Files
metona-ai-desktop/electron/harness/utils/__tests__/token-estimator.test.ts
T
thzxx 2230bcec3f feat: v0.4.0 四阶段迭代 — 安全加固 + 工程基线 + 架构重构 + 双 Provider 扩展
P0 安全修复:
- API Key 加密存储(safeStorage 密钥链,版本化前缀,历史明文平滑兼容)
- 间接提示注入防护(SecurityScanHook 工具结果深扫描,网络工具脱敏/本地工具警示分级)
- error:report IPC 断链修复(渲染进程错误上报落 electron-log + 审计)
- abort 信号贯通工具层(run_command/dev-tools 子进程随会话中断终止)
- run_command 沙箱加固(cd 系统目录/敏感文件读取拦截 + chcp 前缀剥离防解析退化)
- .env 真实生效(dotenv 回退加载,应用内配置优先)

P1 工程基础:
- ESLint 9 flat config + 全部 34 条存量 warnings 清零(零容忍基线)
- 测试基线 118 用例 11 文件(token/文件防护/权限/沙箱/注入/命令/引擎/注册表/审计链/摘要分层)
- test:electron 双模式(ELECTRON_RUN_AS_NODE 跑 Electron ABI,SQLite 套件全执行)
- SessionRecorder 多会话隔离 + 9 种 TRACE 事件补全(含最终轮 iteration_end)
- Provider 故障转移(重试耗尽/不可重试一次性切换 fallback + 前端通知)
- MCP 真就绪(等待全部连接完成再广播 tools:ready)
- SLO/HealthChecker 真实接入(60s 巡检 + 托盘状态)
- CONFIG_DEFAULTS 单一来源(消除 SEED 双源漂移)

P2 架构升级:
- handlers.ts 1940 行拆分为 13 个 IPC 域模块(防重入注册 + 多窗口广播)
- AgentEngineManager 每会话独立引擎(LRU 30 + adapter 工厂隔离 abort 信号)
- TaskOrchestrator EngineProvider 改造 + abortByParent 联动中断 SubAgent
- 会话摘要分层上下文(session_summaries 滚动摘要 + 截断游标清理防因果污染)
- 消息编辑重发/重新生成(truncateAfter IPC + store 动作 + UI)
- Markdown 导出 / WebSearch 并行抓取(并发 3)/ 记忆 TF 缓存 / 版本构建期注入

P3 能力扩展:
- OpenAI Adapter(o 系列推理模型 reasoning_effort/max_completion_tokens)
- Anthropic Adapter(原生 Messages API:tool_use 块/角色合并/thinking budget/图片 base64/SSE 事件机)
- 设置页/Onboarding 六 Provider 全链路接入
2026-08-20 23:17:02 +08:00

69 lines
2.2 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
/**
* Token Estimator 单元测试(P1-14 测试基线)
*/
import { describe, it, expect } from 'vitest';
import { estimateStringTokens, estimateMessagesTokens } from '../token-estimator';
describe('estimateStringTokens', () => {
it('空值返回 0', () => {
expect(estimateStringTokens('')).toBe(0);
expect(estimateStringTokens(null)).toBe(0);
expect(estimateStringTokens(undefined)).toBe(0);
});
it('纯 ASCII4 字符 ≈ 1 token', () => {
// 16 个 ASCII 字符 → 16 * 0.25 = 4 tokens
expect(estimateStringTokens('abcdefghijklmnop')).toBe(4);
});
it('纯中文:1 字符 ≈ 1 token', () => {
expect(estimateStringTokens('你好世界')).toBe(4);
});
it('混合文本按系数分别计算', () => {
// 4 ASCII (1 token) + 2 中文 (2 tokens) = 3 tokens
expect(estimateStringTokens('abcd你好')).toBe(3);
});
it('Emoji 计为 1 token/字符', () => {
expect(estimateStringTokens('🎉🎊')).toBe(2);
});
it('结果向上取整', () => {
// 1 个 ASCII = 0.25 → ceil 为 1
expect(estimateStringTokens('a')).toBe(1);
});
});
describe('estimateMessagesTokens', () => {
it('每条消息计入结构性开销(4 tokens)', () => {
const msgs = [{ content: '' }, { content: '' }];
expect(estimateMessagesTokens(msgs)).toBe(8); // 2 * 4 overhead
});
it('content 为 null 时只计开销(tool_calls 消息场景)', () => {
expect(estimateMessagesTokens([{ content: null }])).toBe(4);
});
it('toolCalls 计入 id/name/args 开销', () => {
const withToolCall = [
{
content: null,
toolCalls: [{ id: 'tc_12345678', name: 'read_file', args: { file_path: '/a/b.ts' } }],
},
];
const withoutToolCall = [{ content: null }];
const diff = estimateMessagesTokens(withToolCall) - estimateMessagesTokens(withoutToolCall);
// id(9 chars→3) + name(9→3) + args(~18→5) + overhead(8) ≈ 19 tokens
expect(diff).toBeGreaterThanOrEqual(15);
expect(diff).toBeLessThanOrEqual(30);
});
it('reasoningContent 计入 token', () => {
const withReasoning = [{ content: '', reasoningContent: 'abcd' }];
const without = [{ content: '' }];
expect(estimateMessagesTokens(withReasoning) - estimateMessagesTokens(without)).toBe(1);
});
});