dot-prompt

Response Contracts

Response contracts define the expected output structure from your LLM. They version together with your prompt and are derived automatically from JSON blocks.

Basic Response Contract

response do
  {
    "response_type": "greeting",
    "content": "..."
  }
end response

The compiler derives the schema from the JSON structure: 

{
  "type": "object",
  "properties": {
    "response_type": { "type": "string", "enum": ["greeting"] },
    "content": { "type": "string" }
  },
  "required": ["response_type", "content"]
}

Nested Objects

response do
  {
    "user": {
      "name": "...",
      "preferences": {
        "theme": "..."
      }
    }
  }
end response

Multiple Response Blocks

You can have multiple response blocks in different branches. The compiler validates compatibility: 

if @is_premium is true do
Premium content here.

response do
  {
    "content": "...",
    "premium_features": ["a", "b", "c"]
  }
end response

else
Free content here.

response do
  {
    "content": "..."
  }
end response

end @is_premium

Compatibility Rules

  • Compatible — Fields in common have matching types
  • Warning — Some fields differ but are compatible
  • Error — Incompatible types for the same field

Using Variables

Reference variables in response contracts: 

response do
  {
    "response_type": "teaching",
    "step": @current_step,
    "next_step": @current_step + 1
  }
end response

Contract Validation

The derived schema is returned with every compiled prompt. Your application can validate LLM responses against the contract: 

{
  "prompt": "...",
  "response_contract": {
    "type": "object",
    "properties": {
      "response_type": { "type": "string" },
      "content": { "type": "string" }
    }
  }
}

Benefits

  • Automatic schema derivation
  • Versioned with the prompt
  • Breaking change detection
  • Runtime validation support

Next Steps