宜博 / lemonboss

import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';

/**
 * Master-Innovator 集成测试
 *
 * 验证 Master Session 的完整工具调用链路：
 *   1. Master 收到用户消息 → 解析为 orchestrate 工具调用
 *   2. orchestrate 返回编排计划
 *   3. Master 按计划调用 call_sub_agent
 *   4. 最终汇总结果回复用户
 *
 * 通过 mock AgentManager.callAgent 模拟 Master 的 Claude CLI 输出。
 */

const { MasterSession } = require('../../core/master-session');
const { ToolRegistry } = require('../../tools/tool-registry');
const { createOrchestrateTool } = require('../../tools/orchestrate-tool');
const { createCallAgentTool, createListAgentsTool } = require('../../tools/call-agent-tool');

// Mock logging
vi.mock('../../core/logging', () => ({
    getGlobalLogManager: () => ({
        initialized: false,
        getOrCreateSession: () => null,
    }),
    waitForGlobalLogManager: () => Promise.resolve(),
}));

// ─── 测试配置 ───

const testConfig = {
    agents: {
        master: { type: 'master-cli', client: 'claude-local', description: 'Master Agent' },
        coder: { type: 'executor', client: 'claude-local', description: '编写代码' },
        writer: { type: 'executor', client: 'claude-local', description: '创意写作' },
        verifier: { type: 'validator', client: 'claude-local', description: '代码验证' },
        'frontend-tester': { type: 'validator', client: 'claude-local', description: '前端测试' },
    },
    agentGroups: {
        development: {
            name: '软件开发',
            description: '编写和验证代码',
            default: true,
            agents: ['coder', 'verifier', 'frontend-tester'],
        },
        writing: {
            name: '创意写作',
            description: '写作和编辑',
            agents: ['writer'],
        },
    },
    clients: {
        'claude-local': { command: 'claude', args: [] },
    },
};

// ─── Helper ───

function createMockSession(callResponses) {
    let callIndex = 0;

    const session = new MasterSession(testConfig, '/tmp/test-master');

    // 直接 mock init 跳过真实 AgentManager 创建
    session._initialized = true;

    // 手动注册工具
    session.toolRegistry = new ToolRegistry();
    session.toolRegistry.registerAll([
        createOrchestrateTool(),
        createCallAgentTool(),
        createListAgentsTool(),
    ]);

    // Mock AgentManager
    const mockCallAgent = vi.fn().mockImplementation(async (agentName, prompt) => {
        if (agentName === 'master') {
            // Master 的模拟回复 — 按序返回预设响应
            const response = callResponses[callIndex++] || { output: '任务完成！' };
            return response;
        }
        // Sub Agent 的模拟回复
        return { output: `[${agentName}] 执行完成: ${prompt.substring(0, 50)}...` };
    });

    session.manager = {
        callAgent: mockCallAgent,
        closeAllSessions: vi.fn(),
        getAgentConfig: vi.fn().mockImplementation((name) => {
            const cfg = testConfig.agents[name];
            if (!cfg) throw new Error(`Agent "${name}" not found`);
            return cfg;
        }),
        listAgents: vi.fn().mockReturnValue(
            Object.entries(testConfig.agents).map(([name, cfg]) => ({ name, ...cfg }))
        ),
    };

    return { session, mockCallAgent };
}

// ─── 测试 ───

describe('Master Session — 工具调用集成测试', () => {
    afterEach(() => {
        vi.restoreAllMocks();
    });

    // ── 1. 直接回复（无工具调用） ──

    describe('直接回复', () => {
        it('should reply directly for simple questions without tool calls', async () => {
            const { session } = createMockSession([
                { output: 'Docker 是一个容器化平台，用于打包和运行应用。' },
            ]);

            const result = await session.handleMessage('Docker 是什么？');

            expect(result.type).toBe('reply');
            expect(result.output).toContain('Docker');
            expect(result.toolCalls).toHaveLength(0);
            expect(result.rounds).toBe(1);
        });
    });

    // ── 2. 单 Agent 调用（call_sub_agent） ──

    describe('单 Agent 调用', () => {
        it('should call sub agent directly and return result', async () => {
            const { session, mockCallAgent } = createMockSession([
                // Round 1: Master 调用 call_sub_agent
                {
                    output: `好的，我来帮你写一首诗。

[TOOL_CALL: call_sub_agent]
{
  "agent": "writer",
  "task": "写一首关于秋天的五言绝句"
}
[/TOOL_CALL]`,
                },
                // Round 2: Master 收到结果后汇总回复
                {
                    output: '这是 writer 创作的秋天诗：\n\n秋风起天末，万叶落纷纷。\n孤雁向南去，寒山映夕云。',
                },
            ]);

            const result = await session.handleMessage('帮我写一首关于秋天的诗');

            expect(result.type).toBe('reply');
            expect(result.rounds).toBe(2);
            expect(result.toolCalls).toHaveLength(1);
            expect(result.toolCalls[0].tool).toBe('call_sub_agent');
            expect(result.toolCalls[0].params.agent).toBe('writer');

            // 验证 Master 被调了 2 次（初始 + 工具结果反馈）
            expect(mockCallAgent).toHaveBeenCalledTimes(3); // 2 master + 1 writer

            // 验证 writer 确实被调用了
            const writerCall = mockCallAgent.mock.calls.find(c => c[0] === 'writer');
            expect(writerCall).toBeTruthy();
            expect(writerCall[1]).toContain('秋天');
        });
    });

    // ── 3. orchestrate + 多步骤 call_sub_agent ──

    describe('多 Agent 编排', () => {
        it('should orchestrate plan then call sub agents step by step', async () => {
            const { session, mockCallAgent } = createMockSession([
                // Round 1: Master 调用 orchestrate 获取编排计划
                {
                    output: `这是一个需要多 Agent 协作的任务，让我先获取编排计划。

[TOOL_CALL: orchestrate]
{
  "task": "开发一个 Python 命令行计算器",
  "agentGroup": "development"
}
[/TOOL_CALL]`,
                },
                // Round 2: Master 收到 plan，调用 coder
                {
                    output: `收到编排计划。按步骤执行，先调用 coder。

[TOOL_CALL: call_sub_agent]
{
  "agent": "coder",
  "task": "开发一个 Python 命令行计算器，支持加减乘除，处理除零和非法输入"
}
[/TOOL_CALL]`,
                },
                // Round 3: coder 完成，调用 verifier
                {
                    output: `coder 已完成代码编写。现在调用 verifier 验证。

[TOOL_CALL: call_sub_agent]
{
  "agent": "verifier",
  "task": "验证 Python 计算器：运行测试，检查除零处理和错误输入处理"
}
[/TOOL_CALL]`,
                },
                // Round 4: 全部通过，汇总回复
                {
                    output: `✅ 任务完成！

**计算器已开发并验证通过：**
- coder 创建了 calc.py，支持加减乘除
- verifier 验证通过，除零和非法输入都有正确处理

所有步骤已完成。`,
                },
            ]);

            const toolCalls = [];
            const result = await session.handleMessage('开发一个 Python 命令行计算器', {
                onToolCall: (tc) => toolCalls.push(tc),
            });

            // 验证最终结果
            expect(result.type).toBe('reply');
            expect(result.rounds).toBe(4);
            expect(result.output).toContain('任务完成');

            // 验证工具调用历史
            expect(result.toolCalls).toHaveLength(3); // orchestrate + coder + verifier
            expect(result.toolCalls[0].tool).toBe('orchestrate');
            expect(result.toolCalls[1].tool).toBe('call_sub_agent');
            expect(result.toolCalls[1].params.agent).toBe('coder');
            expect(result.toolCalls[2].tool).toBe('call_sub_agent');
            expect(result.toolCalls[2].params.agent).toBe('verifier');

            // 验证 orchestrate 返回了有效 plan
            const orchResult = result.toolCalls[0].result;
            expect(orchResult.status).toBe('success');
            expect(orchResult.result.status).toBe('success');
            expect(orchResult.result.plan.agentGroup).toBe('development');
            expect(orchResult.result.plan.steps.length).toBeGreaterThan(0);

            // 验证 onToolCall 回调被触发
            expect(toolCalls).toHaveLength(3);

            // 验证 Sub Agent 被实际调用
            const agentCalls = mockCallAgent.mock.calls.filter(c => c[0] !== 'master');
            expect(agentCalls.map(c => c[0])).toEqual(['coder', 'verifier']);
        });
    });

    // ── 4. 修复循环 ──

    describe('修复循环', () => {
        it('should handle validator failure → executor fix → re-validate', async () => {
            const { session, mockCallAgent } = createMockSession([
                // Round 1: 直接调 coder
                {
                    output: `[TOOL_CALL: call_sub_agent]
{
  "agent": "coder",
  "task": "写一个 sort 函数"
}
[/TOOL_CALL]`,
                },
                // Round 2: 调 verifier
                {
                    output: `[TOOL_CALL: call_sub_agent]
{
  "agent": "verifier",
  "task": "验证 sort 函数"
}
[/TOOL_CALL]`,
                },
                // Round 3: verifier "失败"（通过模拟 Sub Agent 返回失败结果），Master 让 coder 修复
                {
                    output: `verifier 报告了问题，让 coder 修复。

[TOOL_CALL: call_sub_agent]
{
  "agent": "coder",
  "task": "修复 sort 函数：空数组未处理"
}
[/TOOL_CALL]`,
                },
                // Round 4: 再次验证
                {
                    output: `[TOOL_CALL: call_sub_agent]
{
  "agent": "verifier",
  "task": "重新验证 sort 函数"
}
[/TOOL_CALL]`,
                },
                // Round 5: 完成
                {
                    output: '✅ sort 函数已修复并验证通过。',
                },
            ]);

            const result = await session.handleMessage('写一个排序函数');

            expect(result.type).toBe('reply');
            expect(result.rounds).toBe(5);

            // 验证工具调用：coder → verifier → coder(修复) → verifier(重验)
            const agents = result.toolCalls.map(tc => tc.params.agent);
            expect(agents).toEqual(['coder', 'verifier', 'coder', 'verifier']);
        });
    });

    // ── 5. list_agents 工具 ──

    describe('list_agents 工具', () => {
        it('should return available agents when Master calls list_agents', async () => {
            const { session } = createMockSession([
                {
                    output: `[TOOL_CALL: list_agents]
{}
[/TOOL_CALL]`,
                },
                {
                    output: '目前可用的 Agent 有：coder（编写代码）、writer（创意写作）、verifier（代码验证）。',
                },
            ]);

            const result = await session.handleMessage('有哪些 Agent 可用？');

            expect(result.type).toBe('reply');
            expect(result.toolCalls).toHaveLength(1);
            expect(result.toolCalls[0].tool).toBe('list_agents');

            const listResult = result.toolCalls[0].result.result;
            expect(listResult.agents.length).toBeGreaterThan(0);
            expect(listResult.agents.find(a => a.name === 'coder')).toBeTruthy();
            expect(listResult.groups.length).toBeGreaterThan(0);
        });
    });

    // ── 6. 未知工具处理 ──

    describe('错误处理', () => {
        it('should handle unknown tool gracefully', async () => {
            const { session } = createMockSession([
                {
                    output: `[TOOL_CALL: nonexistent_tool]
{"task": "test"}
[/TOOL_CALL]`,
                },
                {
                    output: '工具不存在，我直接回答你的问题。',
                },
            ]);

            const result = await session.handleMessage('测试');

            expect(result.type).toBe('reply');
            expect(result.toolCalls).toHaveLength(1);
            expect(result.toolCalls[0].result.status).toBe('error');
            expect(result.toolCalls[0].result.error).toContain('Unknown tool');
        });

        it('should handle nonexistent agent in call_sub_agent', async () => {
            const { session } = createMockSession([
                {
                    output: `[TOOL_CALL: call_sub_agent]
{
  "agent": "nonexistent_agent",
  "task": "test"
}
[/TOOL_CALL]`,
                },
                {
                    output: '该 Agent 不存在。',
                },
            ]);

            const result = await session.handleMessage('调用不存在的 agent');

            expect(result.type).toBe('reply');
            expect(result.toolCalls).toHaveLength(1);
            expect(result.toolCalls[0].result.status).toBe('error');
            expect(result.toolCalls[0].result.error).toContain('不存在');
        });

        it('should stop at max tool rounds', async () => {
            // 无限循环调用工具
            const infiniteResponses = Array(15).fill({
                output: `[TOOL_CALL: list_agents]
{}
[/TOOL_CALL]`,
            });

            const { session } = createMockSession(infiniteResponses);

            const result = await session.handleMessage('无限循环', {
                maxToolRounds: 3,
            });

            expect(result.type).toBe('max_rounds');
            expect(result.rounds).toBe(3);
        });
    });

    // ── 7. ToolRegistry 单元测试 ──

    describe('ToolRegistry', () => {
        it('should parse multiple tool calls from output', () => {
            const output = `
先做第一步：
[TOOL_CALL: orchestrate]
{"task": "test task"}
[/TOOL_CALL]

还有第二步：
[TOOL_CALL: call_sub_agent]
{"agent": "coder", "task": "write code"}
[/TOOL_CALL]
`;
            const calls = ToolRegistry.parseToolCalls(output);
            expect(calls).toHaveLength(2);
            expect(calls[0].tool).toBe('orchestrate');
            expect(calls[0].params.task).toBe('test task');
            expect(calls[1].tool).toBe('call_sub_agent');
            expect(calls[1].params.agent).toBe('coder');
        });

        it('should handle non-JSON params as raw task', () => {
            const output = `[TOOL_CALL: call_sub_agent]
just a plain text task
[/TOOL_CALL]`;
            const calls = ToolRegistry.parseToolCalls(output);
            expect(calls).toHaveLength(1);
            expect(calls[0].params.task).toBe('just a plain text task');
        });

        it('should return empty array for no tool calls', () => {
            const calls = ToolRegistry.parseToolCalls('Just a normal reply with no tools.');
            expect(calls).toHaveLength(0);
        });

        it('should generate tool prompt with all registered tools', () => {
            const registry = new ToolRegistry();
            registry.registerAll([
                createOrchestrateTool(),
                createCallAgentTool(),
                createListAgentsTool(),
            ]);

            const prompt = registry.generateToolPrompt();
            expect(prompt).toContain('orchestrate');
            expect(prompt).toContain('call_sub_agent');
            expect(prompt).toContain('list_agents');
            expect(prompt).toContain('[TOOL_CALL:');
        });
    });

    // ── 8. orchestrate planning 验证 ──

    describe('orchestrate planning', () => {
        it('should generate valid plan for development group', async () => {
            const tool = createOrchestrateTool();
            const result = await tool.execute(
                { task: '开发计算器', agentGroup: 'development' },
                { config: testConfig }
            );

            expect(result.status).toBe('success');
            expect(result.plan.agentGroup).toBe('development');
            expect(result.plan.steps.length).toBeGreaterThan(0);

            // executor 步骤
            const executorSteps = result.plan.steps.filter(s => s.type === 'executor');
            expect(executorSteps.length).toBe(1); // coder
            expect(executorSteps[0].agent).toBe('coder');

            // validator 步骤
            const validatorSteps = result.plan.steps.filter(s => s.type === 'validator');
            expect(validatorSteps.length).toBe(2); // verifier + frontend-tester

            // conditional 修复步骤
            const conditionalSteps = result.plan.steps.filter(s => s.action === 'conditional');
            expect(conditionalSteps.length).toBe(1);
        });

        it('should return error for nonexistent group', async () => {
            const tool = createOrchestrateTool();
            const result = await tool.execute(
                { task: 'test', agentGroup: 'nonexistent' },
                { config: testConfig }
            );

            expect(result.status).toBe('error');
            expect(result.message).toContain('不存在');
        });

        it('should use default group when not specified', async () => {
            const tool = createOrchestrateTool();
            const result = await tool.execute(
                { task: '开发功能' },
                { config: testConfig }
            );

            expect(result.status).toBe('success');
            expect(result.plan.agentGroup).toBe('development'); // default: true
        });
    });

    // ── 9. 事件发射验证 ──

    describe('事件发射', () => {
        it('should emit round and tool_call events', async () => {
            const { session } = createMockSession([
                {
                    output: `[TOOL_CALL: list_agents]
{}
[/TOOL_CALL]`,
                },
                { output: '完成' },
            ]);

            const events = { round: [], tool_call: [], tool_result: [] };
            session.on('round', (e) => events.round.push(e));
            session.on('tool_call', (e) => events.tool_call.push(e));
            session.on('tool_result', (e) => events.tool_result.push(e));

            await session.handleMessage('test');

            expect(events.round.length).toBe(2); // 2 rounds
            expect(events.tool_call.length).toBe(1);
            expect(events.tool_call[0].tool).toBe('list_agents');
            expect(events.tool_result.length).toBe(1);
        });
    });
});