Struct Focus 

pub struct Focus { /* private fields */ }

A focused LLM call.

Each instance is bound to a system prompt and dedicated to one specific judgment question. Use ask() to send input and receive a structured JSON answer.

Usage

// Simple case: single string input
let classify = Focus::new(client, "You are a task complexity classifier...");
let output = classify.ask::<TaskComplexity>(&user_input, 5s).await?;

// Complex case: multiple context fields
let status_focus = Focus::new(client, "You are a task status judge...");
let ctx = Context::new()
    .add("command", command)
    .add("screen", screen);
let output = status_focus.ask::<TaskStatus>(&ctx, 5s).await?;

Implementations

impl Focus

pub fn new( client: Arc<dyn LlmProvider>, system_prompt: impl Into<String>, ) -> Focus

Create a new Focus instance.

• client: LLM client (shared; multiple Focus instances can reuse the same client)
• system_prompt: The role and judgment rules for this Focus (bound at creation, never changes)

pub async fn ask<T>( &self, input: &(impl FocusInput + ?Sized), timeout: Duration, ) -> Result<FocusOutput<T>, FocusError>

where T: DeserializeOwned,

Make a focused LLM call.

Sends the system prompt (bound at creation) + user input (this call), forces JSON output, and deserializes into T.

Arguments

• input: User input — can be &str or Context
• timeout: Call timeout

Returns

FocusOutput<T> containing the structured result and raw response.

Trait Implementations

impl Clone for Focus

fn clone(&self) -> Focus

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Focus

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

Formats the value using the given formatter.