editor

原文涵盖的主题

• Editor — understand the footage, then edit it to a style
• Is there a video-understanding API key?
• Step 0 — Scan inputs/ for ALL assets
• Step 1 — Understand the footage (local, no key)
• Step 2 — Pick the treatment
• Step 3 — The recipes (each = a composition of other skills)
• Montage — fast, kinetic, music-driven (20–60s)
• Documentary — narrative, slower, cinematic

Full Production Studio note: this skill drives the toolkit in agamm/video-agent. Clone that repo and run your agent inside it for the scripts/paths below to work; the guidance itself is usable anywhere. This is a director/orchestrator skill, not a single ffmpeg move. It does three things in order: (1) understand what the footage is, (2) pick a treatment (genre/format), and (3) compose the other skills into a plan and execute it. Every "signature move" below is just a call into an existing skill or CLI command — this skill decides which, in what order, and why. No separate one — and you don't need it. XAIAPIKEY drives Grok's generative editing only (grok-video-edit). The understanding model here is you (Claude) — you are multimodal. Sample the footage with the local, key-free tools and read it yourself. (Grok's vision chat could caption frames if you were ever running fully headless with no ability to read images, but that's the fallback, not the path — reading the frames directly is better and free.) Before anything else, run ls inputs/ (or whatever directory holds the footage). Look for: Multiple video files (multicam, cutaways, B-roll) Music / audio files (.mp3, .wav, .aac, .m4a)

这里只展示节选，请以完整原文和当前要求为准。