Behaviour for a self-contained, reusable AI task.

A prompt module encapsulates everything needed for a specific AI operation: what to ask (the prompts), how to ask it (the execution profile), and how to process the answer (post-processing).

Callbacks

• set_profile/1 — returns a LangchainPrompt.Profile that selects the adapter and model configuration for this task.
• generate_system_prompt/1 — returns the system prompt string, or nil to omit the system message (useful for models that don't support it).
• generate_user_prompt/1 — returns the user prompt string, or nil for conversational prompts where the last message is already in message_history.
• post_process/2 — transforms the raw LangchainPrompt.Message response into a domain value. Return {:ok, result} or {:error, reason}.

Example

defmodule MyApp.Prompts.Classify do
  @behaviour LangchainPrompt.Prompt

  @impl true
  def set_profile(_assigns), do: MyApp.AIProfiles.get(:fast)

  @impl true
  def generate_system_prompt(_assigns) do
    "Classify the sentiment as :positive, :neutral, or :negative. " <>
    "Reply with exactly one word."
  end

  @impl true
  def generate_user_prompt(%{text: text}), do: text

  @impl true
  def post_process(_assigns, %LangchainPrompt.Message{content: content}) do
    case String.trim(content) do
      "positive" -> {:ok, :positive}
      "neutral"  -> {:ok, :neutral}
      "negative" -> {:ok, :negative}
      other      -> {:error, {:unexpected_response, other}}
    end
  end
end