Your timeline, from a prompt.
BetterEdits writes a guide into every workspace, so Claude Code can read the footage, cut, fill placeholders, caption and export while you watch the app do it.
Three decisions that make it real.
Agents are only useful when the tool is honest about what it can do.
- 01
Every workspace teaches the agent
When you create a workspace, the app writes a guide and a CLAUDE.md into it: the three files a video is made of, every command, how tracks, clips, text and captions behave, and the house rules. Open Claude Code there and it already knows what to do.
- 02
The files first
A video is three JSON files written for reading. The agent reads them, writes the edit in one go, lints it without the app, and the app takes it as one undo step. The commands do the single precise steps.
- 03
Same rules as the window
The agent works through the same code the window uses. It cannot reach a state the window can't, and it cannot skip a confirmation the window would ask for.
Cuts, captions and structures.
The editor's three pillars from a prompt: cut on the transcript, fill a structure, caption it. Every change is one undo step in the running app.
- >Read the transcript of Video 1, then cut out the pauses longer than a second. Leave a little air around each cut and show me a contact sheet when you are done.
- >Add a title in the Strong favorite over the first two seconds that says "Stop doing this", near the top. Then put subtitles on the piece with the default style at the bottom.
- >Here is a template for a three-shot reel: hook, demo, call to action, each five seconds, with music underneath. Write the JSON, import it as a new project called Hook v2, and open it.
- >Read the analysed footage in the Footage folder and lay the clips where someone is talking to the camera on Video 1, in the order they were shot.
- >Open timeline.json, move the title so it starts on the second shot, and shorten every b-roll clip to two seconds. Show me a contact sheet.
An agent can only do what you could do.
The agent works through the same code as the window, so it cannot reach a state the window can't, and it cannot skip a confirmation the window would ask for.
- Undo in the editor
Every change is one step in the app's undo stack, even a rewrite of the whole timeline.
- Reads, then looks
The agent reads the analysis next to the footage when there is one, and draws itself frames and contact sheets to check the cut.
- Works in the running app
The commands drive the open app and launch it in the background when it is not running.
- No keys in the folder
Claude Code brings its own model. The one key the app keeps, for analysing footage, lives in the app's own settings and never in a workspace.
Agent questions
- Does the app have to be open?
- The files can be edited without it. The commands talk to the running app and launch it in the background when it is not running.
- What can it break?
- The same things you can from the window, and each change is one undo away. The reader is strict: a broken file is refused with the place, and the app keeps its copy until it is fixed.
- Does it only work with Claude Code?
- Claude Code is what we build for and test with. Any agent that reads a CLAUDE.md-style guide works the same way.
- Do I need API keys?
- Not for the app. Claude Code uses your own Claude subscription or key, as it does for anything else.
The whole contract is in the manual: Claude Code and the command line.
Open Claude Code in a folder of footage.
BetterEdits is free, and the guide is already in the folder.