Skip to content

Skill Commands

Overview

Skill commands allow skills defined in your skills directory to be exposed as slash commands across ACP, AG-UI, and OpenCode protocols. This enables direct skill invocation via protocol-native interfaces.

What Are Skill Commands?

Skills are reusable instruction sets stored in SKILL.md files. With skill commands, these become directly invocable via:

  • ACP: AvailableCommand[] in capabilities
  • AG-UI: Tools with skill__ prefix
  • OpenCode: slashed Commands with skill: prefix

Configuration

Skills are automatically discovered from your skills directory and exposed as commands. No additional configuration is required.

Example SKILL.md

# Skill: my-skill
A description of what this skill does

## License
MIT

## Compatibility
1.0.0

## Allowed Tools
bash, read, grep

## Instructions
Detailed instructions for the agent...

Protocol-Specific Usage

ACP Protocol

Skills appear as AvailableCommand via the session/update notification with available_commands_update after session creation:

{
  "sessionId": "sess_abc123",
  "update": {
    "sessionUpdate": "available_commands_update",
    "availableCommands": [
      {
        "name": "my-skill",
        "description": "A description...",
        "input": {"hint": "Arguments for skill"}
      }
    ]
  }
}

Note: Per the ACP specification, available commands are declared via session/update after session creation, not in the initialize response.

AG-UI Protocol

Skills appear as Tools with skill__ prefix:

{
  "name": "skill__my-skill",
  "description": "A description...",
  "parameters": {
    "type": "object",
    "properties": {
      "arguments": {"type": "string"}
    }
  }
}

OpenCode Protocol

Skills appear as Commands with skill: prefix:

/skill:my-skill arguments here

Troubleshooting

Skills not appearing

  • Ensure SKILL.md files are valid
  • Check skills directory path in config
  • Verify skills have required metadata (name, description)

Command not executing

  • Check allowed_tools in SKILL.md
  • Verify skill instructions are valid