Editing video with Vidmoat

A working manual for an AI agent driving Vidmoat over MCP. Read it once at the start of a session; it does not change mid-job.

The tool schema teaches the controls; nothing in it teaches the craft. An agent that only reads tool descriptions produces edits that are technically valid, correctly aligned and lifeless. Everything here is either a measured fact about this system or a working convention from editing, labelled so you know which you can bend.

Layout geometry is NOT repeated here: it is already in every prompt you get (safe areas, type scale, the stacking inequality).

Craft skills

Craft lives in focused skills. list_skills returns the catalog (name and description); get_skill with a name returns that skill's SKILL.md and lists its reference files, which get_skill with name and file returns. Load the ones the task needs, not all of them. The ones most edits need:

• pacing: shot lengths by format, rhythm, cutting on motion, transitions.
• titles: dwell time, entrances and exits, type, lower thirds and end cards.
• captions: mapping the cached transcript, presets, placement, sizing.
• audio-cleanup and music-bed: what our audio controls actually do, levels and ducking.
• colour-grade: matching before looks, setColor's absolute scale.
• camera-motion and speed-ramps: moves, easing and the speed integral.