Enum Block 

#[non_exhaustive]
pub enum Block {
    Text(String),
    FileRef {
        path: String,
        hash: Option<String>,
        excerpt: Option<String>,
    },
    Skill {
        name: String,
        body: String,
    },
    ToolCall {
        call_id: String,
        name: String,
        args: Value,
    },
    ToolResult {
        call_id: String,
        content: Value,
    },
    Feedback(Vec<Signal>),
    Reasoning(String),
    Image {
        media_type: String,
        base64: String,
    },
    Audio {
        media_type: String,
        base64: String,
    },
}

A single block of content within the assembled prompt.

Blocks are grouped so that long-stable prefixes (system + guides) stay cacheable across turns (“prompt caching” pattern).

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

Text(String)

Plain prompt text.

FileRef

Reference to a file in the world. The runtime decides whether to inline contents or hand the agent a tool call to read it.

Fields

path: String

hash: Option<String>

excerpt: Option<String>

Skill

Reference to an activated SKILL.md body.

Fields

name: String

body: String

ToolCall

A tool call the assistant requested.

Fields

call_id: String

name: String

args: Value

ToolResult

The result of a previous tool call.

Fields

call_id: String

content: Value

Feedback(Vec<Signal>)

Feedback signals from sensors, rendered for the model.

Reasoning(String)

Provider-specific reasoning trace (DeepSeek reasoning_content, Anthropic thinking blocks). Must be echoed back to the provider on subsequent calls or the API rejects the request.

Image

An inline image for vision-capable models. media_type is a MIME type (e.g. "image/png", "image/jpeg"); base64 is the standard-base64 encoding of the raw image bytes. Each provider adapter renders this into its own multimodal wire shape (OpenAI image_url data-URI, Anthropic image/base64 source, Gemini inline_data).

Fields

media_type: String

base64: String

Audio

Inline audio for models that listen. media_type is a MIME type (e.g. "audio/wav", "audio/webm"); base64 is the standard-base64 encoding of the raw audio bytes. Each provider adapter renders this into its own wire shape (OpenAI input_audio with a bare format name, Gemini inline_data).

This is what a transcript cannot carry: intonation, stress, whether two words ran together, whether a vowel was long. Speech transcribed to text before the model sees it has already lost the part a pronunciation judgement rests on.

Fields

media_type: String

base64: String

Implementations

impl Block

pub fn image_bytes(media_type: impl Into<String>, bytes: &[u8]) -> Self

Build a Block::Image from raw image bytes, base64-encoding them. media_type is a MIME type like "image/png".

pub fn audio_bytes(media_type: impl Into<String>, bytes: &[u8]) -> Self

Build a Block::Audio from raw audio bytes, base64-encoding them. media_type is a MIME type like "audio/wav".

Trait Implementations

impl Clone for Block

fn clone(&self) -> Block

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Block

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for Block

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for Block

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.