Diffusion Studio Skills: Help Agents Understand and Edit Footage

watch does not scan an entire video blindly. It probes the tracks first, uses a waveform and filmstrip to reveal structure, then chooses word-timed transcription, audio understanding, or timestamped frame grabs according to the question. Answers include jumpable time ranges. editor reuses that evidence pass, writes a brief before a nontrivial cut, lays down the A-roll spine, then layers B-roll, captions, sound effects, and generated assets. The project itself is a folder of JSX; saving recompiles it and refreshes the Diffusion Studio canvas. Verification checks more than a successful save: dapi check catches black frames, invisible clips, and missing media, while capture shows what the viewer will actually see. Export happens only when requested. The skills require Diffusion Studio and its bundled dapi CLI; macOS users can install the Homebrew cask, and the full editor can also run from source. The skills repository is MIT-licensed.