Skills¶

Narnia ships with agentic skills — instructions that let an LLM (such as GitHub Copilot CLI or Claude Code) perform complex tasks on your behalf.

Unlike MCP tools (which run inside the MCP server process), skills run in the LLM's own environment. This gives them full visibility into command output, adaptive error handling, and the ability to manage long-running processes like web servers.

Available Skills¶

Skill	Description
narnia-web-server	Start, stop, restart, and check status of the Narnia web UI

Installing as a Plugin¶

Narnia follows the standard plugin layout with plugin.json at the repo root.

Copilot CLI¶

copilot plugin install ncosentino/narnia

Claude Code¶

claude plugin install ncosentino/narnia

From a Local Clone¶

If you already have the repo checked out:

copilot plugin install ./path/to/narnia

Managing the Plugin¶

copilot plugin list              # List installed plugins
copilot plugin update narnia     # Update to latest version
copilot plugin uninstall narnia  # Remove the plugin

Once installed, skills are automatically available to the LLM. Just ask it to perform the task (e.g., "start the Narnia web server").

Why Skills Instead of MCP Tools?¶

The original open_narnia_ui MCP tool had fundamental reliability issues:

Blind fire-and-forget: The MCP tool launched dotnet run but couldn't see build errors or adapt to slow builds
Fixed timeout: A 20-second timeout was often too short for cold builds
No diagnostics: When it failed, neither the LLM nor the user could see what went wrong

Skills solve this by letting the LLM use its native shell tools — it can see build output, retry on failure, and manage processes with full control.