@dommaker/harness 1.5.0 → 1.6.0

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

@@ -1,476 +0,0 @@
-/**
- * harness update-user-model — 用户模型持续演化引擎
- *
- * 不依赖固定规则快照。每天从最新数据中提取增量信号，
- * 对比昨天状态，输出变化，更新模型。
- *
- * 数据源:
- *   ~/.claude/projects/-root--claude/*.jsonl  (对话)
- *   /tmp/claude-knowledge-hooks.log            (行为信号)
- *   studio/.harness/knowledge/                 (知识新鲜度)
- *   ~/.claude/projects/-root-projects/memory/  (规则库)
- *
- * 模型状态: $HARNESS_UUM_STATE_FILE，默认 ~/.claude/user-model-state.json
- * 画像输出: $HARNESS_UUM_PROFILE_FILE，默认 ~/.claude/projects/-root-projects/memory/user_profile.md
- *
- * 工单 19-C：transcript 解析/纠正模式/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 TranscriptFilter,
-  extractCorrectionMatches,
-  cleanCorrectionConcept,
-  jaccardChinese,
-} from '../session-mining';
-import { log, processIO, type CommandIO, type CommandResult } from '../command-contract';
-
-export interface UpdateUserModelOptions {
-  days?: number;    // 只处理最近 N 天（自然日，含今天）的会话；缺省不过滤（向后兼容）
-  json?: boolean;
-  dryRun?: boolean;  // don't update state, just show what would change
-}
-
-interface ModelState {
-  lastUpdated: string;
-  sessionsProcessed: string[];       // session IDs already counted
-  patterns: Record<string, {         // pattern → running stats
-    firstSeen: string;
-    occurrences: number;
-    sessions: string[];
-    trend: 'rising' | 'stable' | 'falling' | 'new' | 'gone';
-    lastSeen: string;
-  }>;
-  lensWeights: Record<string, number>;  // lens → activation count → normalized weight
-  principleWeights: Record<string, number>;
-  evolutionLog: Array<{ date: string; change: string }>;
-}
-
-// 路径集中解析（harness#116）：默认值保持原行为，env 可覆盖（多工作区/测试隔离）。
-// 运行时解析而非模块加载期常量，保证 env 设置后生效。
-export interface UserModelPaths {
-  stateFile: string;
-  profileFile: string;
-}
-
-export function resolveUserModelPaths(env: NodeJS.ProcessEnv = process.env): UserModelPaths {
-  const home = os.homedir();
-  return {
-    stateFile: env.HARNESS_UUM_STATE_FILE
-      || path.join(home, '.claude', 'user-model-state.json'),
-    profileFile: env.HARNESS_UUM_PROFILE_FILE
-      || path.join(home, '.claude', 'projects', '-root-projects', 'memory', 'user_profile.md'),
-  };
-}
-
-export async function updateUserModel(options: UpdateUserModelOptions, io: CommandIO = processIO): Promise<CommandResult> {
-  const transcriptDir = process.env.CLAUDE_TRANSCRIPTS_DIR
-    || path.join(os.homedir(), '.claude', 'projects', '-root--claude');
-
-  // 1. Load previous state
-  const paths = resolveUserModelPaths();
-  const state = loadState(paths);
-
-  // 2. Scan new data
-  const newSessions = findNewSessions(transcriptDir, state.sessionsProcessed, options.days);
-  if (newSessions.length === 0) {
-    log(io, chalk.gray('No new sessions to process'));
-    return { kind: 'skip', reason: '没有新会话可处理' };
-  }
-
-  // 3. Extract signals from new data
-  const signals = extractSignals(newSessions);
-
-  // 4. Build concept clusters (semantic dedup)
-  const mergedConcepts = buildMergedConcepts(signals);
-
-  // 5. Diff against previous state → find changes
-  const changes = diffState(state, mergedConcepts, signals);
-
-  // 6. Update state
-  if (!options.dryRun) {
-    for (const sid of newSessions.map(s => s.id)) {
-      if (!state.sessionsProcessed.includes(sid)) {
-        state.sessionsProcessed.push(sid);
-      }
-    }
-    applySignals(state, signals, mergedConcepts);
-    state.lastUpdated = new Date().toISOString();
-
-    if (state.sessionsProcessed.length > 200) {
-      state.sessionsProcessed = state.sessionsProcessed.slice(-200);
-    }
-
-    saveState(state, paths);
-    updateProfile(state, paths);
-  }
-
-  // 6. Output
-  if (options.json) {
-    log(io, JSON.stringify({ newSessions: newSessions.length, changes }, null, 2));
-    return { kind: 'ok' };
-  }
-
-  log(io, chalk.blue(`📊 Processed ${newSessions.length} new sessions\n`));
-  printChanges(changes, signals, io);
-  return { kind: 'ok' };
-}
-
-// ── State I/O ──
-
-function loadState(paths: UserModelPaths): ModelState {
-  try {
-    if (fs.existsSync(paths.stateFile)) {
-      return JSON.parse(fs.readFileSync(paths.stateFile, 'utf-8'));
-    }
-  } catch {}
-  return {
-    lastUpdated: new Date(0).toISOString(),
-    sessionsProcessed: [],
-    patterns: {},
-    lensWeights: {},
-    principleWeights: {},
-    evolutionLog: [],
-  };
-}
-
-function saveState(state: ModelState, paths: UserModelPaths): void {
-  try {
-    const dir = path.dirname(paths.stateFile);
-    if (!fs.existsSync(dir)) fs.mkdirSync(dir, { recursive: true });
-    fs.writeFileSync(paths.stateFile, JSON.stringify(state, null, 2), 'utf-8');
-  } catch {}
-}
-
-// ── Session scanning ──
-
-interface SimpleSession {
-  id: string;
-  date: string;
-  userText: string;       // all user messages joined
-  assistantText: string;  // all assistant messages joined
-  toolCalls: string[];     // tool names used
-}
-
-function findNewSessions(dir: string, processed: string[], days?: number): SimpleSession[] {
-  // 过滤下推 seam（harness#112）：excludeIds 走文件名、since 走 stat，未命中不 parse
-  const filter: TranscriptFilter = { excludeIds: processed };
-
-  // 「最近 N 天」：自然日窗口（含今天）。days<=0 视为不过滤，与缺省一致。
-  // date 是 mtime 的 UTC 日，date >= cutoff ⟺ mtimeMs >= cutoff 日 UTC 零点
-  if (days !== undefined && days > 0) {
-    const cutoff = new Date(Date.now() - (days - 1) * 86_400_000)
-      .toISOString()
-      .slice(0, 10);
-    filter.since = Date.parse(cutoff);
-  }
-
-  const sessions = readTranscriptSessions(dir, filter);
-
-  sessions.sort((a, b) => a.date.localeCompare(b.date));
-
-  return sessions.map(s => ({
-    id: s.id,
-    date: s.date,
-    userText: s.turns.filter(t => t.role === 'user').map(t => t.content).join('\n'),
-    assistantText: s.turns.filter(t => t.role === 'assistant').map(t => t.content).join('\n'),
-    toolCalls: s.toolCalls,
-  }));
-}
-
-// ── Signal extraction ──
-
-interface SessionSignals {
-  sessionId: string;
-  date: string;
-  correctionPhrases: string[];        // "你又..." etc.
-  concepts: Record<string, number>;   // N-gram → frequency
-  toolPatterns: Record<string, number>; // tool → count
-  sessionDuration: number;            // estimated from turn count
-  turnCount: number;
-  deepAnalysis: boolean;              // EnterPlanMode or Agent(Explore)
-  knowledgeCaptured: boolean;         // Write to knowledge-docs
-}
-
-function extractCorrectionConcepts(userText: string): string[] {
-  const concepts: string[] = [];
-  for (const sentence of extractCorrectionMatches(userText)) {
-    const cleaned = cleanCorrectionConcept(sentence);
-    if (cleaned.length >= 4 && cleaned.length <= 60) concepts.push(cleaned);
-  }
-  return [...new Set(concepts)];
-}
-
-// ── Semantic clustering ──
-
-function clusterConcepts(
-  agg: Record<string, { count: number; sessions: string[] }>
-): Record<string, { count: number; sessions: string[] }> {
-  const entries = Object.entries(agg).filter(([k]) => k.length >= 4);
-  if (entries.length <= 1) return agg;
-
-  const merged: Record<string, { count: number; sessions: string[] }> = {};
-  const used = new Set<number>();
-
-  for (let i = 0; i < entries.length; i++) {
-    if (used.has(i)) continue;
-    const [concept, data] = entries[i];
-    let clusterName = concept;
-    let clusterCount = data.count;
-    const clusterSessions = new Set(data.sessions);
-    used.add(i);
-
-    for (let j = i + 1; j < entries.length; j++) {
-      if (used.has(j)) continue;
-      const [other, otherData] = entries[j];
-      const sim = jaccardChinese(concept, other);
-      if (sim > 0.5) {
-        clusterCount += otherData.count;
-        for (const s of otherData.sessions) clusterSessions.add(s);
-        // Use the shorter name as cluster label
-        if (other.length < clusterName.length) clusterName = other;
-        used.add(j);
-      }
-    }
-
-    merged[clusterName] = { count: clusterCount, sessions: [...clusterSessions] };
-  }
-
-  return merged;
-}
-
-function extractSignals(sessions: SimpleSession[]): SessionSignals[] {
-  return sessions.map(s => {
-    const correctionPhrases = extractCorrectionConcepts(s.userText);
-
-    // Concept extraction (Chinese char N-grams, etc.)
-    const concepts: Record<string, number> = {};
-    const cleanText = s.userText
-      .replace(/```[\s\S]*?```/g, '')
-      .replace(/\{[^{}]*\}/g, '')
-      .replace(/[{}[\]"':,\\]+/g, ' ');
-    let cnSeq = '';
-    for (const char of cleanText) {
-      if (/[\u4e00-\u9fff]/.test(char)) {
-        cnSeq += char;
-      } else {
-        if (cnSeq.length >= 4) concepts[cnSeq] = (concepts[cnSeq] || 0) + 1;
-        cnSeq = '';
-      }
-    }
-    if (cnSeq.length >= 4) concepts[cnSeq] = (concepts[cnSeq] || 0) + 1;
-
-    // Tool patterns
-    const toolPatterns: Record<string, number> = {};
-    for (const t of s.toolCalls) {
-      toolPatterns[t] = (toolPatterns[t] || 0) + 1;
-    }
-
-    return {
-      sessionId: s.id,
-      date: s.date,
-      correctionPhrases,
-      concepts,
-      toolPatterns,
-      sessionDuration: s.userText.length + s.assistantText.length, // proxy
-      turnCount: (s.userText.match(/\n/g) || []).length,
-      deepAnalysis: s.toolCalls.includes('EnterPlanMode') || s.toolCalls.includes('Agent'),
-      knowledgeCaptured: s.toolCalls.includes('Write') && (
-        s.assistantText.includes('.harness/knowledge-docs/') ||
-        s.assistantText.includes('.harness/knowledge/')
-      ),
-    };
-  });
-}
-
-// ── Concept aggregation ──
-
-type ConceptMap = Record<string, { count: number; sessions: string[] }>;
-
-function buildMergedConcepts(signals: SessionSignals[]): ConceptMap {
-  const agg: ConceptMap = {};
-  for (const sig of signals) {
-    for (const [concept, count] of Object.entries(sig.concepts)) {
-      if (!agg[concept]) agg[concept] = { count: 0, sessions: [] };
-      agg[concept].count += count;
-      if (!agg[concept].sessions.includes(sig.sessionId)) {
-        agg[concept].sessions.push(sig.sessionId);
-      }
-    }
-    for (const phrase of sig.correctionPhrases) {
-      if (phrase.length < 4) continue;
-      if (!agg[phrase]) agg[phrase] = { count: 0, sessions: [] };
-      agg[phrase].count += 3;
-      if (!agg[phrase].sessions.includes(sig.sessionId)) {
-        agg[phrase].sessions.push(sig.sessionId);
-      }
-    }
-  }
-  return clusterConcepts(agg);
-}
-
-// ── Diff & Apply ──
-
-interface Change {
-  type: 'new_pattern' | 'rising' | 'falling' | 'gone' | 'lens_shift';
-  key: string;
-  detail: string;
-}
-
-function diffState(state: ModelState, mergedConcepts: ConceptMap, signals: SessionSignals[]): Change[] {
-  const changes: Change[] = [];
-
-  // Detect new/rising/falling/gone concepts (from merged clusters)
-  for (const [concept, agg] of Object.entries(mergedConcepts)) {
-    const existing = state.patterns[concept];
-    if (!existing) {
-      if (agg.sessions.length >= 2) {
-        changes.push({ type: 'new_pattern', key: concept, detail: `Appeared in ${agg.sessions.length} sessions` });
-      }
-    } else if (existing.trend === 'gone' || existing.trend === 'falling') {
-      changes.push({ type: 'rising', key: concept, detail: `Re-emerged after being ${existing.trend}` });
-    }
-  }
-
-  // Detect lens weight shifts (from correction patterns)
-  const totalCorrections = signals.reduce((sum, s) => sum + s.correctionPhrases.length, 0);
-  if (totalCorrections >= 3) {
-    changes.push({ type: 'lens_shift', key: 'completeness',
-      detail: `${totalCorrections} corrections in these sessions — completeness lens may need weight increase` });
-  }
-
-  return changes;
-}
-
-function applySignals(state: ModelState, signals: SessionSignals[], mergedConcepts: ConceptMap): void {
-  const now = new Date().toISOString().slice(0, 10);
-
-  // Update patterns from merged concept clusters
-  // occurrences 唯一来源：mergedConcepts 聚合值（含 correction phrase 的 ×3 加权，
-  // 见 buildMergedConcepts）。不再叠加 per-session 原始 count —— 那是 #113 裁决的双计 bug。
-  for (const [concept, agg] of Object.entries(mergedConcepts)) {
-    if (!state.patterns[concept]) {
-      state.patterns[concept] = {
-        firstSeen: now, occurrences: 0, sessions: [], trend: 'new', lastSeen: now,
-      };
-    }
-    const p = state.patterns[concept];
-    p.occurrences += agg.count;
-    for (const sid of agg.sessions) {
-      if (!p.sessions.includes(sid)) p.sessions.push(sid);
-    }
-    p.lastSeen = now;
-    p.trend = p.occurrences >= 5 ? 'stable' : 'rising';
-  }
-
-  // Update lens weights from correction signals
-  for (const sig of signals) {
-    if (sig.correctionPhrases.length > 0) {
-      state.lensWeights.completeness = (state.lensWeights.completeness || 0) + 1;
-      state.lensWeights.automation = (state.lensWeights.automation || 0) + (sig.correctionPhrases.length > 2 ? 1 : 0);
-    }
-    if (sig.deepAnalysis) {
-      state.lensWeights.first_principles = (state.lensWeights.first_principles || 0) + 1;
-    }
-  }
-
-  // Detect falling/stable shifts
-  for (const [, p] of Object.entries(state.patterns)) {
-    if (p.trend === 'stable' && p.lastSeen < new Date(Date.now() - 7 * 86400000).toISOString().slice(0, 10)) {
-      p.trend = 'falling';
-    }
-  }
-}
-
-// ── Profile update ──
-
-function updateProfile(state: ModelState, paths: UserModelPaths): void {
-  try {
-    let content = fs.readFileSync(paths.profileFile, 'utf-8');
-
-    // Replace Derived Rules section
-    const derivedStart = '## Derived Rules';
-    const derivedEnd = '\n## Evolution';
-    const derivedIdx = content.indexOf(derivedStart);
-
-    const activePatterns = Object.entries(state.patterns)
-      .filter(([, p]) => (p.trend === 'rising' || p.trend === 'stable') && p.sessions.length >= 2)
-      .sort(([, a], [, b]) => b.occurrences - a.occurrences)
-      .slice(0, 10);
-
-    const derivedContent = [
-      '## Derived Rules (auto-generated by update-user-model)',
-      `Last scan: ${state.lastUpdated}. ${state.sessionsProcessed.length} sessions analyzed.`,
-      '',
-      '| Pattern | Trend | Occurrences | Sessions |',
-      '|---------|-------|-------------|----------|',
-      ...activePatterns.map(([k, p]) =>
-        `| ${k} | ${p.trend} | ${p.occurrences} | ${p.sessions.length} |`),
-      '',
-    ].join('\n');
-
-    if (derivedIdx >= 0) {
-      content = content.slice(0, derivedIdx) + derivedContent + '\n' + content.slice(content.indexOf(derivedEnd, derivedIdx));
-    } else {
-      // No derived section yet — append before Evolution
-      const evIdx = content.indexOf('## Evolution');
-      if (evIdx >= 0) {
-        content = content.slice(0, evIdx) + derivedContent + '\n' + content.slice(evIdx);
-      } else {
-        content += '\n' + derivedContent;
-      }
-    }
-
-    fs.writeFileSync(paths.profileFile, content, 'utf-8');
-  } catch (e) {
-    // Profile file might not exist yet — skip
-  }
-}
-
-// ── Output ──
-
-function printChanges(changes: Change[], signals: SessionSignals[], io: CommandIO): void {
-  if (changes.length === 0) {
-    log(io, chalk.green('No significant changes detected'));
-    return;
-  }
-
-  const byType = {
-    new_pattern: changes.filter(c => c.type === 'new_pattern'),
-    rising: changes.filter(c => c.type === 'rising'),
-    lens_shift: changes.filter(c => c.type === 'lens_shift'),
-  };
-
-  if (byType.new_pattern.length > 0) {
-    log(io, chalk.yellow(`🌱 New patterns: ${byType.new_pattern.length}`));
-    for (const c of byType.new_pattern.slice(0, 5)) {
-      log(io, chalk.gray(`   ${c.key}: ${c.detail}`));
-    }
-    log(io);
-  }
-
-  if (byType.rising.length > 0) {
-    log(io, chalk.green(`📈 Re-emerging: ${byType.rising.length}`));
-    for (const c of byType.rising.slice(0, 5)) {
-      log(io, chalk.gray(`   ${c.key}: ${c.detail}`));
-    }
-    log(io);
-  }
-
-  if (byType.lens_shift.length > 0) {
-    log(io, chalk.cyan(`🎯 Lens shifts:`));
-    for (const c of byType.lens_shift) {
-      log(io, chalk.gray(`   ${c.key}: ${c.detail}`));
-    }
-    log(io);
-  }
-
-  log(io, chalk.bold(`Total signals processed:`));
-  log(io, `  Sessions: ${signals.length}`);
-  log(io, `  Correction phrases: ${signals.reduce((s, sig) => s + sig.correctionPhrases.length, 0)}`);
-  log(io, `  Concepts extracted: ${signals.reduce((s, sig) => s + Object.keys(sig.concepts).length, 0)}`);
-}

package/src/cli/session-mining/__tests__/transcript.test.ts

@@ -1,118 +0,0 @@
-/**
- * readTranscriptSessions 坏行/null 记录容错测试（harness#82）
- *
- * transcript 行可 parse 但值为 null（或提取异常）时，
- * 与原逐行 catch 实现一致：该条跳过，不影响其余记录。
- */
-
-import * as fs from 'fs';
-import * as os from 'os';
-import * as path from 'path';
-import { readTranscriptSessions } from '../transcript';
-
-describe('readTranscriptSessions 坏行容错（harness#82）', () => {
-  let dir: string;
-
-  beforeEach(() => {
-    dir = fs.mkdtempSync(path.join(os.tmpdir(), 'harness-transcript-'));
-  });
-
-  afterEach(() => {
-    fs.rmSync(dir, { recursive: true, force: true });
-  });
-
-  it('null 行与坏行只丢该行，合法 turn 照常提取', () => {
-    const lines = [
-      JSON.stringify({ message: { role: 'user', content: 'hello' } }),
-      'null', // parse 合法但值为 null，原实现经逐行 catch 跳过
-      '{"broken',
-      JSON.stringify({ message: { role: 'assistant', content: [{ type: 'text', text: 'hi' }] } }),
-    ];
-    fs.writeFileSync(path.join(dir, 'session-a.jsonl'), lines.join('\n') + '\n', 'utf-8');
-
-    const sessions = readTranscriptSessions(dir);
-    expect(sessions).toHaveLength(1);
-    expect(sessions[0].turns.map(t => t.role)).toEqual(['user', 'assistant']);
-    expect(sessions[0].turns[1].content).toBe('hi');
-  });
-
-  it('目录不可读返回空数组', () => {
-    expect(readTranscriptSessions(path.join(dir, 'missing'))).toEqual([]);
-  });
-});
-
-describe('readTranscriptSessions 过滤（harness#112）', () => {
-  let dir: string;
-
-  beforeEach(() => {
-    dir = fs.mkdtempSync(path.join(os.tmpdir(), 'harness-transcript-filter-'));
-  });
-
-  afterEach(() => {
-    fs.rmSync(dir, { recursive: true, force: true });
-  });
-
-  function writeSession(name: string, text: string, mtime?: Date): void {
-    const p = path.join(dir, name);
-    fs.writeFileSync(
-      p,
-      JSON.stringify({ message: { role: 'user', content: text } }) + '\n',
-      'utf-8',
-    );
-    if (mtime) fs.utimesSync(p, mtime, mtime);
-  }
-
-  it('excludeIds：命中的会话不进结果（文件名级过滤，不 parse）', () => {
-    writeSession('session-a.jsonl', '甲会话内容');
-    writeSession('session-b.jsonl', '乙会话内容');
-
-    const sessions = readTranscriptSessions(dir, { excludeIds: ['session-a'] });
-    expect(sessions.map(s => s.id)).toEqual(['session-b']);
-  });
-
-  it('excludeIds 按截断 40 字符的会话 ID 匹配', () => {
-    const longName = 'a'.repeat(50);
-    writeSession(`${longName}.jsonl`, '长名会话内容');
-    writeSession('session-b.jsonl', '乙会话内容');
-
-    const sessions = readTranscriptSessions(dir, { excludeIds: [longName.slice(0, 40)] });
-    expect(sessions.map(s => s.id)).toEqual(['session-b']);
-  });
-
-  it('since：mtimeMs 早于窗口的文件不进结果（stat 级过滤，不 parse）', () => {
-    writeSession('old.jsonl', '旧会话内容', new Date(Date.now() - 30 * 86_400_000));
-    writeSession('recent.jsonl', '新会话内容');
-
-    const since = Date.now() - 7 * 86_400_000;
-    const sessions = readTranscriptSessions(dir, { since });
-    expect(sessions.map(s => s.id)).toEqual(['recent']);
-  });
-
-  it('since 边界含等于：mtimeMs == since 仍计入', () => {
-    const t = new Date('2026-08-01T12:00:00.000Z');
-    writeSession('edge.jsonl', '边界会话内容', t);
-
-    const sessions = readTranscriptSessions(dir, { since: t.getTime() });
-    expect(sessions.map(s => s.id)).toEqual(['edge']);
-  });
-
-  it('since 与 excludeIds 可叠加', () => {
-    writeSession('old.jsonl', '旧会话内容', new Date(Date.now() - 30 * 86_400_000));
-    writeSession('excluded.jsonl', '被排除会话');
-    writeSession('kept.jsonl', '保留会话内容');
-
-    const sessions = readTranscriptSessions(dir, {
-      since: Date.now() - 7 * 86_400_000,
-      excludeIds: ['excluded'],
-    });
-    expect(sessions.map(s => s.id)).toEqual(['kept']);
-  });
-
-  it('无 filter：全量读取（向后兼容）', () => {
-    writeSession('a.jsonl', '甲会话内容', new Date(Date.now() - 30 * 86_400_000));
-    writeSession('b.jsonl', '乙会话内容');
-
-    const sessions = readTranscriptSessions(dir);
-    expect(sessions).toHaveLength(2);
-  });
-});

package/src/cli/session-mining/corrections.ts

@@ -1,56 +0,0 @@
-/**
- * 纠正信号模式表与概念清洗（工单 19-C）
- *
- * 合并 analyze-sessions 与 update-user-model 两份几乎相同的模式表，
- * 采用较宽窗口（update-user-model 版本）作为唯一实现。
- */
-
-/** 用户纠正/重复强调语句的匹配模式（两命令合并后的超集，取较宽窗口） */
-export const CORRECTION_PATTERNS: RegExp[] = [
-  /你(?:又|还是)(?:在)?(?:犯|忘|没|不).{0,30}?[了]?/g,
-  /我不是(?:说|让|让做|讲)[了过]?.{0,50}?[了吗?？]?/g,
-  /怎么(?:又|还|老)(?:是|在)?.{0,30}?[了]?/g,
-  /我(?:一直|反复|总是)(?:在|说|强调)?.{0,30}?[了]?/g,
-  /老(?:是|在|犯|忘)(?:了)?.{0,20}?[了]?/g,
-  /这(?:个|种)(?:问题|模式|错误).{0,20}?[.。！]?/g,
-  /(?:第[一二三四五六七八九十\d]+次|反复|重复)(?:说|提醒|强调).{0,20}?/g,
-  /不(?:要|能|想|愿意).{0,20}?老.{0,10}?[了]?/g,
-  /(?:补上|加上|记上|修复).{0,10}?[了吗?？]?/g,
-  /(?:沉淀|监控|日志|记录).{0,5}?[了吗?？]?/g,
-];
-
-/** 纠正语句前缀（提取概念时剥离） */
-const CORRECTION_PREFIX_STRIP: RegExp[] = [
-  /你(?:又|还是)(?:在)?(?:犯|忘|没|不)/g,
-  /我不是(?:说|让|让做|讲)[了过]?/g,
-  /怎么(?:又|还|老)(?:是|在)?/g,
-  /我(?:一直|反复|总是)(?:在|说|强调)?/g,
-  /老(?:是|在|犯|忘)(?:了)?/g,
-];
-
-/**
- * 从纠正语句中提取概念：剥离纠正前缀与标点
- */
-export function cleanCorrectionConcept(sentence: string): string {
-  let cleaned = sentence;
-  for (const re of CORRECTION_PREFIX_STRIP) {
-    re.lastIndex = 0;
-    cleaned = cleaned.replace(re, '');
-  }
-  return cleaned.replace(/[，。！？、；：""''（）\s]+/g, '').trim();
-}
-
-/**
- * 用模式表在文本中抽取纠正语句（去重）
- */
-export function extractCorrectionMatches(text: string): string[] {
-  const matches: string[] = [];
-  for (const pattern of CORRECTION_PATTERNS) {
-    pattern.lastIndex = 0;
-    let match: RegExpExecArray | null;
-    while ((match = pattern.exec(text)) !== null) {
-      matches.push(match[0].trim());
-    }
-  }
-  return matches;
-}

package/src/cli/session-mining/index.ts

@@ -1,13 +0,0 @@
-/**
- * 会话挖掘共享层（工单 19-C）
- *
- * update-user-model 与 analyze-sessions 两个命令此前各自维护一份
- * transcript 解析 / 纠正模式 / 分词与相似度实现，收敛到本目录：
- *   - transcript.ts  Claude Code transcript（.jsonl）读取
- *   - corrections.ts 纠正信号模式表与概念清洗
- *   - text.ts        分词 / 停用词 / Jaccard / 文本清洗
- */
-
-export * from './transcript';
-export * from './corrections';
-export * from './text';