Schema

Schema

Schema is a key component of any plan. A Frags schema, is a JsonSchema that describes in great detail the shape of the data that will be produced by the LLM.

In addition to all the default fields you find in the specification, two Frags extensions are available:

• x-session (required): defines which session is in charge of generating that piece of data

Note: the x-session field can be applied ONLY at to the sub-fields of the root object.

Structure

At the very first level, the schema MUST be an object and all the children are required to have at least the x-session fields, referencing the name of an existing session.

Good practices

• fields should have a name that's evocative of the data they describe
• Use the description field to instruct the LLM on what the data is about. Don't bloat your prompts, use the data descriptions as an extension of the prompt!

YAML

Example:

schema:
  type: object
  required:
    - classification
    - other_family_members
  properties:
    classification:
      x-session: s1
      description: the classification of the animal
      type: object
      required:
        - scientific_name
        - common_name
        - taxonomy
      properties:
        scientific_name:
          type: string
        common_name:
          type: string
        taxonomy:
          type: object
          description: the full taxonomy of the animal
          required:
            - kingdom
            - phylum
            - class
            - order
          properties:
            kingdom:
              type: string
            phylum:
              type: string
            class:
              type: string
            order:
              type: string
            family:
              type: string
            genus:
              type: string
            species:
              type: string
    other_family_member:
      type: array
      x-session: s2
      description: other animals of the same family
      items:
        type: string

Go

Same example, but in Go. Here you may appreciate one extremely important distinction: while the CLI unmarshals to a generic data structure (ProgMap, a thread-safe incremental map), things can be different in Go, as you're free to unmarshal to any map or struct you want, as long as it's been properly described by the schema.

frags.Schema{
		Type:     "object",
		Required: []string{"classification", "other_family_members"},
		Properties: map[string]*frags.Schema{
			"classification": {
				Type:     "object",
				XSession: &session1,
				Required: []string{"scientific_name", "common_name", "taxonomy"},
				Properties: map[string]*frags.Schema{
					"scientific_name": {Type: "string"},
					"common_name":     {Type: "string"},
					"taxonomy": {
						Type:     "object",
						Required: []string{"kingdom", "phylum", "class", "order", "family", "genus", "species"},
						Properties: map[string]*frags.Schema{
							"kingdom": {Type: "string"},
							"phylum":  {Type: "string"},
							"class":   {Type: "string"},
							"order":   {Type: "string"},
							"family":  {Type: "string"},
							"genus":   {Type: "string"},
							"species": {Type: "string"},
						},
					},
				}},
			"other_family_members": {
				Type:     "array",
				XSession: &session2,
				Items:    &frags.Schema{Type: "string"},
			},
		},
}