bamboo_engine/runtime/task_evaluation/
executor.rs

1use std::sync::Arc;
2
3use tokio::sync::mpsc;
4
5use bamboo_agent_core::{AgentError, AgentEvent, Session};
6use bamboo_domain::ReasoningEffort;
7use bamboo_domain::TaskItemStatus;
8use bamboo_infrastructure::{LLMProvider, LLMRequestOptions};
9
10use super::super::task_context::TaskLoopContext;
11use super::message_builder::build_task_evaluation_messages;
12use super::schema::get_task_evaluation_tools;
13use super::token_estimation::estimate_prompt_tokens;
14use super::TaskEvaluationResult;
15
16mod outcomes;
17
18fn has_tool_activity(ctx: &TaskLoopContext) -> bool {
19    ctx.items.iter().any(|item| !item.tool_calls.is_empty())
20}
21
22fn skipped_evaluation(reasoning: &str) -> TaskEvaluationResult {
23    TaskEvaluationResult {
24        needs_evaluation: false,
25        updates: Vec::new(),
26        reasoning: reasoning.to_string(),
27        prompt_tokens: 0,
28        completion_tokens: 0,
29    }
30}
31
32fn normalize_lightweight_reasoning_effort(
33    reasoning_effort: Option<ReasoningEffort>,
34) -> Option<ReasoningEffort> {
35    reasoning_effort.map(|effort| match effort {
36        ReasoningEffort::Xhigh | ReasoningEffort::Max => ReasoningEffort::High,
37        other => other,
38    })
39}
40
41/// 执行 TaskList 评估
42pub async fn evaluate_task_progress(
43    ctx: &TaskLoopContext,
44    session: &Session,
45    llm: Arc<dyn LLMProvider>,
46    event_tx: &mpsc::Sender<AgentEvent>,
47    session_id: &str,
48    model: &str,
49    reasoning_effort: Option<ReasoningEffort>,
50) -> Result<TaskEvaluationResult, AgentError> {
51    use crate::runtime::stream::handler::consume_llm_stream_silent;
52
53    let in_progress_count = ctx
54        .items
55        .iter()
56        .filter(|item| matches!(item.status, TaskItemStatus::InProgress))
57        .count();
58
59    if in_progress_count == 0 {
60        return Ok(skipped_evaluation("No in-progress tasks to evaluate"));
61    }
62
63    let is_plan_mode = session
64        .agent_runtime_state
65        .as_ref()
66        .and_then(|s| s.plan_mode.as_ref())
67        .is_some();
68
69    if !is_plan_mode && !has_tool_activity(ctx) {
70        return Ok(skipped_evaluation(
71            "No tool executions yet; skipping task evaluation.",
72        ));
73    }
74
75    tracing::info!(
76        "[{}] Evaluating {} in-progress task items",
77        session_id,
78        in_progress_count
79    );
80
81    let _ = event_tx
82        .send(AgentEvent::TaskEvaluationStarted {
83            session_id: session_id.to_string(),
84            items_count: in_progress_count,
85        })
86        .await;
87
88    let messages = build_task_evaluation_messages(ctx, session);
89    let prompt_tokens = estimate_prompt_tokens(&messages);
90    let tools = get_task_evaluation_tools();
91
92    // Use model from parameter (passed from config), not from session.
93    tracing::debug!("[{}] Task evaluation using model: {}", session_id, model);
94
95    let request_reasoning_effort = normalize_lightweight_reasoning_effort(reasoning_effort);
96    if request_reasoning_effort != reasoning_effort {
97        tracing::debug!(
98            "[{}] Task evaluation downgraded reasoning effort from {:?} to {:?} for lightweight request",
99            session_id,
100            reasoning_effort,
101            request_reasoning_effort
102        );
103    }
104
105    let request_options = LLMRequestOptions {
106        session_id: Some(session_id.to_string()),
107        reasoning_effort: request_reasoning_effort,
108        parallel_tool_calls: None,
109        responses: None,
110    };
111    match llm
112        .chat_stream_with_options(&messages, &tools, Some(500), model, Some(&request_options))
113        .await
114    {
115        Ok(stream) => {
116            let stream_output = consume_llm_stream_silent(
117                stream,
118                &tokio_util::sync::CancellationToken::new(),
119                session_id,
120            )
121            .await
122            .map_err(|error| AgentError::LLM(error.to_string()))?;
123
124            Ok(
125                outcomes::build_success_result(stream_output, event_tx, session_id, prompt_tokens)
126                    .await,
127            )
128        }
129        Err(error) => {
130            tracing::warn!("[{}] Task evaluation failed: {}", session_id, error);
131            Ok(skipped_evaluation(&format!("Evaluation failed: {}", error)))
132        }
133    }
134}
bamboo_engine/runtime/task_evaluation/executor.rs

bamboo_engine/runtime/task_evaluation/
executor.rs
