echo_agent 0.1.1

AI Agent framework with ReAct loop, multi-provider LLM, tool execution, and A2A HTTP server
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104

use std::sync::Arc;

use futures::future::BoxFuture;
use serde_json::Value;

use crate::error::ToolError;
use crate::human_loop::{HumanLoopProvider, HumanLoopRequest, HumanLoopResponse};
use crate::tools::{Tool, ToolParameters, ToolResult};

/// LLM 触发的人工介入工具。
///
/// 当 LLM 不确定用户意图、需要额外信息或需要用户确认时调用。
/// 通过注入的 [`HumanLoopProvider`] 以异步方式向用户请求输入，
/// 支持命令行、HTTP Webhook、WebSocket 等多种渠道。
pub struct HumanInLoop {
    provider: Arc<dyn HumanLoopProvider>,
}

impl HumanInLoop {
    pub fn new(provider: Arc<dyn HumanLoopProvider>) -> Self {
        Self { provider }
    }
}

impl Tool for HumanInLoop {
    fn name(&self) -> &str {
        "human_in_loop"
    }

    fn description(&self) -> &str {
        "当你不确定用户意图、需要额外信息、或需要用户确认时使用此工具。"
    }

    fn parameters(&self) -> Value {
        serde_json::json!({
            "type": "object",
            "properties": {
                "reasoning": {
                    "type": "string",
                    "description": "触发原因：为什么需要人工介入？意图不明确时由 LLM 给出；工具存在风险时由用户确认。"
                },
                "tool": {
                    "type": "string",
                    "description": "引起触发的工具名称（可选）"
                },
                "approval_type": {
                    "type": "string",
                    "description": "触发类型：LLM 主动触发时填 'LLM'，工具触发时填 'tool'"
                }
            },
            "required": ["reasoning", "approval_type"]
        })
    }

    fn execute(
        &self,
        parameters: ToolParameters,
    ) -> BoxFuture<'_, crate::error::Result<ToolResult>> {
        Box::pin(async move {
            let approval_type = parameters
                .get("approval_type")
                .and_then(|t| t.as_str())
                .ok_or_else(|| ToolError::MissingParameter("approval_type".to_string()))?;

            let reasoning = parameters
                .get("reasoning")
                .and_then(|t| t.as_str())
                .ok_or_else(|| ToolError::MissingParameter("reasoning".to_string()))?;

            let tool = parameters
                .get("tool")
                .and_then(|t| t.as_str())
                .unwrap_or("无");

            let prompt = format!(
                "需要你给予帮助。\n触发类型：{approval_type}\n触发原因：{reasoning}\n触发工具：{tool}\n\n请直接回复你的意见或确认："
            );

            let req = HumanLoopRequest::input(prompt);
            let result_text = match self.provider.request(req).await? {
                HumanLoopResponse::Text(text) => text,
                HumanLoopResponse::Approved => "用户已确认".to_string(),
                HumanLoopResponse::ApprovedWithScope { scope } => {
                    format!("用户已确认（scope: {:?}）", scope)
                }
                HumanLoopResponse::ModifiedArgs { args, scope } => {
                    format!("用户修改参数后确认（args: {}, scope: {:?}）", args, scope)
                }
                HumanLoopResponse::Rejected { reason } => {
                    format!(
                        "用户已拒绝{}",
                        reason.map(|r| format!("，原因：{r}")).unwrap_or_default()
                    )
                }
                HumanLoopResponse::Timeout => "等待用户输入超时".to_string(),
                HumanLoopResponse::Deferred => "用户推迟决策".to_string(),
            };

            Ok(ToolResult::success(format!(
                "用户回复（触发原因：{reasoning}，工具：{tool}）：{result_text}"
            )))
        })
    }
}