@dommaker/harness 1.5.0 → 1.7.0

package/src/cli/commands/__tests__/update-user-model.test.ts

@@ -1,268 +0,0 @@
-/**
- * update-user-model 命令测试（O1：--days flag + --json/--dry-run 兼容）
- *
- * Seam：updateUserModel(options) 公开入口。
- * 隔离面：
- *   - os.homedir → 测试临时目录（state/profile 落 TEST_HOME，不触碰真实 ~/.claude）
- *   - readTranscriptSessions → fixture 会话（extractCorrectionMatches 等纯函数保持真实）
- */
-
-import * as fs from 'fs';
-import { captureIO, lastJsonOutput, type CapturingIO } from '../../command-contract';
-import * as path from 'path';
-import { updateUserModel, resolveUserModelPaths } from '../update-user-model';
-import { readTranscriptSessions, type MinedSession } from '../../session-mining';
-
-const TEST_HOME = '/tmp/harness-uum-test-home';
-
-jest.mock('os', () => ({
-  ...jest.requireActual('os'),
-  homedir: () => '/tmp/harness-uum-test-home',
-}));
-
-jest.mock('../../session-mining', () => ({
-  ...jest.requireActual('../../session-mining'),
-  readTranscriptSessions: jest.fn(),
-}));
-
-jest.mock('chalk', () => {
-  const id = (s: string) => s;
-  const chalkFn = Object.assign(id, {
-    gray: id,
-    blue: id,
-    green: id,
-    yellow: id,
-    cyan: id,
-    bold: id,
-    red: id,
-  });
-  return {
-    __esModule: true,
-    default: chalkFn,
-    gray: id,
-    blue: id,
-    green: id,
-    yellow: id,
-    cyan: id,
-    bold: id,
-    red: id,
-  };
-});
-
-const mockReadTranscriptSessions = readTranscriptSessions as jest.MockedFunction<
-  typeof readTranscriptSessions
->;
-
-const STATE_FILE = path.join(TEST_HOME, '.claude', 'user-model-state.json');
-const PROFILE_FILE = path.join(TEST_HOME, '.claude', 'projects', '-root-projects', 'memory', 'user_profile.md');
-
-function todayStr(offsetDays = 0): string {
-  return new Date(Date.now() - offsetDays * 86_400_000).toISOString().slice(0, 10);
-}
-
-function mkSession(partial: Partial<MinedSession> = {}): MinedSession {
-  return {
-    id: 'session-1',
-    date: todayStr(),
-    mtimeMs: Date.now(),
-    turns: [
-      { role: 'user', content: '数据库迁移方案需要执行' },
-      { role: 'assistant', content: '好的' },
-    ],
-    toolCalls: ['Read'],
-    ...partial,
-  };
-}
-
-let io: CapturingIO;
-beforeEach(() => {
-  io = captureIO();
-});
-
-describe('update-user-model command', () => {
-
-  beforeEach(() => {
-    jest.clearAllMocks();
-    fs.rmSync(TEST_HOME, { recursive: true, force: true });
-    fs.mkdirSync(path.join(TEST_HOME, '.claude', 'projects', '-root-projects', 'memory'), { recursive: true });
-    process.env.CLAUDE_TRANSCRIPTS_DIR = path.join(TEST_HOME, 'transcripts');
-  });
-
-  afterEach(() => {
-    delete process.env.CLAUDE_TRANSCRIPTS_DIR;
-    delete process.env.HARNESS_UUM_STATE_FILE;
-    delete process.env.HARNESS_UUM_PROFILE_FILE;
-  });
-
-  // 路径解析（harness#116）：默认值保持原行为，env 覆盖生效
-  test('resolveUserModelPaths 默认：homedir 下的原路径（现行为不变）', () => {
-    expect(resolveUserModelPaths({})).toEqual({
-      stateFile: STATE_FILE,
-      profileFile: PROFILE_FILE,
-    });
-  });
-
-  test('resolveUserModelPaths env 覆盖：HARNESS_UUM_STATE_FILE / HARNESS_UUM_PROFILE_FILE', () => {
-    expect(resolveUserModelPaths({
-      HARNESS_UUM_STATE_FILE: '/custom/state.json',
-      HARNESS_UUM_PROFILE_FILE: '/custom/profile.md',
-    })).toEqual({
-      stateFile: '/custom/state.json',
-      profileFile: '/custom/profile.md',
-    });
-  });
-
-  test('env 覆盖端到端：state 落 env 指定路径，不写默认路径', async () => {
-    const customState = path.join(TEST_HOME, 'custom', 'state.json');
-    process.env.HARNESS_UUM_STATE_FILE = customState;
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'session-a' })]);
-
-    await updateUserModel({}, io);
-
-    expect(fs.existsSync(customState)).toBe(true);
-    expect(fs.existsSync(STATE_FILE)).toBe(false);
-    const state = JSON.parse(fs.readFileSync(customState, 'utf-8'));
-    expect(state.sessionsProcessed).toContain('session-a');
-  });
-
-  test('无新会话：提示 No new sessions to process', async () => {
-    mockReadTranscriptSessions.mockReturnValue([]);
-
-    await updateUserModel({}, io);
-
-    expect(io.outText()).toContain('No new sessions to process');
-    expect(fs.existsSync(STATE_FILE)).toBe(false);
-  });
-
-  test('--json：输出 newSessions 与 new_pattern 变化', async () => {
-    mockReadTranscriptSessions.mockReturnValue([
-      mkSession({ id: 'session-a' }),
-      mkSession({ id: 'session-b' }),
-    ]);
-
-    await updateUserModel({ json: true, dryRun: true }, io);
-
-    const output = lastJsonOutput(io);
-    expect(output.newSessions).toBe(2);
-    expect(output.changes).toEqual(
-      expect.arrayContaining([
-        expect.objectContaining({
-          type: 'new_pattern',
-          key: '数据库迁移方案需要执行',
-        }),
-      ]),
-    );
-  });
-
-  test('--dry-run：只展示变化，不写 state 文件', async () => {
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'session-a' })]);
-
-    await updateUserModel({ dryRun: true }, io);
-
-    expect(io.outText()).toContain('Processed 1 new sessions');
-    expect(fs.existsSync(STATE_FILE)).toBe(false);
-  });
-
-  test('默认（非 dry-run）：落盘 state 并记录已处理会话', async () => {
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'session-a' })]);
-
-    await updateUserModel({}, io);
-
-    expect(fs.existsSync(STATE_FILE)).toBe(true);
-    const state = JSON.parse(fs.readFileSync(STATE_FILE, 'utf-8'));
-    expect(state.sessionsProcessed).toContain('session-a');
-  });
-
-  test('--days 1：since 过滤下推到 seam（自然日窗口，含今天）', async () => {
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'today' })]);
-
-    await updateUserModel({ json: true, dryRun: true, days: 1 }, io);
-
-    const output = lastJsonOutput(io);
-    expect(output.newSessions).toBe(1);
-    // date >= 今日 ⟺ mtimeMs >= 今日 UTC 零点，过滤在 seam 内 stat 级完成
-    expect(mockReadTranscriptSessions).toHaveBeenCalledWith(
-      process.env.CLAUDE_TRANSCRIPTS_DIR,
-      { excludeIds: [], since: Date.parse(todayStr()) },
-    );
-  });
-
-  test('缺省 --days：处理全部未处理会话（向后兼容）', async () => {
-    mockReadTranscriptSessions.mockReturnValue([
-      mkSession({ id: 'today' }),
-      mkSession({
-        id: 'three-days-ago',
-        date: todayStr(3),
-        turns: [
-          { role: 'user', content: '旧会话概念内容测试' },
-          { role: 'assistant', content: '' },
-        ],
-      }),
-    ]);
-
-    await updateUserModel({ json: true, dryRun: true }, io);
-
-    const output = lastJsonOutput(io);
-    expect(output.newSessions).toBe(2);
-    // 缺省 days：不下推 since（向后兼容全量）
-    expect(mockReadTranscriptSessions).toHaveBeenCalledWith(
-      process.env.CLAUDE_TRANSCRIPTS_DIR,
-      { excludeIds: [] },
-    );
-  });
-
-  test('同一 concept 只计一次：occurrences 等于 merged 聚合值（#113 双计 bug）', async () => {
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'session-a' })]);
-
-    await updateUserModel({}, io);
-
-    const state = JSON.parse(fs.readFileSync(STATE_FILE, 'utf-8'));
-    // userText 整条中文序列恰为一个 concept（count=1）；聚合后只应出现一次，
-    // 双计 bug 下会被 merged 循环 + per-session 循环各加一次变成 2
-    expect(state.patterns['数据库迁移方案需要执行'].occurrences).toBe(1);
-    expect(state.patterns['数据库迁移方案需要执行'].sessions).toEqual(['session-a']);
-  });
-
-  test('correction phrase 的 ×3 加权不受影响（buildMergedConcepts 内）', async () => {
-    mockReadTranscriptSessions.mockReturnValue([
-      mkSession({
-        id: 'session-a',
-        turns: [
-          { role: 'user', content: '这种问题反复出现' },
-          { role: 'assistant', content: '' },
-        ],
-      }),
-    ]);
-
-    await updateUserModel({}, io);
-
-    const state = JSON.parse(fs.readFileSync(STATE_FILE, 'utf-8'));
-    // 纠正语句提取的 concept 按 ×3 计入（裁决：权重在 buildMergedConcepts，删循环不影响）
-    expect(state.patterns['这种问题'].occurrences).toBe(3);
-  });
-
-  test('sessionsProcessed 去重：已处理会话经 excludeIds 下推排除', async () => {
-    fs.writeFileSync(
-      STATE_FILE,
-      JSON.stringify({
-        lastUpdated: '',
-        sessionsProcessed: ['session-a'],
-        patterns: {},
-        lensWeights: {},
-        principleWeights: {},
-        evolutionLog: [],
-      }),
-      'utf-8',
-    );
-    mockReadTranscriptSessions.mockReturnValue([mkSession({ id: 'session-b' })]);
-
-    await updateUserModel({ json: true, dryRun: true, days: 7 }, io);
-
-    const output = lastJsonOutput(io);
-    expect(output.newSessions).toBe(1);
-    expect(mockReadTranscriptSessions).toHaveBeenCalledWith(
-      process.env.CLAUDE_TRANSCRIPTS_DIR,
-      { excludeIds: ['session-a'], since: expect.any(Number) },
-    );
-  });
-});

package/src/cli/commands/analyze-sessions.ts

@@ -1,335 +0,0 @@
-/**
- * harness analyze-sessions — 对话模式发现引擎
- *
- * 扫描 transcript 文件，不预设关键词，挖掘：
- *   1. 纠正信号（用户反复指出的问题）
- *   2. 高频 N-gram（跨会话重复出现的概念）
- *   3. 规则缺口（高频模式未被现有 memory 覆盖）
- *
- * 输出候选规则建议，供用户审核后写入 memory。
- *
- * 工单 19-C：transcript 解析/纠正模式/分词相似度收敛至 cli/session-mining/。
- */
-
-import chalk from 'chalk';
-import * as fs from 'fs';
-import * as path from 'path';
-import * as os from 'os';
-import {
-  readTranscriptSessions,
-  type MinedSession,
-  extractCorrectionMatches,
-  cleanCorrectionConcept,
-  tokenize,
-  stripCodeBlocks,
-  jaccardSimilarity,
-  isPunctuation,
-  isCodeNoise,
-  hasSemanticContent,
-  STOP_WORDS,
-} from '../session-mining';
-import { log, processIO, type CommandIO, type CommandResult } from '../command-contract';
-
-export interface AnalyzeSessionsOptions {
-  days?: number;
-  projectPath?: string;
-  json?: boolean;
-}
-
-interface Correction {
-  sentence: string;
-  sessionId: string;
-  timestamp: string;
-}
-
-interface PatternCandidate {
-  pattern: string;           // 被纠正/重复的概念
-  frequency: number;          // 跨会话出现次数
-  sessions: string[];         // 出现的会话 ID
-  source: 'correction' | 'ngram';  // 来源
-  confidence: number;         // 0-1
-  suggestedRule: string;      // 建议的规则名
-}
-
-// ── Main ──
-
-export async function analyzeSessions(options: AnalyzeSessionsOptions, io: CommandIO = processIO): Promise<CommandResult> {
-  const transcriptDir = process.env.CLAUDE_TRANSCRIPTS_DIR
-    || path.join(os.homedir(), '.claude', 'projects', '-root--claude');
-  const memoryDir = path.join(os.homedir(), '.claude', 'projects', '-root-projects', 'memory');
-  const days = options.days || 7;
-
-  if (!fs.existsSync(transcriptDir)) {
-    log(io, chalk.yellow('No transcripts directory found'));
-    return { kind: 'skip', reason: `会话记录目录不存在: ${transcriptDir}` };
-  }
-
-  // 1. Scan transcripts（since 下推 seam（harness#112）：stat 级过滤，窗口外不 parse；倒序）
-  const cutoff = Date.now() - days * 24 * 60 * 60 * 1000;
-  const sessions = readTranscriptSessions(transcriptDir, { since: cutoff })
-    .sort((a, b) => b.date.localeCompare(a.date));
-
-  if (sessions.length === 0) {
-    log(io, chalk.yellow(`No sessions found in the last ${days} days`));
-    return { kind: 'skip', reason: `最近 ${days} 天没有会话` };
-  }
-
-  // 2. Extract corrections + N-grams
-  const corrections = extractCorrections(sessions);
-  const ngrams = extractNGrams(sessions, { minLen: 4, maxLen: 24, minFreq: 2 });
-
-  // 3. Load existing memory rules for gap detection
-  const existingRules = loadExistingRules(memoryDir);
-
-  // 4. Generate candidates
-  const candidates = generateCandidates(corrections, ngrams, existingRules);
-
-  // 5. Output
-  if (options.json) {
-    log(io, JSON.stringify({ sessions: sessions.length, corrections: corrections.length, candidates }, null, 2));
-    return { kind: 'ok' };
-  }
-
-  log(io, chalk.blue(`🔍 Analyzing ${sessions.length} sessions (last ${days} days)...\n`));
-
-  printCorrectionSummary(corrections, io);
-  printCandidates(candidates, io);
-  return { kind: 'ok' };
-}
-
-// ── Correction Extraction ──
-
-function extractCorrections(sessions: MinedSession[]): Correction[] {
-  const corrections: Correction[] = [];
-
-  for (const session of sessions) {
-    for (const turn of session.turns) {
-      if (turn.role !== 'user') continue;
-      const text = stripCodeBlocks(turn.content);
-
-      for (const sentence of extractCorrectionMatches(text)) {
-        if (sentence.length > 3 && sentence.length < 200) {
-          corrections.push({
-            sentence,
-            sessionId: session.id,
-            timestamp: session.date,
-          });
-        }
-      }
-    }
-  }
-
-  return corrections;
-}
-
-// ── N-gram Mining ──
-
-interface NGramOptions {
-  minLen: number;
-  maxLen: number;
-  minFreq: number;
-}
-
-interface NGramResult {
-  phrase: string;
-  frequency: number;
-  sessions: string[];
-}
-
-function extractNGrams(sessions: MinedSession[], opts: NGramOptions): NGramResult[] {
-  const phraseMap = new Map<string, { count: number; sessions: Set<string> }>();
-
-  for (const session of sessions) {
-    const userTexts = session.turns
-      .filter(t => t.role === 'user' && !t.content.startsWith('{') && !t.content.startsWith('[') && t.content.length > 10)
-      .map(t => stripCodeBlocks(t.content));
-
-    const combined = userTexts.join(' ');
-
-    // Extract meaningful word sequences (Chinese chars + meaningful words)
-    const words = tokenize(combined);
-
-    for (let n = 2; n <= 5; n++) {
-      for (let i = 0; i <= words.length - n; i++) {
-        const phrase = words.slice(i, i + n).join('');
-        const clean = phrase.replace(/[，。！？、；：""''（）\s]/g, '');
-
-        if (clean.length < opts.minLen || clean.length > opts.maxLen) continue;
-        if (/^\d+$/.test(clean)) continue;
-        if (STOP_WORDS.has(clean)) continue;
-        if (isPunctuation(clean)) continue;
-
-        const existing = phraseMap.get(clean) || { count: 0, sessions: new Set() };
-        existing.count++;
-        existing.sessions.add(session.id);
-        phraseMap.set(clean, existing);
-      }
-    }
-  }
-
-  // Filter by minimum frequency and sort
-  return Array.from(phraseMap.entries())
-    .filter(([, v]) => v.count >= opts.minFreq && v.sessions.size >= 2)
-    .map(([phrase, v]) => ({ phrase, frequency: v.count, sessions: Array.from(v.sessions) }))
-    .filter(r => !isCodeNoise(r.phrase) && hasSemanticContent(r.phrase))
-    .sort((a, b) => b.frequency - a.frequency || b.sessions.length - a.sessions.length);
-}
-
-// ── Existing Rules ──
-
-function isInExistingRules(concept: string, existingRules: Set<string>): boolean {
-  const lower = concept.toLowerCase();
-  if (existingRules.has(lower)) return true;
-  // Partial match: concept appears in any rule name
-  for (const rule of existingRules) {
-    if (rule.includes(lower) || lower.includes(rule)) return true;
-  }
-  return false;
-}
-
-function loadExistingRules(memoryDir: string): Set<string> {
-  const concepts = new Set<string>();
-  try {
-    if (!fs.existsSync(memoryDir)) return concepts;
-    for (const file of fs.readdirSync(memoryDir)) {
-      if (!file.endsWith('.md')) continue;
-      const content = fs.readFileSync(path.join(memoryDir, file), 'utf-8');
-      // Extract key concepts from rule titles
-      const titleMatch = content.match(/#\s+(.+)/);
-      if (titleMatch) {
-        const words = tokenize(titleMatch[1]);
-        for (const w of words) {
-          if (w.length >= 4) concepts.add(w.toLowerCase());
-        }
-      }
-    }
-  } catch {}
-  return concepts;
-}
-
-// ── Candidate Generation ──
-
-function generateCandidates(
-  corrections: Correction[],
-  ngrams: NGramResult[],
-  existingRules: Set<string>,
-): PatternCandidate[] {
-  const candidates: PatternCandidate[] = [];
-
-  // From corrections: cluster similar sentences
-  const correctionClusters = clusterCorrections(corrections);
-  for (const cluster of correctionClusters) {
-    // Must appear at least 3 times total
-    if (cluster.sentences.length < 3) continue;
-    const concept = extractConcept(cluster.sentences);
-    if (isInExistingRules(concept, existingRules)) continue;
-
-    candidates.push({
-      pattern: concept,
-      frequency: cluster.sentences.length,
-      sessions: cluster.sessions,
-      source: 'correction',
-      confidence: Math.min(1, cluster.sentences.length / 10),
-      suggestedRule: `feedback_auto_${concept.toLowerCase().replace(/[^a-z0-9\u4e00-\u9fff]/g, '_').slice(0, 40)}.md`,
-    });
-  }
-
-  // From N-grams: high-frequency concepts not in corrections
-  const correctionConcepts = new Set(correctionClusters.map(c => extractConcept(c.sentences).toLowerCase()));
-  for (const ng of ngrams) {
-    const concept = ng.phrase;
-    if (correctionConcepts.has(concept.toLowerCase())) continue; // already covered
-    if (isInExistingRules(concept, existingRules)) continue;
-    if (ng.frequency < 2) continue;
-    const crossSessionStrength = ng.sessions.length >= 2;
-    const highFreqSingle = ng.sessions.length === 1 && ng.frequency >= 4;
-    if (!crossSessionStrength && !highFreqSingle) continue;
-
-    candidates.push({
-      pattern: concept,
-      frequency: ng.frequency,
-      sessions: ng.sessions,
-      source: 'ngram',
-      confidence: Math.min(1, ng.frequency / 10),
-      suggestedRule: `feedback_recurring_${concept.toLowerCase().replace(/[^a-z0-9\u4e00-\u9fff]/g, '_').slice(0, 40)}.md`,
-    });
-  }
-
-  return candidates.sort((a, b) => b.confidence - a.confidence);
-}
-
-interface CorrectionCluster {
-  sentences: string[];
-  sessions: string[];
-}
-
-function clusterCorrections(corrections: Correction[]): CorrectionCluster[] {
-  const clusters: CorrectionCluster[] = [];
-  const used = new Set<number>();
-
-  for (let i = 0; i < corrections.length; i++) {
-    if (used.has(i)) continue;
-    const cluster: CorrectionCluster = {
-      sentences: [corrections[i].sentence],
-      sessions: [corrections[i].sessionId],
-    };
-    used.add(i);
-
-    for (let j = i + 1; j < corrections.length; j++) {
-      if (used.has(j)) continue;
-      const similarity = jaccardSimilarity(corrections[i].sentence, corrections[j].sentence);
-      if (similarity > 0.3) {
-        cluster.sentences.push(corrections[j].sentence);
-        if (!cluster.sessions.includes(corrections[j].sessionId)) {
-          cluster.sessions.push(corrections[j].sessionId);
-        }
-        used.add(j);
-      }
-    }
-
-    clusters.push(cluster);
-  }
-
-  return clusters;
-}
-
-function extractConcept(sentences: string[]): string {
-  // Take the longest sentence and extract the noun phrase after the correction marker
-  const longest = sentences.reduce((a, b) => a.length > b.length ? a : b, '');
-  const cleaned = cleanCorrectionConcept(longest);
-  return cleaned.slice(0, 60) || longest.slice(0, 60);
-}
-
-// ── Output ──
-
-function printCorrectionSummary(corrections: Correction[], io: CommandIO): void {
-  if (corrections.length === 0) {
-    log(io, chalk.green('No correction patterns found'));
-    return;
-  }
-
-  log(io, chalk.yellow(`📢 ${corrections.length} correction signals detected\n`));
-}
-
-function printCandidates(candidates: PatternCandidate[], io: CommandIO): void {
-  if (candidates.length === 0) {
-    log(io, chalk.green('✅ No new pattern candidates — all recurring concepts already have rules\n'));
-    return;
-  }
-
-  log(io, chalk.bold('🔔 Pattern Candidates (suggested rules)\n'));
-
-  const topK = candidates.slice(0, 10);
-  for (const c of topK) {
-    const icon = c.source === 'correction' ? '🔴' : '🟡';
-    const confBar = '█'.repeat(Math.round(c.confidence * 10)) + '░'.repeat(10 - Math.round(c.confidence * 10));
-    log(io, chalk.bold(`${icon} ${c.pattern}`));
-    log(io, chalk.gray(`   Source: ${c.source} | Freq: ${c.frequency} | Sessions: ${c.sessions.length} | Confidence: ${confBar}`));
-    log(io, chalk.cyan(`   → Rule: ${c.suggestedRule}`));
-    log(io);
-  }
-
-  if (candidates.length > 10) {
-    log(io, chalk.gray(`... and ${candidates.length - 10} more candidates. Run with --json for full list.`));
-  }
-}