Agents / GitHub Copilot

oxframe in GitHub Copilot

GitHub Copilot writes the scenes, runs oxframe check and fixes what the report names, then renders the video. Three steps set it up.

1. Install the CLI and the skills

cargo install oxframe-cli
npx skills add oxygenframe/oxframe-sdk

Point Copilot at the project notes: add a line to .github/copilot-instructions.md that says to follow AGENTS.md. The skills add workflows for 11 kinds of video on top of the main oxframe skill.

2. Connect the MCP server

MCP server in .vscode/mcp.json:

{
  "servers": {
    "oxframe": { "type": "stdio", "command": "oxframe", "args": ["mcp"] }
  }
}

The server gives GitHub Copilot check, inspect, render_frame, render_graph, render, list_components and docs_search as tools.

3. Ask for a video

Make a 20-second explainer of binary search: a title, a sorted array where
the search narrows to the middle step by step, and an end card. Run oxframe
check until it is clean, then render it.

Start in an empty folder: the agent runs oxframe init. For other kinds of video see the skills; the gallery has a brief for every example.

When it goes wrong

• A compile error the agent loops on. The project notes list the usual ones: a bind closure that needs move and cloned signals, a closure ending in ; that returns no animation, all((a, b)) with double parentheses.
• The report is clean but the picture is wrong. Ask the agent to open out/check/storyboard.png, or to render one frame with oxframe snapshot --at 2.5.
• The agent starts oxframe dev. That is a server for you, not for the agent; the project notes tell it to use check instead.