.json, or three hex colors plus a font and a voice. The kit is saved as brand-kit.json in the HyperFrames project root. One piece of advice from the skill: skip LinkedIn blue #0a66c2 as the accent, since it disappears into the feed UI and the clip ends up looking like a screenshot.

Scripting follows feed grammar — 30 to 75 seconds, captions burned in at no more than 8 words per card, no music, no press-release phrasing. Node.js 22 or newer and FFmpeg 6 or newer have to be on the machine; without them the HyperFrames project will not build. Publishing is optional, and each publish call asks for confirmation first.