LemonAgent.Types.AgentToolResult (lemon_agent v0.1.0)

Result returned from tool execution.

Contains content blocks (text/images) to be displayed, optional details for UI rendering or logging, and trust metadata.

:content is what the model sees — it becomes the tool-result message in the transcript, so it should read as an answer rather than as a status line. :details is not sent to the model: it is for the UI and the logs, and is the right place for structured data (ids, counts, the raw response) that would only cost tokens in :content.

:trust marks content the tool did not author. Anything fetched from the network, read from a user-supplied file, or returned by a third-party API is :untrusted, which tells downstream rendering that instructions inside it must not be obeyed. LemonAgent.Security.ExternalContent wraps such content with the right framing and metadata — use it rather than setting the field by hand.

Summary

Types

content_block()

t()

Types

content_block()

@type content_block() ::
  LemonAi.Types.TextContent.t() | LemonAi.Types.ImageContent.t()

t()

@type t() :: %LemonAgent.Types.AgentToolResult{
  content: [content_block()],
  details: any(),
  trust: LemonAi.Types.trust_level()
}