Struct Turn 

pub struct Turn {
    pub role: Role,
    pub timestamp: Timestamp,
    pub model: Option<String>,
    pub effort: Option<String>,
    pub content: Content,
    pub usage: Option<Usage>,
    pub is_sidechain: bool,
    pub is_meta: bool,
}

A conversation turn, with the role lifted out of the JSONL’s type tag.

Fields

role: Role

timestamp: Timestamp

model: Option<String>

effort: Option<String>

How hard the model was asked to think, where the harness records it: a refinement of the model, not a separate fact about the turn.

content: Content

usage: Option<Usage>

What the turn cost, for the assistant turns that report it. A user turn has none, and transcripts written before the harness recorded usage carry none either.

is_sidechain: bool

True for turns belonging to a subagent running inside this session.

is_meta: bool

True for turns the harness injected rather than the user typing them.

Implementations

impl Turn

pub fn is_tool_response(&self) -> bool

True when a user-role turn carries only tool results: the harness returning tool output, not the user typing. These read as a continuation of the assistant’s turn, not a message of their own.

pub fn is_clear_command(&self) -> bool

True when this turn is the /clear slash command, which resets the context: a session boundary the harness records as a user turn, with no conversation of its own worth showing.

Trait Implementations

impl Debug for Turn

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

Formats the value using the given formatter.