Built for AI Agents, end to end.
skills.video ships a full REST API plus the skillsvideo agent-native CLI. Every provider, every model — callable from HTTP, any script, or dropped straight into an agent runtime.
Ships with first-class support for the agents you already run.
Anthropic's terminal-native coding agent.
OpenAI's autonomous code agent for your shell.
The AI that actually does things.
An agent that grows with you.
Open-source OpenClaw desktop app for Feishu.
An agentic system with specialized processes for thinking, working, and remembering.
Pick your install method
Choose the shell you actually work in: Terminal on macOS, Linux bash, native Windows PowerShell, or Git Bash / WSL on Windows.
After install, run skillsvideo --version to verify.
Three commands to confirm everything works
PKCE OAuth via a 127.0.0.1 loopback. No tokens to copy-paste, no API keys to rotate by hand.
# pkce + loopback callback
Every provider and model on the skills.video dashboard, addressable from your shell.
# parity with web app
Tokens refresh in the background. Revoke a session from the dashboard at any time.
# auto-refresh · revocable
Examples
Real one-liners for the things you do every day on skills.video.
Every job goes through the same generate command — pick a text-to-video model and ship a clip in one line.
Pass a local image as the first frame and add a motion prompt — the CLI uploads it and waits for the run.
Same generate command, just point --model at an image model. There is no separate image subcommand.
Combine --wait and --output to use the CLI as a one-liner inside any script or CI job.
More than generate
The CLI also handles default model selection, agent setup, task follow-up, and headless auth for CI.
Pin the image and video models you reach for most. After that, generate can fall back to your first preferred model automatically.
Install for the current agent runtime first, or write into Claude Code and Codex roots in one shot. The bundled SKILL.md also fits OpenClaw-style setups.
List recent cached runs when you need to recover a task id, then fetch the latest state for a specific job.
Browser login for humans, API keys for headless runners. Save a token once, then verify the active account before a build starts.
Browser login, end to end
The CLI never sees your password. Authorization happens on skills.video in your browser and the code is exchanged with PKCE.
The CLI binds a local callback on 127.0.0.1 and asks the API for an authorization URL.
You land on skills.video, confirm which CLI and workspace you are authorizing, then approve.
The browser redirects back to the loopback callback. The CLI verifies state, exchanges the code with PKCE, and stores a refreshable credential.
Ready to ship from your terminal?
Install once, sign in from any browser, and skills.video is at your prompt.