@pcircle/memesh 2.10.1 → 3.0.0

package/scripts/hooks/post-tool-use.js

@@ -1,895 +0,0 @@
-#!/usr/bin/env node
-
-/**
- * PostToolUse Hook - Claude Code Event-Driven Hooks
- *
- * Triggered after each tool execution in Claude Code.
- *
- * Features (Silent Observer):
- * - Reads tool execution data from stdin
- * - Detects patterns (READ_BEFORE_EDIT, Git workflows, Frontend work, Search patterns)
- * - Detects anomalies (slow execution, high tokens, failures, quota warnings)
- * - Updates recommendations.json incrementally
- * - Updates current-session.json
- * - Auto-saves key points to MeMesh when token threshold reached
- * - Runs silently (no console output - non-intrusive)
- */
-
-import {
-  STATE_DIR,
-  MEMESH_DB_PATH,
-  THRESHOLDS,
-  readJSONFile,
-  writeJSONFile,
-  writeJSONFileAsync,
-  sqliteBatchEntity,
-  readStdin,
-  logError,
-  logMemorySave,
-  getDateString,
-  isPlanFile,
-  parsePlanSteps,
-  derivePlanName,
-  sqliteQueryJSON,
-  updateEntityMetadata,
-  addObservation,
-} from './hook-utils.js';
-import { isTestCommand, extractTestFailureContext, buildTestFailureQuery, buildErrorQuery } from './post-tool-use-recall-utils.js';
-import fs from 'fs';
-import path from 'path';
-
-// ============================================================================
-// File Paths
-// ============================================================================
-
-const RECOMMENDATIONS_FILE = path.join(STATE_DIR, 'recommendations.json');
-const CURRENT_SESSION_FILE = path.join(STATE_DIR, 'current-session.json');
-const SESSION_CONTEXT_FILE = path.join(STATE_DIR, 'session-context.json');
-
-// ============================================================================
-// Pattern Detection
-// ============================================================================
-
-/**
- * Pattern Detector - Analyzes tool usage patterns
- */
-class PatternDetector {
-  constructor() {
-    this.recentTools = [];
-  }
-
-  /**
-   * Add a tool call to the recent tools list
-   * @param {Object} toolData - Tool execution data
-   */
-  addToolCall(toolData) {
-    this.recentTools.push({
-      toolName: toolData.toolName,
-      args: toolData.arguments,
-      timestamp: new Date().toISOString(),
-    });
-
-    // Keep only last 10 (using slice instead of shift for better performance)
-    if (this.recentTools.length > 10) {
-      this.recentTools = this.recentTools.slice(-10);
-    }
-  }
-
-  /**
-   * Detect patterns in recent tool usage
-   * @param {Object} currentTool - Current tool execution data
-   * @returns {Array} Detected patterns
-   */
-  detectPatterns(currentTool) {
-    const patterns = [];
-
-    if (!currentTool || !currentTool.toolName) {
-      return patterns;
-    }
-
-    const toolArgs = currentTool.arguments || {};
-
-    // Pattern 1: READ_BEFORE_EDIT
-    if (currentTool.toolName === 'Edit') {
-      const filePath = toolArgs.file_path;
-      if (filePath) {
-        const recentReads = this.recentTools.filter(t =>
-          t.toolName === 'Read' && t.args?.file_path === filePath
-        );
-        if (recentReads.length > 0) {
-          patterns.push({
-            type: 'READ_BEFORE_EDIT',
-            description: 'Read before Edit - correct behavior',
-            severity: 'info',
-          });
-        } else {
-          patterns.push({
-            type: 'EDIT_WITHOUT_READ',
-            description: `Edit ${path.basename(filePath)} without prior Read`,
-            severity: 'warning',
-          });
-        }
-      }
-    }
-
-    // Pattern 2: Git Workflow
-    const gitCommands = ['git add', 'git commit', 'git push', 'git branch', 'git checkout'];
-    if (currentTool.toolName === 'Bash' && toolArgs.command) {
-      const cmd = toolArgs.command;
-      if (gitCommands.some(gitCmd => cmd.includes(gitCmd))) {
-        const recentGitOps = this.recentTools.filter(t =>
-          t.toolName === 'Bash' && gitCommands.some(gc => t.args?.command?.includes(gc))
-        ).length;
-
-        if (recentGitOps >= 2) {
-          patterns.push({
-            type: 'GIT_WORKFLOW',
-            description: `Git workflow detected (${recentGitOps + 1} operations)`,
-            severity: 'info',
-            suggestion: 'Consider loading devops-git-workflows skill',
-          });
-        }
-      }
-    }
-
-    // Pattern 3: Frontend Work
-    const frontendExtensions = ['.tsx', '.jsx', '.vue', '.svelte', '.css', '.scss'];
-    if (['Edit', 'Write', 'Read'].includes(currentTool.toolName)) {
-      const filePath = toolArgs.file_path;
-      if (filePath && frontendExtensions.some(ext => filePath.endsWith(ext))) {
-        const recentFrontendOps = this.recentTools.filter(t =>
-          ['Edit', 'Write', 'Read'].includes(t.toolName) &&
-          frontendExtensions.some(ext => t.args?.file_path?.endsWith(ext))
-        ).length;
-
-        if (recentFrontendOps >= 2) {
-          patterns.push({
-            type: 'FRONTEND_WORK',
-            description: `Frontend work detected (${recentFrontendOps + 1} files)`,
-            severity: 'info',
-            suggestion: 'Consider loading frontend-design skill',
-          });
-        }
-      }
-    }
-
-    // Pattern 4: Intensive Search
-    if (['Grep', 'Glob'].includes(currentTool.toolName)) {
-      const recentSearches = this.recentTools.filter(t =>
-        ['Grep', 'Glob'].includes(t.toolName)
-      ).length;
-
-      if (recentSearches >= 3) {
-        patterns.push({
-          type: 'INTENSIVE_SEARCH',
-          description: `Multiple search operations (${recentSearches + 1} times)`,
-          severity: 'info',
-        });
-      }
-    }
-
-    return patterns;
-  }
-}
-
-// ============================================================================
-// Anomaly Detection
-// ============================================================================
-
-/**
- * Detect anomalies in tool execution
- * @param {Object} toolData - Tool execution data
- * @param {Object} sessionContext - Session context with quota info
- * @returns {Array} Detected anomalies
- */
-function detectAnomalies(toolData, sessionContext) {
-  const anomalies = [];
-
-  // Anomaly 1: Slow Execution
-  if (toolData.duration && toolData.duration > THRESHOLDS.SLOW_EXECUTION) {
-    anomalies.push({
-      type: 'SLOW_EXECUTION',
-      description: `${toolData.toolName} took ${(toolData.duration / 1000).toFixed(1)}s (slow)`,
-      severity: 'warning',
-    });
-  }
-
-  // Anomaly 2: High Token Usage
-  if (toolData.tokensUsed && toolData.tokensUsed > THRESHOLDS.HIGH_TOKENS) {
-    anomalies.push({
-      type: 'HIGH_TOKENS',
-      description: `${toolData.toolName} used ${toolData.tokensUsed} tokens (high usage)`,
-      severity: 'warning',
-    });
-  }
-
-  // Anomaly 3: Execution Failure
-  if (toolData.success === false) {
-    anomalies.push({
-      type: 'EXECUTION_FAILURE',
-      description: `${toolData.toolName} execution failed`,
-      severity: 'error',
-    });
-  }
-
-  // Anomaly 4: Quota Warning
-  if (sessionContext.tokenQuota) {
-    const quotaUsed = sessionContext.tokenQuota.used + (toolData.tokensUsed || 0);
-    const quotaPercentage = quotaUsed / sessionContext.tokenQuota.limit;
-
-    if (quotaPercentage > THRESHOLDS.QUOTA_WARNING) {
-      anomalies.push({
-        type: 'QUOTA_WARNING',
-        description: `Token quota at ${(quotaPercentage * 100).toFixed(1)}%`,
-        severity: 'warning',
-      });
-    }
-  }
-
-  return anomalies;
-}
-
-// ============================================================================
-// Recommendations Update
-// ============================================================================
-
-/**
- * Update recommendations based on detected patterns and anomalies
- * @param {Array} patterns - Detected patterns
- * @param {Array} anomalies - Detected anomalies
- */
-function updateRecommendations(patterns, anomalies) {
-  const recommendations = readJSONFile(RECOMMENDATIONS_FILE, {
-    recommendedSkills: [],
-    detectedPatterns: [],
-    warnings: [],
-    lastUpdated: null,
-  });
-
-  // Add new skills based on patterns
-  patterns.forEach(pattern => {
-    if (pattern.suggestion && pattern.suggestion.includes('skill')) {
-      const skillMatch = pattern.suggestion.match(/loading\s+(.+?)\s+skill/);
-      if (skillMatch) {
-        const skillName = skillMatch[1];
-        const existing = recommendations.recommendedSkills.find(s => s.name === skillName);
-        if (!existing) {
-          recommendations.recommendedSkills.push({
-            name: skillName,
-            reason: pattern.description,
-            priority: 'medium',
-          });
-        }
-      }
-    }
-  });
-
-  // Add detected patterns (keep last 10)
-  patterns.forEach(pattern => {
-    if (pattern && pattern.description) {
-      recommendations.detectedPatterns.unshift({
-        description: pattern.description,
-        suggestion: pattern.suggestion || '',
-        timestamp: new Date().toISOString(),
-      });
-    }
-  });
-  recommendations.detectedPatterns = recommendations.detectedPatterns.slice(0, 10);
-
-  // Add warnings from anomalies (keep last 5)
-  anomalies
-    .filter(a => a.severity === 'warning' || a.severity === 'error')
-    .forEach(anomaly => {
-      recommendations.warnings.unshift(anomaly.description);
-    });
-  recommendations.warnings = recommendations.warnings.slice(0, 5);
-
-  recommendations.lastUpdated = new Date().toISOString();
-
-  // Async write — non-blocking, caller awaits via pendingWrites
-  return writeJSONFileAsync(RECOMMENDATIONS_FILE, recommendations);
-}
-
-// ============================================================================
-// Session Update
-// ============================================================================
-
-/**
- * Update current session with tool call data
- * @param {Object} toolData - Tool execution data
- * @param {Array} patterns - Detected patterns
- * @param {Array} anomalies - Detected anomalies
- */
-function updateCurrentSession(toolData, patterns, anomalies) {
-  const currentSession = readJSONFile(CURRENT_SESSION_FILE, {
-    startTime: new Date().toISOString(),
-    toolCalls: [],
-    patterns: [],
-  });
-
-  // Add tool call record
-  currentSession.toolCalls.push({
-    timestamp: new Date().toISOString(),
-    toolName: toolData.toolName,
-    arguments: toolData.arguments,
-    duration: toolData.duration,
-    success: toolData.success,
-    tokenUsage: toolData.tokensUsed,
-  });
-
-  // Cap toolCalls to prevent unbounded growth in long sessions
-  const MAX_TOOL_CALLS = 1000;
-  if (currentSession.toolCalls.length > MAX_TOOL_CALLS) {
-    currentSession.toolCalls = currentSession.toolCalls.slice(-MAX_TOOL_CALLS);
-  }
-
-  // Track file modifications and test executions (for dry-run gate)
-  trackFileModifications(toolData, currentSession);
-  trackTestExecutions(toolData, currentSession);
-
-  // Update pattern counts
-  patterns.forEach(pattern => {
-    const existing = currentSession.patterns.find(p => p.type === pattern.type);
-    if (!existing) {
-      currentSession.patterns.push({
-        type: pattern.type,
-        count: 1,
-        firstDetected: new Date().toISOString(),
-      });
-    } else {
-      existing.count++;
-    }
-  });
-
-  // Async write — session file is read on next call, eventual consistency is fine
-  // Return promise so caller can track it in pendingWrites
-  const writePromise = writeJSONFileAsync(CURRENT_SESSION_FILE, currentSession);
-
-  return { session: currentSession, writePromise };
-}
-
-/**
- * Update session context (quota tracking)
- * @param {Object} toolData - Tool execution data
- * @returns {{ sessionContext: Object, writePromise: Promise<boolean> }}
- */
-function updateSessionContext(toolData) {
-  const sessionContext = readJSONFile(SESSION_CONTEXT_FILE, {
-    tokenQuota: { used: 0, limit: 200000 },
-    learnedPatterns: [],
-    lastSessionDate: null,
-    lastSaveTokens: 0,
-  });
-
-  // Update token usage
-  if (toolData.tokensUsed) {
-    sessionContext.tokenQuota.used += toolData.tokensUsed;
-  }
-
-  sessionContext.lastSessionDate = new Date().toISOString();
-
-  const writePromise = writeJSONFileAsync(SESSION_CONTEXT_FILE, sessionContext);
-
-  return { sessionContext, writePromise };
-}
-
-// ============================================================================
-// MeMesh Key Points Auto-Save
-// ============================================================================
-
-/**
- * Extract key points from session state
- * @param {Object} sessionState - Current session state
- * @returns {Array<string>} Extracted key points
- */
-function extractKeyPoints(sessionState) {
-  const keyPoints = [];
-
-  if (!sessionState?.toolCalls?.length) {
-    return keyPoints;
-  }
-
-  // 1. Identify completed file operations
-  const fileOps = {};
-  sessionState.toolCalls.forEach(tc => {
-    if (['Edit', 'Write'].includes(tc.toolName) && tc.arguments?.file_path) {
-      const filePath = tc.arguments.file_path;
-      fileOps[filePath] = (fileOps[filePath] || 0) + 1;
-    }
-  });
-
-  const modifiedFiles = Object.keys(fileOps);
-  if (modifiedFiles.length > 0) {
-    const summary = modifiedFiles.length > 5
-      ? `${modifiedFiles.slice(0, 5).map(f => path.basename(f)).join(', ')} (+${modifiedFiles.length - 5} more)`
-      : modifiedFiles.map(f => path.basename(f)).join(', ');
-    keyPoints.push(`[TASK] Modified files: ${summary}`);
-  }
-
-  // 2. Identify failures
-  const failures = sessionState.toolCalls.filter(tc => tc.success === false);
-  if (failures.length > 0) {
-    const failedTools = [...new Set(failures.map(f => f.toolName))];
-    keyPoints.push(`[PROBLEM] ${failures.length} tool failures: ${failedTools.join(', ')}`);
-  }
-
-  // 3. Git operations
-  const gitCommits = sessionState.toolCalls.filter(tc =>
-    tc.toolName === 'Bash' && tc.arguments?.command?.includes('git commit')
-  );
-  if (gitCommits.length > 0) {
-    keyPoints.push(`[DECISION] Made ${gitCommits.length} git commit(s)`);
-  }
-
-  // 4. Detected patterns
-  if (sessionState.patterns?.length > 0) {
-    const patternSummary = sessionState.patterns
-      .filter(p => p.count > 2)
-      .map(p => `${p.type}(${p.count})`)
-      .join(', ');
-    if (patternSummary) {
-      keyPoints.push(`[PATTERN] Recurring patterns: ${patternSummary}`);
-    }
-  }
-
-  // 5. Work scope indicator
-  const toolCounts = {};
-  sessionState.toolCalls.forEach(tc => {
-    toolCounts[tc.toolName] = (toolCounts[tc.toolName] || 0) + 1;
-  });
-  const topTools = Object.entries(toolCounts)
-    .sort((a, b) => b[1] - a[1])
-    .slice(0, 3)
-    .map(([tool, count]) => `${tool}:${count}`)
-    .join(', ');
-  keyPoints.push(`[SCOPE] Tool usage: ${topTools}, total: ${sessionState.toolCalls.length}`);
-
-  return keyPoints;
-}
-
-/**
- * Save conversation key points to MeMesh knowledge graph.
- * Uses sqliteBatchEntity for performance (3 spawns instead of N).
- * @param {Object} sessionState - Current session state
- * @param {Object} sessionContext - Session context
- * @returns {boolean} True if saved successfully
- */
-function saveConversationKeyPoints(sessionState, sessionContext) {
-  try {
-    if (!fs.existsSync(MEMESH_DB_PATH)) {
-      logError('saveConversationKeyPoints', `MeMesh DB not found: ${MEMESH_DB_PATH}`);
-      return false;
-    }
-
-    const keyPoints = extractKeyPoints(sessionState);
-    if (keyPoints.length === 0) {
-      return false;
-    }
-
-    const entityName = `session_keypoints_${Date.now()}`;
-
-    const metadata = JSON.stringify({
-      tokensSaved: sessionContext.tokenQuota?.used || 0,
-      toolCount: sessionState.toolCalls?.length || 0,
-      saveReason: 'token_threshold',
-    });
-
-    const tags = ['auto_saved', 'token_trigger', getDateString()];
-
-    const entityId = sqliteBatchEntity(
-      MEMESH_DB_PATH,
-      { name: entityName, type: 'session_keypoint', metadata },
-      keyPoints,
-      tags
-    );
-
-    if (entityId === null) {
-      return false;
-    }
-
-    logMemorySave(`🧠 MeMesh: Saved ${keyPoints.length} key points (tokens: ${sessionContext.tokenQuota?.used})`);
-
-    return true;
-  } catch (error) {
-    logError('saveConversationKeyPoints', error);
-    return false;
-  }
-}
-
-/**
- * Check if token threshold reached and save key points
- * @param {Object} sessionState - Current session state
- * @param {Object} sessionContext - Session context
- * @returns {boolean} True if saved
- */
-function checkAndSaveKeyPoints(sessionState, sessionContext) {
-  try {
-    const lastSaveTokens = sessionContext.lastSaveTokens || 0;
-    const currentTokens = sessionContext.tokenQuota?.used || 0;
-    const tokensSinceLastSave = currentTokens - lastSaveTokens;
-
-    if (tokensSinceLastSave >= THRESHOLDS.TOKEN_SAVE) {
-      const saved = saveConversationKeyPoints(sessionState, sessionContext);
-
-      if (saved) {
-        sessionContext.lastSaveTokens = currentTokens;
-        writeJSONFile(SESSION_CONTEXT_FILE, sessionContext);
-      }
-
-      return saved;
-    }
-
-    return false;
-  } catch (error) {
-    logError('checkAndSaveKeyPoints', error);
-    return false;
-  }
-}
-
-// ============================================================================
-// File Modification & Test Tracking (for dry-run gate in pre-tool-use.js)
-// ============================================================================
-
-/**
- * Track file modifications from Write/Edit tool calls.
- * Stores modified file paths in session state.
- * @param {Object} toolData - Normalized tool data
- * @param {Object} currentSession - Current session state (mutated in place)
- */
-function trackFileModifications(toolData, currentSession) {
-  if (!['Edit', 'Write'].includes(toolData.toolName)) return;
-
-  const filePath = toolData.arguments?.file_path;
-  if (!filePath) return;
-
-  if (!currentSession.modifiedFiles) {
-    currentSession.modifiedFiles = [];
-  }
-
-  const MAX_MODIFIED_FILES = 100;
-  if (!currentSession.modifiedFiles.includes(filePath)) {
-    if (currentSession.modifiedFiles.length >= MAX_MODIFIED_FILES) {
-      currentSession.modifiedFiles.shift(); // Remove oldest entry
-    }
-    currentSession.modifiedFiles.push(filePath);
-  }
-}
-
-/** Patterns that indicate test execution in a Bash command */
-const TEST_PATTERNS = [
-  /vitest\s+(run|watch)?/,
-  /jest\b/,
-  /npm\s+test/,
-  /npm\s+run\s+test/,
-  /npx\s+vitest/,
-  /npx\s+jest/,
-  /tsc\s+--noEmit/,
-  /node\s+--check\s/,
-  /bun\s+test/,
-  /pytest\b/,
-];
-
-/**
- * Track test executions from Bash tool calls.
- * Marks tested files/directories in session state.
- * @param {Object} toolData - Normalized tool data
- * @param {Object} currentSession - Current session state (mutated in place)
- */
-function trackTestExecutions(toolData, currentSession) {
-  if (toolData.toolName !== 'Bash') return;
-  if (!toolData.success) return;
-
-  const cmd = toolData.arguments?.command || '';
-  const isTest = TEST_PATTERNS.some(pattern => pattern.test(cmd));
-  if (!isTest) return;
-
-  if (!currentSession.testedFiles) {
-    currentSession.testedFiles = [];
-  }
-
-  currentSession.lastTestRun = new Date().toISOString();
-
-  // Extract test target path if provided
-  // e.g., "vitest run src/auth" → mark all modified files under src/auth/ as tested
-  const pathMatch = cmd.match(/(?:vitest|jest|node\s+--check)\s+(?:run\s+)?(\S+)/);
-  const testTarget = pathMatch ? pathMatch[1] : null;
-
-  if (testTarget && currentSession.modifiedFiles) {
-    // Mark files under the test target directory/path as tested
-    // Use path-prefix match: "src/auth" matches "src/auth/middleware.ts" but NOT "src/auth-utils/helper.ts"
-    for (const modFile of currentSession.modifiedFiles) {
-      const isMatch = modFile === testTarget ||
-        modFile.startsWith(testTarget + '/') ||
-        modFile.startsWith(testTarget + path.sep);
-      if (isMatch && !currentSession.testedFiles.includes(modFile)) {
-        currentSession.testedFiles.push(modFile);
-      }
-    }
-  } else if (!testTarget && currentSession.modifiedFiles) {
-    // Full test run (no specific target) — mark all modified files as tested
-    for (const modFile of currentSession.modifiedFiles) {
-      if (!currentSession.testedFiles.includes(modFile)) {
-        currentSession.testedFiles.push(modFile);
-      }
-    }
-  }
-}
-
-// ============================================================================
-// Code Review Tracking
-// ============================================================================
-
-/**
- * Check if this tool call is a code review invocation.
- * Detects both Skill tool usage and Task tool dispatching code reviewers.
- * @param {Object} toolData - Normalized tool data
- * @returns {boolean}
- */
-function isCodeReviewInvocation(toolData) {
-  // Skill tool with code review
-  if (toolData.toolName === 'Skill') {
-    const name = toolData.arguments?.name || toolData.arguments?.skill_name || '';
-    return /code.?review|comprehensive.?code.?review/i.test(name);
-  }
-
-  // Task tool dispatching code reviewer subagent
-  if (toolData.toolName === 'Task') {
-    const subagentType = toolData.arguments?.subagent_type || '';
-    return /code.?review/i.test(subagentType);
-  }
-
-  return false;
-}
-
-/**
- * Mark code review as done in session state.
- * This flag is checked by pre-tool-use.js before git commits.
- */
-function markCodeReviewDone() {
-  const session = readJSONFile(CURRENT_SESSION_FILE, {});
-  session.codeReviewDone = true;
-  session.codeReviewTimestamp = new Date().toISOString();
-  writeJSONFile(CURRENT_SESSION_FILE, session);
-}
-
-// ============================================================================
-// Tool Data Normalization
-// ============================================================================
-
-/**
- * Normalize tool data from Claude Code format
- * @param {Object} raw - Raw tool data from stdin
- * @returns {Object} Normalized tool data
- */
-function normalizeToolData(raw) {
-  return {
-    toolName: raw.tool_name || raw.toolName || 'unknown',
-    arguments: raw.tool_input || raw.arguments || {},
-    duration: raw.duration_ms || raw.duration || 0,
-    success: raw.success !== false,
-    tokensUsed: raw.tokens_used || raw.tokensUsed || 0,
-    _raw: raw,
-  };
-}
-
-// ============================================================================
-// Plan File Detection (Beta)
-// ============================================================================
-
-/**
- * Detect plan file creation and save to KG.
- * Triggered when Write tool targets a plan file path.
- * @param {Object} toolData - Normalized tool data
- */
-function detectPlanFile(toolData) {
-  if (toolData.toolName !== 'Write') return;
-
-  const filePath = toolData.arguments?.file_path;
-  if (!isPlanFile(filePath)) return;
-
-  try {
-    // Read the file content
-    if (!fs.existsSync(filePath)) return;
-    const content = fs.readFileSync(filePath, 'utf-8');
-
-    const steps = parsePlanSteps(content);
-    if (steps.length === 0) return;
-
-    if (!fs.existsSync(MEMESH_DB_PATH)) return;
-
-    const planName = derivePlanName(filePath);
-    const entityName = `Plan: ${planName}`;
-
-    const newMetadata = {
-      sourceFile: filePath,
-      totalSteps: steps.length,
-      completed: steps.filter(s => s.completed).length,
-      status: 'active',
-      stepsDetail: steps,
-    };
-
-    // Check if plan entity already exists (re-save scenario)
-    const existing = sqliteQueryJSON(MEMESH_DB_PATH,
-      'SELECT id FROM entities WHERE name = ?', [entityName]);
-
-    if (existing && existing.length > 0) {
-      // Upsert: update metadata and add observation for the re-save
-      updateEntityMetadata(MEMESH_DB_PATH, entityName, newMetadata);
-      addObservation(MEMESH_DB_PATH, entityName,
-        `Plan re-saved: ${steps.length} steps (${steps.filter(s => s.completed).length} completed)`);
-      logMemorySave(`Plan updated: ${planName} (${steps.length} steps)`);
-    } else {
-      // First save: create entity with observations and tags
-      const observations = steps.map(s => `Step ${s.number}: ${s.description}`);
-      const tags = ['plan', 'active', `plan:${planName}`, 'scope:project'];
-
-      sqliteBatchEntity(MEMESH_DB_PATH,
-        { name: entityName, type: 'workflow_checkpoint', metadata: JSON.stringify(newMetadata) },
-        observations, tags
-      );
-      logMemorySave(`Plan detected: ${planName} (${steps.length} steps)`);
-    }
-  } catch (error) {
-    logError('detectPlanFile', error);
-  }
-}
-
-// ============================================================================
-// Proactive Recall on Test Failure / Error Detection
-// ============================================================================
-
-/**
- * Trigger proactive recall on test failure or error detection.
- * Writes results to proactive-recall.json for HookToolHandler.
- * @param {Object} toolData - Normalized tool data
- */
-function triggerProactiveRecall(toolData) {
-  try {
-    const recallFile = path.join(STATE_DIR, 'proactive-recall.json');
-    let query = null;
-    let trigger = null;
-
-    // Test failure detection
-    if (toolData.toolName === 'Bash' && toolData.arguments?.command) {
-      if (isTestCommand(toolData.arguments.command) && !toolData.success) {
-        const ctx = extractTestFailureContext(toolData._raw?.output || '');
-        if (ctx) {
-          query = buildTestFailureQuery(ctx.testName, ctx.errorMessage);
-          trigger = 'test-failure';
-        }
-      }
-    }
-
-    // Error detection (non-test failures)
-    if (!trigger && !toolData.success && toolData._raw?.output) {
-      const errorMatch = toolData._raw.output.match(/(\w*Error):\s*(.+)/);
-      if (errorMatch) {
-        query = buildErrorQuery(errorMatch[1], errorMatch[2]);
-        trigger = 'error-detection';
-      }
-    }
-
-    if (!query || !trigger) return;
-
-    // Build FTS5 query
-    const ftsTokens = query.split(/\s+/)
-      .filter(t => t.length > 2)
-      .slice(0, 8)
-      .map(t => `"${t.replace(/"/g, '""')}"*`)
-      .join(' OR ');
-
-    if (!ftsTokens) return;
-
-    const sql = `
-      SELECT e.name,
-        (SELECT json_group_array(content) FROM observations o WHERE o.entity_id = e.id) as observations_json
-      FROM entities e
-      JOIN entities_fts ON entities_fts.rowid = e.id
-      WHERE entities_fts MATCH ?
-      ORDER BY bm25(entities_fts, 10.0, 5.0)
-      LIMIT 3
-    `;
-
-    const result = sqliteQueryJSON(MEMESH_DB_PATH, sql, [ftsTokens]);
-    if (!result || result.length === 0) return;
-
-    const recallData = {
-      trigger,
-      query,
-      timestamp: Date.now(),
-      results: result.map(r => ({
-        name: r.name,
-        observations: JSON.parse(r.observations_json || '[]').filter(Boolean).slice(0, 2),
-      })),
-    };
-
-    writeJSONFile(recallFile, recallData);
-  } catch (error) {
-    logError('proactive-recall-trigger', error);
-  }
-}
-
-// ============================================================================
-// Main PostToolUse Logic
-// ============================================================================
-
-async function postToolUse() {
-  try {
-    // Read stdin with timeout
-    const input = await readStdin(3000);
-
-    if (!input || input.trim() === '') {
-      process.exit(0);
-    }
-
-    // Parse and normalize tool data
-    const rawData = JSON.parse(input);
-    const toolData = normalizeToolData(rawData);
-
-    // Track code review invocations (for pre-commit enforcement)
-    if (isCodeReviewInvocation(toolData)) {
-      markCodeReviewDone();
-    }
-
-    // Detect plan file creation (beta)
-    detectPlanFile(toolData);
-
-    // Initialize pattern detector
-    const detector = new PatternDetector();
-
-    // Load recent tools from current session
-    const currentSession = readJSONFile(CURRENT_SESSION_FILE, { toolCalls: [] });
-    currentSession.toolCalls.slice(-10).forEach(tc => {
-      detector.addToolCall({
-        toolName: tc.toolName,
-        arguments: tc.arguments || {},
-      });
-    });
-
-    // Add current tool
-    detector.addToolCall(toolData);
-
-    // Detect patterns
-    const patterns = detector.detectPatterns(toolData);
-
-    // Update session context (for quota tracking) — returns sync data + async write promise
-    const { sessionContext, writePromise: contextWritePromise } = updateSessionContext(toolData);
-
-    // Detect anomalies
-    const anomalies = detectAnomalies(toolData, sessionContext);
-
-    // Trigger proactive recall on test failure or error
-    triggerProactiveRecall(toolData);
-
-    // Fire async writes in parallel
-    const pendingWrites = [contextWritePromise];
-
-    // Update recommendations incrementally
-    if (patterns.length > 0 || anomalies.length > 0) {
-      pendingWrites.push(updateRecommendations(patterns, anomalies));
-    }
-
-    // Update current session (async write)
-    const { session: updatedSession, writePromise: sessionWritePromise } =
-      updateCurrentSession(toolData, patterns, anomalies);
-    pendingWrites.push(sessionWritePromise);
-
-    // Check token threshold and save key points if needed
-    checkAndSaveKeyPoints(updatedSession, sessionContext);
-
-    // Wait for all async writes to complete before exit
-    await Promise.all(pendingWrites);
-
-    // Silent exit
-    process.exit(0);
-  } catch (error) {
-    logError('PostToolUse', error);
-    process.exit(0); // Never block Claude Code on hook errors
-  }
-}
-
-// ============================================================================
-// Execute
-// ============================================================================
-
-postToolUse();