Files
portal/tools/router-classifier.test.mjs
T
Дмитрий b3af39bdbf feat(router): classifier Layer 2 — Sonnet escalation + cache (stage 3 task 3)
buildLLMPrompt сериализует активные узлы + chains в prompt.
classify() — гибрид regex + LLM с кэшем per-prompt-hash.
callAnthropicAPI через built-in fetch (без SDK).
shouldEscalate: confidence<0.7 AND not micro.
Fallback на regex-result при ошибке LLM.

NB: real-API verification отложена — нет ANTHROPIC_API_KEY на dev-машине;
Phase A 'вариант 2': mock-тесты only. Когда ключ появится, код заработает
без изменений.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-24 10:18:22 +03:00

181 lines
7.0 KiB
JavaScript

import { describe, it, expect, beforeEach } from 'vitest';
import { classifyByRegex } from './router-classifier.mjs';
const fakeRegistry = {
nodes: [
{ id: '#19', name: 'Superpowers', status: 'active', triggers: [
{ classification: 'feature', weight: 1.0 },
{ classification: 'planning', weight: 1.0 },
] },
{ id: '#62', name: 'billing-audit', status: 'active', triggers: [
{ keyword: 'списание', weight: 1.0 },
{ keyword: 'биллинг', weight: 1.0 },
{ classification: 'bugfix', weight: 0.5 },
] },
{ id: '#74', name: 'marketing', status: 'active', triggers: [
{ keyword: 'email-рассылка', weight: 1.0 },
{ keyword: 'кампания', weight: 1.0 },
{ classification: 'marketing', weight: 1.0 },
] },
{ id: '#11', name: 'pint', status: 'active', triggers: [
{ classification: 'refactor', weight: 1.0 },
{ classification: 'cleanup', weight: 1.0 },
] },
],
};
describe('classifyByRegex — task type', () => {
it('detects feature from RU keyword «фича»', () => {
const r = classifyByRegex('давай сделаем новую фичу для биллинга', fakeRegistry);
expect(r.taskType).toBe('feature');
});
it('detects planning from RU «план»', () => {
const r = classifyByRegex('напиши план рефакторинга модуля X', fakeRegistry);
expect(r.taskType).toBe('planning');
});
it('detects bugfix from EN «bug»', () => {
const r = classifyByRegex('there is a bug in the auth flow', fakeRegistry);
expect(r.taskType).toBe('bugfix');
});
it('detects micro for typo', () => {
const r = classifyByRegex('опечатка в файле X', fakeRegistry);
expect(r.micro).toBe(true);
});
it('detects micro for rename', () => {
const r = classifyByRegex('переименуй функцию foo в bar', fakeRegistry);
expect(r.micro).toBe(true);
});
it('returns taskType=unknown when no signal', () => {
const r = classifyByRegex('просто привет', fakeRegistry);
expect(r.taskType).toBe('unknown');
expect(r.micro).toBe(false);
});
});
describe('classifyByRegex — domain node match', () => {
it('picks #62 billing-audit on «списание»', () => {
const r = classifyByRegex('почини двойное списание лида', fakeRegistry);
expect(r.recommendedNode).toBe('#62');
});
it('picks #74 marketing on «email-рассылка»', () => {
const r = classifyByRegex('составь email-рассылку для тарифа Бизнес', fakeRegistry);
expect(r.recommendedNode).toBe('#74');
});
it('falls back to classification trigger when no keyword match', () => {
const r = classifyByRegex('рефакторинг кода', fakeRegistry);
// 'рефакторинг' → classification: refactor → #11 pint
expect(r.recommendedNode).toBe('#11');
});
it('returns null when no node matched', () => {
const r = classifyByRegex('просто вопрос', fakeRegistry);
expect(r.recommendedNode).toBeNull();
});
it('case-insensitive keyword match', () => {
const r = classifyByRegex('СПИСАНИЕ дублируется', fakeRegistry);
expect(r.recommendedNode).toBe('#62');
});
});
describe('classifyByRegex — source tag', () => {
it('always marks source: regex', () => {
const r = classifyByRegex('test', fakeRegistry);
expect(r.source).toBe('regex');
});
});
describe('classifyByRegex — confidence', () => {
it('returns confidence>=0.8 for clean keyword match', () => {
const r = classifyByRegex('списание дублируется', fakeRegistry);
expect(r.confidence).toBeGreaterThanOrEqual(0.8);
});
it('returns confidence<0.5 when ambiguous (no clean match)', () => {
const r = classifyByRegex('что-то непонятное', fakeRegistry);
expect(r.confidence).toBeLessThan(0.5);
});
});
import { buildLLMPrompt, parseLLMResponse, shouldEscalate, classify } from './router-classifier.mjs';
describe('buildLLMPrompt', () => {
it('serializes active nodes with id+name+top-3 triggers', () => {
const prompt = buildLLMPrompt('почини списание', fakeRegistry);
expect(prompt).toMatch(/#62/);
expect(prompt).toMatch(/billing-audit/);
expect(prompt).toMatch(/списание/);
expect(prompt).toMatch(/почини списание/);
});
it('excludes inactive nodes', () => {
const reg = { nodes: [...fakeRegistry.nodes, { id: '#999', name: 'gone', status: 'historic', triggers: [] }] };
const prompt = buildLLMPrompt('test', reg);
expect(prompt).not.toMatch(/#999/);
});
});
describe('parseLLMResponse', () => {
it('parses JSON object', () => {
const r = parseLLMResponse('{"taskType":"bugfix","micro":false,"recommendedNode":"#62","confidence":0.9,"recommendedChain":null,"reasoning":"keyword списание"}');
expect(r.taskType).toBe('bugfix');
expect(r.recommendedNode).toBe('#62');
expect(r.confidence).toBe(0.9);
});
it('parses JSON wrapped in ```json``` block', () => {
const r = parseLLMResponse('```json\n{"taskType":"feature","micro":false,"recommendedNode":"#19","confidence":0.8}\n```');
expect(r.taskType).toBe('feature');
});
it('returns null on unparseable response', () => {
expect(parseLLMResponse('I cannot help with this')).toBeNull();
});
});
describe('shouldEscalate', () => {
it('escalates when confidence < 0.7', () => {
expect(shouldEscalate({ confidence: 0.6, taskType: 'bugfix' })).toBe(true);
});
it('does NOT escalate on micro', () => {
expect(shouldEscalate({ confidence: 0.4, taskType: 'unknown', micro: true })).toBe(false);
});
it('does NOT escalate when confidence >= 0.7', () => {
expect(shouldEscalate({ confidence: 0.9, taskType: 'bugfix' })).toBe(false);
});
});
describe('classify — full integration (with mock LLM)', () => {
it('returns regex result when confidence high', async () => {
const r = await classify('почини списание дублируется', fakeRegistry, { llmCall: () => { throw new Error('should not call LLM'); } });
expect(r.source).toBe('regex');
expect(r.recommendedNode).toBe('#62');
});
it('escalates to LLM when confidence low', async () => {
const r = await classify('что-то непонятное', fakeRegistry, {
llmCall: async () => ({ taskType: 'question', micro: false, recommendedNode: null, confidence: 0.95, recommendedChain: null })
});
expect(r.source).toBe('llm');
expect(r.taskType).toBe('question');
});
it('uses cache on second call with same prompt', async () => {
let calls = 0;
const llmCall = async () => { calls++; return { taskType: 'feature', micro: false, recommendedNode: '#19', confidence: 0.9, recommendedChain: 'L1' }; };
const cache = new Map();
await classify('ambiguous query', fakeRegistry, { llmCall, cache });
await classify('ambiguous query', fakeRegistry, { llmCall, cache });
expect(calls).toBe(1); // Second hit cache.
});
});