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
CI / 全量测试 (Electron ABI) (push) Failing after 5m21s
CI / 产物编译验证 (push) Successful in 10m6s
CI / 类型检查 + Lint + 单元测试 (push) Failing after 5m42s

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:
2026-08-21 22:11:11 +08:00
parent c727ad075c
commit e4d693c1c7
8 changed files with 406 additions and 45 deletions
+64 -27
View File
@@ -53,21 +53,25 @@ export class DeepSeekAdapter extends BaseAdapter {
const body = this.toNativeRequest(request, false);
// #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch,确保 timer 清理
const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${this.config.apiKey}`,
...this.config.headers,
const response = await this.fetchWithTimeout(
`${this.config.baseURL}/chat/completions`,
{
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${this.config.apiKey}`,
...this.config.headers,
},
body: JSON.stringify(body),
},
body: JSON.stringify(body),
}, this.config.timeoutMs ?? 120_000);
this.config.timeoutMs ?? 120_000,
);
if (!response.ok) {
await this.throwHttpError(response, 'DeepSeek API error');
}
const data = await response.json() as Record<string, unknown>;
const data = (await response.json()) as Record<string, unknown>;
const parsed = parseOpenAICompatibleResponse(data);
return {
@@ -93,15 +97,19 @@ export class DeepSeekAdapter extends BaseAdapter {
const body = this.toNativeRequest(request, true);
// #24 修复: 使用 fetchWithTimeout 替代 getFetchSignal + fetch,确保 timer 清理
const response = await this.fetchWithTimeout(`${this.config.baseURL}/chat/completions`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${this.config.apiKey}`,
...this.config.headers,
const response = await this.fetchWithTimeout(
`${this.config.baseURL}/chat/completions`,
{
method: 'POST',
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${this.config.apiKey}`,
...this.config.headers,
},
body: JSON.stringify(body),
},
body: JSON.stringify(body),
}, this.config.timeoutMs ?? 300_000);
this.config.timeoutMs ?? 300_000,
);
if (!response.ok || !response.body) {
await this.throwHttpError(response, 'DeepSeek stream error');
@@ -132,7 +140,7 @@ export class DeepSeekAdapter extends BaseAdapter {
signal: AbortSignal.timeout(10_000),
});
if (response.ok) {
const data = await response.json() as { data?: Array<{ id: string }> };
const data = (await response.json()) as { data?: Array<{ id: string }> };
if (data.data?.length) {
// 合并 API 返回的模型 ID 与本地元数据
return data.data.map((m) => DeepSeekAdapter.MODEL_INFO[m.id] ?? { id: m.id });
@@ -164,6 +172,18 @@ export class DeepSeekAdapter extends BaseAdapter {
// ===== GET /user/balance =====
/**
* 查询账户余额
*
* v0.5.2 修复: DeepSeek 官方 API 实际返回 `balance_infos` 数组格式:
* { "is_available": true, "balance_infos": [{ "currency": "CNY",
* "total_balance": "110.00", "granted_balance": "10.00", "topped_up_balance": "100.00" }] }
* 此前按扁平字段解析(data.total_balance)→ 永远取到 undefined → 恒显示 0。
* 现优先取 balance_infos[0],回退扁平格式(兼容网关/代理的简化响应)。
*
* URL 规范化: 余额端点为 {root}/user/balance(无 /v1 前缀)。用户配置的
* baseURL 可能带 /v1 或尾斜杠(chat 端点两种写法都合法),此处剥离后拼接。
*/
async getBalance(): Promise<{
currency: string;
totalBalance: string;
@@ -171,20 +191,34 @@ export class DeepSeekAdapter extends BaseAdapter {
toppedUpBalance: string;
} | null> {
try {
const response = await fetch(`${this.config.baseURL}/user/balance`, {
// 规范化 baseURL:去尾斜杠、去尾 /v1(余额端点在根路径下)
const root = this.config.baseURL.replace(/\/+$/, '').replace(/\/v1$/, '');
const response = await fetch(`${root}/user/balance`, {
headers: { Authorization: `Bearer ${this.config.apiKey}` },
signal: AbortSignal.timeout(10_000),
});
if (!response.ok) return null;
const data = await response.json() as {
currency?: string; total_balance?: string;
granted_balance?: string; topped_up_balance?: string;
const data = (await response.json()) as {
is_available?: boolean;
balance_infos?: Array<{
currency?: string;
total_balance?: string;
granted_balance?: string;
topped_up_balance?: string;
}>;
// 扁平格式字段(网关/代理兼容)
currency?: string;
total_balance?: string;
granted_balance?: string;
topped_up_balance?: string;
};
// 优先官方 balance_infos 数组,回退扁平格式
const info = data.balance_infos?.[0] ?? data;
return {
currency: data.currency ?? 'CNY',
totalBalance: data.total_balance ?? '0',
grantedBalance: data.granted_balance ?? '0',
toppedUpBalance: data.topped_up_balance ?? '0',
currency: info.currency ?? 'CNY',
totalBalance: info.total_balance ?? '0',
grantedBalance: info.granted_balance ?? '0',
toppedUpBalance: info.topped_up_balance ?? '0',
};
} catch {
return null;
@@ -228,7 +262,10 @@ export class DeepSeekAdapter extends BaseAdapter {
} else if (request.params.thinkingEnabled) {
body.thinking = { type: 'enabled' };
const effortMap: Record<string, string> = {
low: 'high', medium: 'high', high: 'high', max: 'max',
low: 'high',
medium: 'high',
high: 'high',
max: 'max',
};
// DeepSeek API 仅支持 high / max 两档,low/medium 映射为 high
body.reasoning_effort = effortMap[request.params.thinkingEffort ?? 'high'] ?? 'high';