Agent tools
The typed operations the Edit Agent can call, and which of them ask for approval first.
Every capability the agent has is one of these named tools in the main process. Nothing else is reachable from a conversation.
| The agent can | Tool | Approval |
|---|---|---|
| Read a project timeline and asset metadata | getProjectTimeline | Runs immediately |
| Watch footage as sampled frames it can see | watchProjectVideo | Runs immediately |
| Place a clip on the timeline | addClipToTimeline | Asks first |
| Trim a timeline clip | trimTimelineClip | Asks first |
| Restyle a clip's effects | updateClipEffects | Asks first |
| Generate speech | createSpeechJob | Asks first |
| Generate video | createVideoJob | Asks first |
| Follow a generation job | getJobStatus | Runs immediately |
| Import a finished generation into the project | importGeneratedResult | Asks first |
| Start a local export | exportProjectVideo | Asks first |
What is not in the list
- No shell or arbitrary command execution
- No filesystem access outside the project the conversation belongs to
- No provider credential access — keys stay in main-process safe storage
- No network calls except to a provider you connected, for a request you asked for
Because the tools are the same operations the interface uses, anything the agent does is something you could have done by hand, and shows up in the timeline the same way.