Declarative YAML Configuration

Decouple metadata from Python code using YAML declarations.

Declarative Configuration (YAML)

For enterprise scenarios, separate protocol metadata from code using a landmarks.yaml file:

yaml

global_aliases:
  user_id: [uid, customer_id, user]

landmarks:
  - id: compute
    description: "Virtual machine management"
    type: navigation
    tools:
      - id: compute:toggle_power
        description: "Toggle VM power state"
        type: action
        remedy: "Check valid instance IDs with inspect_landmark"
        parameters:
          - name: instance_id
            type: string
            description: "The unique VM identifier"
            required: true
          - name: state
            type: string
            description: "Desired state: 'on' or 'off'"
            options: ["on", "off"]

Clean Python Decorators (Metadata Inheritance)

TIP

When using MetadataRegistry with a YAML file, you do not need to duplicate metadata in the Python decorators. The decorators only require the action_id. Elemm automatically inherits description, parameters, remedies, and instructions from the YAML file.

This separates the API declaration from the implementation, making your Python code extremely clean:

python

# The YAML file defines description, remedy, and parameter constraints.
# The Python decorator only binds the function to the ID.
@manager.landmark("compute:toggle_power")
async def toggle_power(instance_id: str, state: str):
    # Business logic here...
    return {"status": "success", "new_state": state}

Available YAML Fields

Field	Type	Description
`id`	`string`	Unique ID. Use `namespace:action_name` for hierarchy
`description`	`string`	Description shown to the agent
`type`	`string`	`"navigation"` (namespace) or `"action"` (executable)
`remedy`	`string`	SmartRepair hint on errors
`instructions`	`string`	Behavioral guidelines injected on inspection
`priority`	`int`	Sort weight (1 = highest priority)
`returns`	`string`	Description of the return value
`response_schema`	`object`	JSON Schema of the return object
`parameters`	`list`	Parameter definitions