Tracking Premiere, CapCut & External Renders
Use motionmind track to autolog renders from your existing video editing tools into the version registry.
MotionMind doesn't come with its own AI. It's a skill your agent reads, plus a CLI it runs. Claude Code, Codex, Gemini CLI, OpenCode: whichever you already use, it follows the same four gates and the same rules, because they come from the same SKILL.md.
Where the Skill Lives
motionmind setup already installs the skill into the new workspace. Re-run it after an update, or install it for your whole user account:
motionmind skill install # this workspace: .agents/skills + a link in .claude/skills
motionmind skill install --global # your home folder: ~/.agents/skills and ~/.claude/skills
The workspace install copies the skill to .agents/skills/motionmind/ and links .claude/skills/motionmind to it, so there's one copy. The workspace also gets an AGENTS.md and a CLAUDE.md that tell any agent opened there to use the skill. motionmind doctor shows whether the skill is installed and which agent CLIs it finds on your PATH; on this machine, all four.
Where each agent looks:
| Agent | Where it looks |
|---|---|
| Claude Code | .claude/skills/motionmind in the project, or ~/.claude/skills/motionmind |
| Codex CLI · OpenCode | .agents/skills/motionmind or ~/.agents/skills/motionmind, or a reference in AGENTS.md |
| Gemini CLI · Cursor · others | a copy in the project, referenced from GEMINI.md, .cursor/rules or the system prompt |
For Gemini CLI, a short GEMINI.md in the workspace is enough:
<!-- GEMINI.md in the workspace -->
For any video task, follow .agents/skills/motionmind/SKILL.md and run its commands from this folder.
Install one copy per scope: the workspace or your home folder, not both. skill install won't overwrite a folder it didn't create unless you pass --force. The skill also needs the motionmind command on your PATH, because the agent runs it.
Start the Agent in the Workspace
cd my-studio
claude # or: codex · gemini · opencode
The agent reads the skill from the folder it starts in, so open it in the workspace, the folder with motionmind.store.json. Keep motionmind ui running in another terminal to watch the results.
Let the Studio Write the First Prompt
Starter kit in MotionMind Studio turns a short form into a prompt: new video, new version, revision, or Instagram feed. Fill it in, press Copy prompt, paste it into any of the agents. This is the prompt from the recording:
Use the motionmind skill: new video, follow all four gates, HyperFrames engine.
NEW campaign: demo/morning-15s, run v1-first-cut
Goal / key message: Hot coffee before 7 a.m.
Audience: Morning commuters
CTA: demo.example
Format: 9:16 (1080×1920, Reels/TikTok), 15 seconds
Audio: music + SFX, no VO
Materials: brand assets in shared/brands/demo/ (brand.json, logos, fonts, captures); product repos are read-only. Music from `motionmind music`, SFX from `motionmind sfx list`.
Start with 3 distinct concepts in plan/concepts.md, then wait for my pick (Gate 1).
The kits are Markdown templates in the package's starter-kits/ folder; edit one, or add a .md file to make your own.
Read What the Agent Reads
Docs in the Studio shows the skill and its references (workflow, HyperFrames, motion, sound, QA, screen demos) exactly as the agent reads them. When the agent does something you didn't expect, this is where to look.
Updating
After a new release, update the CLI, then the skill copy:
npm i -g motionmind@latest
motionmind skill install
Add --global to the second command if that's where you installed it.