use std::sync::Arc;
use std::sync::atomic::{AtomicUsize, Ordering};
use echo_agent::human_loop::{HumanLoopEvent, HumanLoopManager};
use echo_agent::prelude::*;
use echo_agent::testing::MockLlmClient;
use echo_agent::tool;
#[tool(name = "add", description = "两数相加")]
async fn add(a: f64, b: f64) -> Result<ToolResult> {
Ok(ToolResult::success(format!("{} + {} = {}", a, b, a + b)))
}
#[tool(name = "subtract", description = "两数相减")]
async fn subtract(a: f64, b: f64) -> Result<ToolResult> {
Ok(ToolResult::success(format!("{} - {} = {}", a, b, a - b)))
}
#[tool(name = "multiply", description = "两数相乘")]
async fn multiply(a: f64, b: f64) -> Result<ToolResult> {
Ok(ToolResult::success(format!("{} * {} = {}", a, b, a * b)))
}
#[tool(name = "divide", description = "两数相除")]
async fn divide(a: f64, b: f64) -> Result<ToolResult> {
if b == 0.0 {
return Ok(ToolResult::error("除数不能为 0".to_string()));
}
Ok(ToolResult::success(format!("{} / {} = {}", a, b, a / b)))
}
fn math_tools() -> Vec<Box<dyn Tool>> {
vec![
Box::new(AddTool),
Box::new(SubtractTool),
Box::new(MultiplyTool),
Box::new(DivideTool),
]
}
fn spawn_human_loop_worker(
manager: Arc<HumanLoopManager>,
input_count: Arc<AtomicUsize>,
) -> tokio::task::JoinHandle<()> {
tokio::spawn(async move {
while let Some(event) = manager.recv_event().await {
match event {
HumanLoopEvent::ApprovalRequest {
tool_name,
prompt,
responder,
..
} => {
println!("🔔 审批请求: {tool_name} -> {prompt}");
responder.approve();
}
HumanLoopEvent::InputRequest { prompt, responder } => {
input_count.fetch_add(1, Ordering::Relaxed);
println!("💬 Human-in-the-loop: {prompt}");
responder.respond("用户补充：今天下雨，请按雨天折扣计算。".to_string());
}
}
}
})
}
#[tokio::main]
async fn main() -> Result<()> {
tracing_subscriber::fmt()
.with_max_level(tracing::Level::INFO)
.init();
println!("🧪 demo04 - SubAgent 编排 + Human-in-the-Loop 演示\n");
let human_loop = Arc::new(HumanLoopManager::new());
let input_count = Arc::new(AtomicUsize::new(0));
let worker = spawn_human_loop_worker(human_loop.clone(), input_count.clone());
let weather_llm = Arc::new(
MockLlmClient::new()
.with_model_name("weather-agent-mock")
.then_tool_call(
"weather-hitl-1",
"human_in_loop",
r#"{
"reasoning":"用户没有说明购物当天是否下雨，无法决定是否触发雨天 85 折。",
"approval_type":"LLM",
"tool":"weather-context"
}"#,
)
.then_tool_call(
"weather-final-1",
"final_answer",
r#"{"answer":"已通过 human_in_loop 向用户确认：今天下雨，因此后续计算需要叠加雨天 85 折。"}"#,
),
);
let math_llm = Arc::new(
MockLlmClient::new()
.with_model_name("math-agent-mock")
.then_tool_call("math-1", "multiply", r#"{"a":10,"b":15}"#)
.then_tool_call("math-2", "multiply", r#"{"a":16,"b":8}"#)
.then_tool_call("math-3", "multiply", r#"{"a":2,"b":98}"#)
.then_tool_call("math-4", "multiply", r#"{"a":500,"b":0.8}"#)
.then_tool_call("math-5", "add", r#"{"a":150,"b":128}"#)
.then_tool_call("math-6", "add", r#"{"a":278,"b":196}"#)
.then_tool_call("math-7", "add", r#"{"a":474,"b":400}"#)
.then_tool_call("math-8", "multiply", r#"{"a":874,"b":0.9}"#)
.then_tool_call("math-9", "multiply", r#"{"a":786.6,"b":0.85}"#)
.then_tool_call("math-10", "subtract", r#"{"a":1000,"b":668.61}"#)
.then_tool_call(
"math-final-1",
"final_answer",
r#"{"answer":"计算结果：本子 150 元，笔 128 元，玩具 196 元，衣服满 500 打八折后为 400 元；小计 874 元，满 800 打九折后为 786.6 元；已确认下雨，再打 85 折，实付 668.61 元，最终剩余 331.39 元。"}"#,
),
);
let orchestrator_llm = Arc::new(
MockLlmClient::new()
.with_model_name("orchestrator-mock")
.then_tool_call(
"main-1",
"agent_tool",
r#"{
"agent_name":"weather-agent",
"task":"先向用户确认购物当天是否下雨，并只返回与折扣判断相关的结论。"
}"#,
)
.then_tool_call(
"main-2",
"agent_tool",
r#"{
"agent_name":"math-agent",
"task":"预算是 1000 元。已确认今天下雨。请按先单品折扣、再满减折扣、最后雨天折扣的顺序计算剩余金额，并返回清晰的计算结论。"
}"#,
)
.then_tool_call(
"main-final-1",
"final_answer",
r#"{"answer":"weather-agent 先通过 human_in_loop 完成了天气澄清，确认今天下雨；math-agent 随后完成折扣计算，最终实付 668.61 元，还剩 331.39 元。"}"#,
),
);
let weather_agent = ReactAgentBuilder::new()
.name("weather-agent")
.model("weather-agent-mock")
.system_prompt(
"你负责识别是否缺少天气上下文。若用户未说明是否下雨，必须先调用 human_in_loop 获取补充信息，再给出简洁结论。",
)
.enable_tools()
.enable_human_in_loop()
.approval_provider(human_loop.clone() as Arc<dyn echo_agent::human_loop::HumanLoopProvider>)
.llm_client(weather_llm)
.max_iterations(6)
.build()?;
let mut math_agent = ReactAgentBuilder::new()
.name("math-agent")
.model("math-agent-mock")
.system_prompt("你是折扣计算专家，只使用数学工具逐步计算，再给出最终金额。")
.enable_tools()
.llm_client(math_llm)
.max_iterations(16)
.build()?;
math_agent.add_tools(math_tools());
let mut main_agent = ReactAgentBuilder::new()
.model("orchestrator-mock")
.name("main_agent")
.system_prompt(
"你是主编排 Agent。遇到缺失上下文的专业问题先分派给对应 subagent，再汇总结果，不要自己直接计算。",
)
.role(AgentRole::Orchestrator)
.enable_subagent()
.llm_client(orchestrator_llm)
.max_iterations(8)
.enable_tools()
.build()?;
main_agent.register_agent(Box::new(weather_agent));
main_agent.register_agent(Box::new(math_agent));
let result = main_agent
.execute(
"我有1000元，要买 10 个 15 元的本子、16 个 8 元的笔、2 个 98 元的玩具和一件 500 元的衣服。单品满 500 打八折，总价满 800 打九折；如果当天下雨，总价再打 85 折。请问我还剩多少？",
)
.await?;
if result.trim().is_empty() {
return Err(echo_agent::error::ReactError::Other(
"demo04 验收失败：SubAgent 编排示例返回空结果".to_string(),
));
}
if input_count.load(Ordering::Relaxed) == 0 {
return Err(echo_agent::error::ReactError::Other(
"demo04 验收失败：未触发 human_in_loop 输入请求".to_string(),
));
}
if !result.contains("331.39") {
return Err(echo_agent::error::ReactError::Other(
"demo04 验收失败：最终结果未包含预期剩余金额 331.39".to_string(),
));
}
println!("\n✅ 最终结果:\n{}", result);
worker.abort();
Ok(())
}