Adds a skill discovery and loading system that lets the LLM agent
autonomously find and follow pre-built editing recipes. Skills are
exposed as two new tools: list_skills (search/discover) and load_skill
(get full instructions).
Built-in skills:
- viral-short: high-retention TikTok/Reels/Shorts workflow with hook,
retention cuts, zoom effects, captions, and CTA
- pitch-video: professional pitch/explainer with 5-act structure
(opening, problem, solution, metrics, CTA)
- redo: re-applies the last undone action via core.command.redo()
- toggle_track_mute: toggles mute on video/audio tracks via
ToggleTrackMuteCommand
- toggle_track_visibility: toggles visibility on overlay tracks via
ToggleTrackVisibilityCommand
- update_clip now supports trimStart/trimEnd for slip-trimming clips
without moving them on the timeline
- undo: reverts the last editing action via core.command.undo()
- duplicate_elements: duplicates elements onto new tracks using
DuplicateElementsCommand
- list_timeline now includes hasMask, hasEffects, and isHidden hints
on elements so the agent can identify clips needing deeper inspection
Add two new agent tools for deep element inspection and property mutation:
- get_element: returns full type-specific metadata for any timeline element
(video, image, text, graphic, sticker, audio, effect)
- update_clip: mutates any visual property on an element including masks,
opacity, transform, blendMode, hidden, volume, muted, and name
Masks are handled as a nested property with add/update/remove actions,
aligned with the data model where masks are properties of elements rather
than standalone timeline entities.
Luis Esteban Acevedo Ladino