fix: v0.5.2 紧急修复 — 工具调用完全失效(引擎懒创建不带工具)+ DeepSeek 余额解析错误
P0 工具调用失效(用户实测反馈:Agnes/DeepSeek 均无法调用工具): - 根因:AgentEngineManager.createEngine 未调用 setTools。引擎是懒创建的 (首次 sendMessage 时 getEngine),启动期的 setToolsAll 调用时 engines Map 为空(全是 no-op)→ 新引擎 this.tools=[] → LLM 请求不带 tools → 模型无法发起 tool_call。症状与用户反馈完全吻合:模型口头说要调工具 (模仿历史消息中的工具调用模式),实际不调,凭记忆瞎编结果。 - 引入点:v0.4.0 P2-10 每会话引擎重构(v0.3.x 全局单引擎时代 setTools 直接作用于唯一引擎,无此问题)。 - 修复:createEngine 从 toolRegistry 拉取当前启用工具(registry 是启用 状态的唯一事实源,MCP 后注册/工具开关场景均一致)。 P1 DeepSeek 余额显示错误(用户实测反馈:显示的不是真实余额): - 根因:DeepSeek 官方 /user/balance 实际返回 balance_infos 数组格式, 此前按扁平字段解析(data.total_balance)→ 恒为 undefined → 界面恒显示 0。 - 修复:优先解析 balance_infos[0],回退扁平格式(网关兼容);URL 规范化 (剥离尾斜杠与 /v1 前缀 — 余额端点在根路径,chat 端点两种写法都合法)。 测试(215 → 224 用例): - 新增 AgentEngineManager 回归测试 ×4:懒创建引擎的 LLM 请求必须携带 registry 工具定义(本次事故的直接拦截测试)/ 禁用工具不出现 / setToolsAll 热更新 / 无 registry 时行为不回归 - 新增 DeepSeek 余额解析测试 ×5:官方数组格式 / 扁平回退 / URL 规范化 / 非 2xx / 网络异常 - 补强引擎链路测试:mock adapter 记录请求并断言 tools 契约 — 此前 mock 无条件吐 tool_call 事件,掩盖了"请求未携带工具定义"的缺陷(复检盲区 的直接教训:mock 必须断言请求契约,否则测试是道具) 验证: lint 0 / typecheck 双工程 0 / test:electron 224 全过 / build 成功
This commit is contained in:
@@ -0,0 +1,180 @@
|
||||
/**
|
||||
* AgentEngineManager 回归测试(v0.5.2)
|
||||
*
|
||||
* 背景:v0.4.0 P2-10 引入每会话懒创建引擎后,createEngine 未调用 setTools,
|
||||
* 且启动期 setToolsAll 调用时 engines Map 为空(全是 no-op)→ 新引擎 tools=[]
|
||||
* → LLM 请求不带工具定义 → 模型无法发起 tool_call,只能凭历史记忆"口头说要
|
||||
* 调工具"并瞎编结果。v0.5.2 在 createEngine 中从 registry 拉取工具修复。
|
||||
*
|
||||
* 本测试直接断言"传给 adapter.sendStream 的请求携带工具"——这是唯一能
|
||||
* 拦住该类回归的测试层级(引擎单测的 mock adapter 无条件吐 tool_call
|
||||
* 事件,会掩盖请求契约缺陷)。
|
||||
*/
|
||||
|
||||
import { describe, it, expect, vi } from 'vitest';
|
||||
|
||||
vi.mock('electron-log', () => ({
|
||||
default: { info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() },
|
||||
}));
|
||||
|
||||
import { AgentEngineManager } from '../agent-engine-manager.service';
|
||||
import type {
|
||||
IMetonaProviderAdapter,
|
||||
MetonaRequest,
|
||||
MetonaResponse,
|
||||
MetonaStreamEvent,
|
||||
} from '../../harness/types';
|
||||
import { MetonaStreamEventType } from '../../harness/types';
|
||||
import { ToolRegistry } from '../../harness/tools/registry';
|
||||
import type { IMetonaTool } from '../../harness/types/metona-tool';
|
||||
import { MetonaToolCategory, MetonaRiskLevel } from '../../harness/types';
|
||||
|
||||
/** 记录请求的 mock adapter — runStream 完成后检查 requests 数组 */
|
||||
function createRecordingAdapter(requests: MetonaRequest[]): IMetonaProviderAdapter {
|
||||
return {
|
||||
providerId: 'mock',
|
||||
supportedModels: ['mock-model'],
|
||||
supportsToolCalling: true,
|
||||
supportsThinking: false,
|
||||
getContextWindow: () => 1_000_000,
|
||||
send: vi.fn(
|
||||
async (): Promise<MetonaResponse> => ({
|
||||
meta: { requestId: 'r', provider: 'mock', model: 'm', latencyMs: 1, timestamp: Date.now() },
|
||||
content: 'ok',
|
||||
usage: { inputTokens: 1, outputTokens: 1, totalTokens: 2 },
|
||||
finishReason: 'stop' as never,
|
||||
}),
|
||||
),
|
||||
sendStream: vi.fn(async function* (req: MetonaRequest): AsyncIterable<MetonaStreamEvent> {
|
||||
requests.push(req);
|
||||
yield {
|
||||
type: MetonaStreamEventType.TEXT_DELTA,
|
||||
requestId: 'r',
|
||||
sessionId: req.meta.sessionId,
|
||||
iteration: req.meta.iteration,
|
||||
seq: 0,
|
||||
timestamp: Date.now(),
|
||||
delta: 'done',
|
||||
};
|
||||
yield {
|
||||
type: MetonaStreamEventType.DONE,
|
||||
requestId: 'r',
|
||||
sessionId: req.meta.sessionId,
|
||||
iteration: req.meta.iteration,
|
||||
seq: 1,
|
||||
timestamp: Date.now(),
|
||||
};
|
||||
}),
|
||||
setAbortSignal: vi.fn(),
|
||||
healthCheck: async () => true,
|
||||
};
|
||||
}
|
||||
|
||||
function makeTool(name: string): IMetonaTool {
|
||||
return {
|
||||
definition: {
|
||||
name,
|
||||
description: `${name} (test fixture)`,
|
||||
parameters: { type: 'object', properties: {}, required: [] },
|
||||
category: MetonaToolCategory.FILESYSTEM,
|
||||
riskLevel: MetonaRiskLevel.SAFE,
|
||||
requiresPermission: false,
|
||||
timeoutMs: 5_000,
|
||||
},
|
||||
async execute() {
|
||||
return { ok: true };
|
||||
},
|
||||
};
|
||||
}
|
||||
|
||||
const userMessage = { role: 'user' as const, content: 'hello', timestamp: Date.now() };
|
||||
const systemPrompt = { roleDefinition: '', outputConstraints: '', safetyGuidelines: '' };
|
||||
|
||||
describe('AgentEngineManager — 懒创建引擎的工具传递(v0.5.2 回归修复)', () => {
|
||||
it('新会话首次 sendMessage 时,LLM 请求必须携带 registry 中的工具定义', async () => {
|
||||
const requests: MetonaRequest[] = [];
|
||||
const adapter = createRecordingAdapter(requests);
|
||||
const registry = new ToolRegistry();
|
||||
registry.registerBuiltin(makeTool('read_file'));
|
||||
registry.registerBuiltin(makeTool('run_command'));
|
||||
|
||||
// 模拟真实启动流程:manager 创建时 registry 已有工具,
|
||||
// 但引擎在首次 sendMessage 才懒创建(setToolsAll 此时是 no-op)
|
||||
const manager = new AgentEngineManager({
|
||||
buildAdapter: () => adapter,
|
||||
baseConfig: {},
|
||||
toolRegistry: registry,
|
||||
});
|
||||
|
||||
const engine = manager.getEngine('sess-lazy-1');
|
||||
await engine.runStream(userMessage, 'sess-lazy-1', [], systemPrompt);
|
||||
|
||||
// 核心断言:请求携带工具定义(回归点 — 修复前为 undefined)
|
||||
expect(requests).toHaveLength(1);
|
||||
expect(requests[0].tools).toBeDefined();
|
||||
expect(requests[0].tools!.map((t) => t.name)).toContain('read_file');
|
||||
expect(requests[0].tools!.map((t) => t.name)).toContain('run_command');
|
||||
});
|
||||
|
||||
it('已禁用的工具不出现在懒创建引擎的请求中', async () => {
|
||||
const requests: MetonaRequest[] = [];
|
||||
const adapter = createRecordingAdapter(requests);
|
||||
const registry = new ToolRegistry();
|
||||
registry.registerBuiltin(makeTool('read_file'));
|
||||
registry.registerBuiltin(makeTool('web_browser'));
|
||||
registry.setToolEnabled('web_browser', false);
|
||||
|
||||
const manager = new AgentEngineManager({
|
||||
buildAdapter: () => adapter,
|
||||
baseConfig: {},
|
||||
toolRegistry: registry,
|
||||
});
|
||||
|
||||
const engine = manager.getEngine('sess-lazy-2');
|
||||
await engine.runStream(userMessage, 'sess-lazy-2', [], systemPrompt);
|
||||
|
||||
expect(requests[0].tools).toBeDefined();
|
||||
expect(requests[0].tools!.map((t) => t.name)).toContain('read_file');
|
||||
expect(requests[0].tools!.map((t) => t.name)).not.toContain('web_browser');
|
||||
});
|
||||
|
||||
it('setToolsAll 仍可热更新已存在的引擎(工具开关变更场景)', async () => {
|
||||
const requests: MetonaRequest[] = [];
|
||||
const adapter = createRecordingAdapter(requests);
|
||||
const registry = new ToolRegistry();
|
||||
registry.registerBuiltin(makeTool('read_file'));
|
||||
|
||||
const manager = new AgentEngineManager({
|
||||
buildAdapter: () => adapter,
|
||||
baseConfig: {},
|
||||
toolRegistry: registry,
|
||||
});
|
||||
|
||||
const engine = manager.getEngine('sess-hot');
|
||||
await engine.runStream(userMessage, 'sess-hot', [], systemPrompt);
|
||||
expect(requests[0].tools!.map((t) => t.name)).toContain('read_file');
|
||||
|
||||
// 热更新:新注册工具 + setToolsAll 同步
|
||||
registry.registerBuiltin(makeTool('new_tool'));
|
||||
manager.setToolsAll(registry.listTools());
|
||||
|
||||
requests.length = 0;
|
||||
await engine.runStream(userMessage, 'sess-hot', [], systemPrompt);
|
||||
expect(requests[0].tools!.map((t) => t.name)).toContain('new_tool');
|
||||
});
|
||||
|
||||
it('无 toolRegistry 时请求不带 tools(纯对话配置,行为不回归)', async () => {
|
||||
const requests: MetonaRequest[] = [];
|
||||
const adapter = createRecordingAdapter(requests);
|
||||
|
||||
const manager = new AgentEngineManager({
|
||||
buildAdapter: () => adapter,
|
||||
baseConfig: {},
|
||||
});
|
||||
|
||||
const engine = manager.getEngine('sess-no-tools');
|
||||
await engine.runStream(userMessage, 'sess-no-tools', [], systemPrompt);
|
||||
|
||||
expect(requests[0].tools).toBeUndefined();
|
||||
});
|
||||
});
|
||||
Reference in New Issue
Block a user