bamboo-agent 2026.4.2

A fully self-contained AI agent backend framework with built-in web services, multi-LLM provider support, and comprehensive tool execution
Documentation
//! Per-round prelude helpers for the agent loop runner.

use tokio_util::sync::CancellationToken;

use crate::agent::core::tools::ToolExecutor;
use crate::agent::core::{AgentError, Session};
use crate::agent::loop_module::config::AgentLoopConfig;
use crate::agent::loop_module::task_context::TaskLoopContext;
use crate::agent::metrics::MetricsCollector;

mod cancellation;
mod prompt_updates;
mod round_state;

use cancellation::ensure_not_cancelled;
use prompt_updates::refresh_round_prompt_context;
use round_state::{build_round_id, log_round_start, update_task_round_state};

pub(super) async fn prepare_round(
    session: &mut Session,
    task_context: &mut Option<TaskLoopContext>,
    round: usize,
    max_rounds: usize,
    cancel_token: &CancellationToken,
    metrics_collector: Option<&MetricsCollector>,
    session_id: &str,
    model_name: &str,
    debug_enabled: bool,
    config: &AgentLoopConfig,
    _tools: &dyn ToolExecutor,
) -> Result<String, AgentError> {
    let _ = (config, session_id);
    refresh_round_prompt_context(session).await;
    update_task_round_state(task_context, round, max_rounds);

    let round_id = build_round_id(session_id, round);

    log_round_start(
        debug_enabled,
        session_id,
        round,
        max_rounds,
        session.messages.len(),
    );

    ensure_not_cancelled(
        cancel_token,
        metrics_collector,
        session_id,
        session.messages.len(),
    )?;

    super::metrics_lifecycle::record_round_started(
        metrics_collector,
        &round_id,
        session_id,
        model_name,
    );

    Ok(round_id)
}

#[cfg(test)]
mod tests;