Triggers

The triggers block defines slash commands that users can invoke interactively. Triggers let users kick off specific agents, run scripts, or start sub-flows on demand.

Basic Syntax

triggers {
  /deploy {
    pattern: "/deploy"
    argument: "ENVIRONMENT"
  }

  /review {
    pattern: "/review"
  }
}

Field Reference

Argument Format

Arguments use UPPERCASE names to indicate a required user-provided value:

triggers {
  /deploy {
    pattern: "/deploy"
    argument: "ENVIRONMENT"
  }
}

When the user types /deploy production, the value production is passed as the ENVIRONMENT argument.

Multiple Triggers

Define as many triggers as your topology needs:

triggers {
  /analyze {
    pattern: "/analyze"
    argument: "FILE_PATH"
  }

  /fix {
    pattern: "/fix"
    argument: "ISSUE_ID"
  }

  /status {
    pattern: "/status"
  }
}

Connecting Triggers to Agents

Triggers work with agents that have invocation: manual. When a trigger fires, it activates the associated agent:

topology dev-tools : [pipeline] {
  agent analyzer {
    model: sonnet
    tools: [Read, Grep]
    invocation: manual
  }

  agent fixer {
    model: sonnet
    tools: [Read, Write]
    invocation: manual
  }

  triggers {
    /analyze {
      pattern: "/analyze"
      argument: "FILE_PATH"
    }

    /fix {
      pattern: "/fix"
      argument: "ISSUE_ID"
    }
  }
}

Tips

• Trigger patterns always start with /.
• Argument names should be descriptive and use UPPERCASE with underscores (e.g. FILE_PATH, BRANCH_NAME).
• Triggers are especially useful in interactive environments like Claude Code where users can type commands directly.
• An agent with invocation: auto does not need a trigger — it runs automatically as part of the flow.