Agentic Video Editing
Let an AI agent transcribe your recording, cut the filler, generate animated B-roll, and apply scene layouts — all from a single conversation.
You can let a coding agent edit your ScreenKite project. The agent uses the same editing tools as the editor window, and each tool edit is one undo step, like a click. Use the agent you already have: Claude Code, ChatGPT Codex, Antigravity, Grok Build or Kimi.
This page covers agents that run in a terminal. To use an agent inside the editor, see AI Chat Assistant.
Set Up
1. Install the command line tool
Choose ScreenKite → Install Command Line Tool.... This installs the screenkite command. You can see its status in Settings → Editor → AI Agent Editing.
2. Set up transcription
Cuts by transcript need word timings. Set up a transcription engine in Settings → Captions before you ask for transcript cuts. See Word-Level Generated Captions.
3. Connect your agent
In the editor, click CLI / MCP in the top toolbar. Choose Show Connection Details to see how to connect your agent. Each open editor window runs its own local server on your Mac.
Ask Your Agent
You write a sentence. The agent calls ScreenKite tools for you. For example:
Open ~/Desktop/Recording.skbundle and transcribe the microphone track. Propose cuts for false starts and filler words. Show me the list before you apply it.
Add B-roll to the three main points of the talk. Use a corner layout.
What the Agent Calls
The CLI and the agent use the same commands. You can run them yourself:
# Open a project
screenkite project open --path ~/Desktop/Recording.skbundle --json
# List the tools and read one tool's options
screenkite tool list
screenkite tool describe --name edit-timeline
# Read the project state
screenkite tool call --name stat --input-json '{"scope":"summary"}' --json
Add --json to get output that a program can read.
Propose, Then Apply
Some actions take a confirm field:
| Tool | Action |
|---|---|
edit-timeline | cut |
edit-words | cutSilences, removeWords, removeDeadAir |
markers | generateChapters |
fx | captions/import, captions/patch, voiceover/replace, camera/replace |
With confirm:false, the action returns its plan and changes nothing. A second call with confirm:true applies the plan.
Other actions do not take confirm. For example, rm-track, rm-asset, edit-timeline delete and fx broll_slot/accept apply at once.
Transcript Cuts
The agent can:
- Check that transcription is ready.
- Transcribe the microphone track with word timings.
- Propose cuts for false starts, filler words and silences.
- Apply the cuts with
confirm:true.
The same word timings make captions. See Word-Level Generated Captions.
B-Roll and Layouts
The agent can propose B-roll slots on the timeline, make animated visuals with HyperFrames, and set scene layouts with layout-set. Proposing a slot and accepting it are two separate agent actions.
In the editor, a proposed slot shows as a B-Roll Slot card on the timeline. To replace a B-roll clip yourself, right-click it and choose Replace B-roll....
Skills
Skills teach an agent a full workflow. List the skills that ship with the app:
screenkite skills
For community workflows and prompts, see github.com/ScreenKite/awesome-ai-video-editing. Install its skills with:
npx skills add ScreenKite/awesome-ai-video-editing