short-drama-edit takes the shots generated for an episode and assembles them into one finished video. Every cut is written in an edit list with its source, in and out points, the reason for each trim, subtitles copied word for word from the script, sound and on-screen text.
Three commands do the technical work with ffmpeg: check validates the edit list against the files and the script, render builds the video with subtitles and unified loudness, and verify measures the result and flags jumps at cut points for a human to review. It never changes the approved script or prompts. The skill’s text is in Chinese.
What You Get
- An edit list Markdown file with a delivery spec and one entry per cut
- A rendered episode with burned-in subtitles, on-screen text layers, sound effects, dubbing and ambience
- A check report of timing, source IDs, media readability, aspect ratio and script-exact subtitle text
- A verify report of total duration, loudness, true peak, and flagged brightness or volume jumps at cuts
- Intermediate segment files kept next to the final video
Common Use Cases
- Assemble the clips of one episode into a vertical cut with subtitles and normalized loudness
- Try a partial cut of a single scene before the full episode is generated
- Add system-panel or task-panel style on-screen text to a game-style drama
What Makes It Unique
Each trim needs a written reason tied to a specific frame action, and subtitles must match the script text exactly, with transcription used only to find timing. The checker blocks mixed aspect ratios and overlapping source ranges unless a replay is justified, and the skill will not edit approved scripts or prompts; gaps go back to the stage that owns them.
Under The Hood
A bundled Python script, edit_tool.py, runs the work on the local machine and requires ffmpeg and ffprobe on the path. It reads clips, images and audio from the episode’s output folder and writes segment files and the final video under the episode folder. For on-screen text or Remotion subtitles it uses a bundled Remotion workspace with Node, which needs its font packages installed with npm first. The agent also edits the edit list file and runs shell commands. The instructions do not send project data to web services.