agentsys 5.0.3 → 5.1.0

package/plugins/debate/lib/patterns/pipeline.js

@@ -0,0 +1,948 @@
+/**
+ * Slop Detection Pipeline
+ *
+ * 3-phase detection pipeline orchestrator:
+ * - Phase 1 (built-in): regex patterns + multi-pass analyzers - always runs
+ * - Phase 2 (optional): CLI tools (jscpd, madge, escomplex) - if available
+ * - Phase 3 (LLM handoff): certainty-tagged findings for agent review
+ *
+ * Inherits modes from deslop: report (analyze only) vs apply (fix issues)
+ *
+ * @module patterns/pipeline
+ * @author Avi Fenesh
+ * @license MIT
+ */
+
+const path = require('path');
+const fs = require('fs');
+const fsPromises = require('fs').promises;
+const slopPatterns = require('./slop-patterns');
+const analyzers = require('./slop-analyzers');
+
+// Concurrent file reads per batch
+const FILE_READ_BATCH_SIZE = 50;
+
+/**
+ * Global exclusions - files that should NEVER be flagged
+ * These are meta-files that define detection patterns, so they naturally
+ * contain the patterns they're detecting (not actual slop)
+ */
+const GLOBAL_EXCLUSIONS = [
+  '**/slop-patterns.js',
+  '**/slop-analyzers.js',
+  '**/cli-enhancers.js',
+  '**/pipeline.js',
+  '**/review-patterns.js',
+  '**/detect.js',
+  '**/*-patterns.js',  // Any pattern definition file
+  '**/*-analyzers.js'  // Any analyzer definition file
+];
+
+/**
+ * Certainty levels for findings
+ * HIGH: Single regex match - definitive
+ * MEDIUM: Multi-pass analysis - requires context
+ * LOW: Heuristic/CLI tool - needs verification
+ */
+const CERTAINTY = {
+  HIGH: 'HIGH',
+  MEDIUM: 'MEDIUM',
+  LOW: 'LOW'
+};
+
+/**
+ * Thoroughness levels
+ * quick: Phase 1 regex only - fastest
+ * normal: Phase 1 + multi-pass analyzers - balanced
+ * deep: Phase 1 + Phase 2 CLI tools (if available) - thorough
+ */
+const THOROUGHNESS = {
+  QUICK: 'quick',
+  NORMAL: 'normal',
+  DEEP: 'deep'
+};
+
+// Default timeout for pipeline execution (5 minutes)
+const DEFAULT_TIMEOUT_MS = 5 * 60 * 1000;
+
+/**
+ * Read multiple files asynchronously in batches
+ * @param {string[]} files - Array of file paths
+ * @param {number} batchSize - Concurrent reads per batch
+ * @returns {Promise<Map<string, {content: string|null, error: Error|null}>>}
+ */
+async function batchReadFiles(files, batchSize = FILE_READ_BATCH_SIZE) {
+  const results = new Map();
+
+  for (let i = 0; i < files.length; i += batchSize) {
+    const batch = files.slice(i, i + batchSize);
+    const batchResults = await Promise.all(
+      batch.map(async (file) => {
+        try {
+          const content = await fsPromises.readFile(file, 'utf8');
+          return { file, content, error: null };
+        } catch (err) {
+          return { file, content: null, error: err };
+        }
+      })
+    );
+
+    for (const result of batchResults) {
+      results.set(result.file, { content: result.content, error: result.error });
+    }
+  }
+
+  return results;
+}
+
+/**
+ * Run the slop detection pipeline
+ *
+ * @param {string} repoPath - Repository root path
+ * @param {Object} options - Pipeline options
+ * @param {string} [options.thoroughness='normal'] - quick | normal | deep
+ * @param {string[]} [options.targetFiles] - Specific files to analyze (defaults to all source files)
+ * @param {string} [options.language] - Filter to specific language
+ * @param {string} [options.mode='report'] - report | apply
+ * @param {Object} [options.cliTools] - Pre-detected CLI tools (from detectAvailableTools)
+ * @param {number} [options.timeout] - Timeout in ms (default: 5 minutes)
+ * @returns {Promise<Object>} Pipeline results: { findings, summary, phase3Prompt, missingTools }
+ */
+async function runPipeline(repoPath, options = {}) {
+  const thoroughness = options.thoroughness || THOROUGHNESS.NORMAL;
+  const mode = options.mode || 'report';
+  const language = options.language || null;
+  const timeout = options.timeout || DEFAULT_TIMEOUT_MS;
+  const startTime = Date.now();
+
+  const findings = [];
+  const missingTools = [];
+  let cliTools = options.cliTools || null;
+  let timedOut = false;
+
+  // Helper to check if timeout exceeded
+  function isTimedOut() {
+    return Date.now() - startTime > timeout;
+  }
+
+  // Get target files - limit to 200 to prevent memory exhaustion
+  // Users can pass targetFiles explicitly for larger scans
+  let targetFiles = options.targetFiles;
+  const maxFiles = options.maxFiles || 200;
+  if (!targetFiles || targetFiles.length === 0) {
+    const result = analyzers.countSourceFiles(repoPath, {
+      maxFiles,
+      includeTests: false
+    });
+    targetFiles = result.files;
+  }
+
+  // Pre-load all file contents asynchronously (used by Phase 1 and Phase 1b)
+  let fileContents;
+  try {
+    fileContents = await batchReadFiles(targetFiles);
+  } catch (err) {
+    console.error('[WARN] File pre-loading failed:', err.message);
+    fileContents = new Map();
+  }
+
+  // Phase 1: Built-in regex patterns (always runs)
+  // Wrapped in try-catch to prevent crashes on malformed files
+  try {
+    const phase1Results = runPhase1(repoPath, targetFiles, language, fileContents);
+    findings.push(...phase1Results);
+  } catch (err) {
+    // Log but continue with empty phase1 results rather than crash
+    console.error('[WARN] Phase 1 failed:', err.message);
+  }
+
+  // Check timeout after Phase 1
+  if (isTimedOut()) {
+    console.error('[WARN] Pipeline timeout after Phase 1');
+    timedOut = true;
+  }
+
+  // Phase 1b: Multi-pass analyzers (if normal or deep)
+  // Wrapped in try-catch as these are expensive and can fail
+  if (!timedOut && thoroughness !== THOROUGHNESS.QUICK) {
+    try {
+      // runMultiPassAnalyzers is async to support non-blocking I/O (PERF-007)
+      const multiPassResults = await runMultiPassAnalyzers(repoPath, targetFiles, fileContents);
+      findings.push(...multiPassResults);
+    } catch (err) {
+      // Log but continue with partial results rather than crash
+      console.error('[WARN] Phase 1b failed:', err.message);
+    }
+
+    // Check timeout after Phase 1b
+    if (isTimedOut()) {
+      console.error('[WARN] Pipeline timeout after Phase 1b');
+      timedOut = true;
+    }
+  }
+
+  // Phase 2: CLI tools (only if deep and tools available)
+  // Detect project languages for language-aware tool recommendations
+  let detectedLanguages = [];
+  if (!timedOut && thoroughness === THOROUGHNESS.DEEP) {
+    // Lazy-load CLI enhancers to avoid circular dependencies
+    const cliEnhancers = require('./cli-enhancers');
+
+    // Detect project languages
+    detectedLanguages = cliEnhancers.detectProjectLanguages(repoPath);
+
+    if (!cliTools) {
+      // Get tools relevant for detected languages
+      cliTools = cliEnhancers.detectAvailableTools(detectedLanguages);
+    }
+
+    // Track missing tools (only those relevant for project languages)
+    const relevantTools = cliEnhancers.getToolsForLanguages(detectedLanguages);
+    for (const toolName of Object.keys(relevantTools)) {
+      if (!cliTools[toolName]) {
+        missingTools.push(toolName);
+      }
+    }
+
+    const phase2Results = runPhase2(repoPath, cliTools, targetFiles);
+    findings.push(...phase2Results);
+  }
+
+  // Build summary
+  const summary = buildSummary(findings);
+
+  // Generate Phase 3 handoff prompt
+  const phase3Prompt = formatHandoffPrompt(findings, mode);
+
+  return {
+    findings,
+    summary,
+    phase3Prompt,
+    missingTools,
+    detectedLanguages,
+    metadata: {
+      repoPath,
+      thoroughness,
+      mode,
+      filesAnalyzed: targetFiles.length,
+      timestamp: new Date().toISOString(),
+      timedOut,
+      elapsedMs: Date.now() - startTime
+    }
+  };
+}
+
+/**
+ * Phase 1: Run built-in regex patterns against target files
+ *
+ * @param {string} repoPath - Repository root
+ * @param {string[]} targetFiles - Files to analyze
+ * @param {string|null} language - Optional language filter
+ * @param {Map<string, {content: string|null, error: Error|null}>} [fileContents] - Pre-loaded file contents (optional)
+ * @returns {Array} Findings with HIGH certainty
+ */
+function runPhase1(repoPath, targetFiles, language, fileContents) {
+  const findings = [];
+  const contentMap = fileContents || new Map();
+
+  // Get patterns (filtered by language if specified)
+  const patterns = language
+    ? slopPatterns.getPatternsForLanguage(language)
+    : slopPatterns.slopPatterns;
+
+  for (const file of targetFiles) {
+    // Skip globally excluded files (pattern definition files)
+    if (slopPatterns.isFileExcluded(file, GLOBAL_EXCLUSIONS)) {
+      continue;
+    }
+
+    // Detect file language once per file
+    const fileLanguage = analyzers.detectLanguage(file);
+
+    // Skip if language filter doesn't match file extension
+    if (language) {
+      // For JS/TS language filter, accept both 'javascript' and 'js' detection results
+      const isJsFamily = (language === 'javascript' || language === 'typescript') && fileLanguage === 'js';
+      if (fileLanguage !== language && !isJsFamily) continue;
+    }
+
+    // Get pre-loaded content from map, fallback to synchronous read
+    const filePath = path.isAbsolute(file) ? file : path.join(repoPath, file);
+    const readResult = contentMap.get(file) || contentMap.get(filePath);
+
+    let content;
+    if (readResult && !readResult.error && readResult.content !== null) {
+      content = readResult.content;
+    } else {
+      // Fallback to synchronous read (for backward compatibility with direct calls)
+      try {
+        content = fs.readFileSync(filePath, 'utf8');
+      } catch {
+        continue; // Skip unreadable files
+      }
+    }
+
+    const lines = content.split('\n');
+
+    for (const [patternName, pattern] of Object.entries(patterns)) {
+      // Skip multi-pass patterns (handled separately)
+      if (pattern.requiresMultiPass) continue;
+
+      // Skip if no regex pattern
+      if (!pattern.pattern) continue;
+
+      // Skip if file matches exclude patterns
+      if (slopPatterns.isFileExcluded(file, pattern.exclude)) continue;
+
+      // Skip language-specific patterns that don't match file's language
+      if (pattern.language) {
+        const patternLang = pattern.language;
+        // Map detection results to pattern language names
+        // Note: 'js' from detectLanguage covers both JavaScript and TypeScript
+        const langMatch = (patternLang === 'javascript' && fileLanguage === 'js') ||
+                          (patternLang === 'python' && fileLanguage === 'python') ||
+                          (patternLang === 'rust' && fileLanguage === 'rust') ||
+                          (patternLang === 'go' && fileLanguage === 'go') ||
+                          (patternLang === 'java' && fileLanguage === 'java') ||
+                          patternLang === fileLanguage;  // Direct match fallback
+        if (!langMatch) continue;
+      }
+
+      // Handle patterns requiring consecutive line blocks
+      if (pattern.minConsecutiveLines) {
+        const minLines = pattern.minConsecutiveLines;
+        let consecutiveStart = -1;
+        let consecutiveCount = 0;
+
+        for (let i = 0; i <= lines.length; i++) {
+          const line = i < lines.length ? lines[i] : ''; // Empty line at end to flush
+          const matches = i < lines.length && pattern.pattern.test(line);
+
+          if (matches) {
+            if (consecutiveStart === -1) {
+              consecutiveStart = i;
+            }
+            consecutiveCount++;
+          } else {
+            // End of consecutive block - check if it meets threshold
+            if (consecutiveCount >= minLines) {
+              findings.push({
+                file,
+                line: consecutiveStart + 1,
+                patternName,
+                severity: pattern.severity,
+                certainty: CERTAINTY.HIGH,
+                description: `${pattern.description} (${consecutiveCount} consecutive lines)`,
+                autoFix: pattern.autoFix,
+                content: `Lines ${consecutiveStart + 1}-${consecutiveStart + consecutiveCount}`,
+                phase: 1,
+                details: { startLine: consecutiveStart + 1, endLine: consecutiveStart + consecutiveCount, lineCount: consecutiveCount }
+              });
+            }
+            consecutiveStart = -1;
+            consecutiveCount = 0;
+          }
+        }
+      } else {
+        // Standard per-line matching
+        for (let i = 0; i < lines.length; i++) {
+          const line = lines[i];
+          if (pattern.pattern.test(line)) {
+            findings.push({
+              file,
+              line: i + 1,
+              patternName,
+              severity: pattern.severity,
+              certainty: CERTAINTY.HIGH,
+              description: pattern.description,
+              autoFix: pattern.autoFix,
+              content: line.trim().substring(0, 100),
+              phase: 1
+            });
+          }
+        }
+      }
+    }
+  }
+
+  return findings;
+}
+
+/**
+ * Run multi-pass analyzers (doc/code ratio, verbosity, etc.)
+ *
+ * @param {string} repoPath - Repository root
+ * @param {string[]} targetFiles - Files to analyze
+ * @param {Map<string, {content: string|null, error: Error|null}>} [fileContents] - Pre-loaded file contents (optional)
+ * @returns {Promise<Array>} Findings with MEDIUM certainty
+ */
+async function runMultiPassAnalyzers(repoPath, targetFiles, fileContents) {
+  const findings = [];
+  const contentMap = fileContents || new Map();
+
+  // Skip expensive analyzers for large file sets to prevent memory exhaustion
+  const isLargeRepo = targetFiles.length > 100;
+
+  // Get multi-pass pattern definitions for thresholds
+  const multiPassPatterns = slopPatterns.getMultiPassPatterns();
+
+  // Supported languages for doc/code and verbosity analysis
+  const docCodeLangs = /\.(js|jsx|ts|tsx|mjs|cjs|py|rs|java|go)$/i;
+
+  for (const file of targetFiles) {
+    if (!file.match(docCodeLangs)) continue;
+    if (analyzers.isTestFile(file)) continue;
+    // Skip globally excluded files (pattern definition files)
+    if (slopPatterns.isFileExcluded(file, GLOBAL_EXCLUSIONS)) continue;
+
+    // Get pre-loaded content from map, fallback to synchronous read
+    const filePath = path.isAbsolute(file) ? file : path.join(repoPath, file);
+    const readResult = contentMap.get(file) || contentMap.get(filePath);
+
+    let content;
+    if (readResult && !readResult.error && readResult.content !== null) {
+      content = readResult.content;
+    } else {
+      // Fallback to synchronous read (for backward compatibility with direct calls)
+      try {
+        content = fs.readFileSync(filePath, 'utf8');
+      } catch {
+        continue;
+      }
+    }
+
+    // Doc/code ratio analysis (multi-language)
+    const docCodePattern = multiPassPatterns.doc_code_ratio_js;
+    if (docCodePattern) {
+      const docRatioViolations = analyzers.analyzeDocCodeRatio(content, {
+        minFunctionLines: docCodePattern.minFunctionLines || 3,
+        maxRatio: docCodePattern.maxRatio || 3.0,
+        filePath: file
+      });
+
+      for (const v of docRatioViolations) {
+        findings.push({
+          file,
+          line: v.line,
+          patternName: 'doc_code_ratio',
+          severity: docCodePattern.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: `${docCodePattern.description} (${v.docLines} doc lines / ${v.codeLines} code lines = ${v.ratio}x)`,
+          autoFix: docCodePattern.autoFix,
+          content: v.functionName ? `${v.functionName}()` : `Function at line ${v.line}`,
+          phase: 1,
+          details: { docLines: v.docLines, codeLines: v.codeLines, ratio: v.ratio, functionName: v.functionName }
+        });
+      }
+    }
+
+    // Verbosity ratio analysis (multi-language)
+    const verbosityPattern = multiPassPatterns.verbosity_ratio;
+    if (verbosityPattern) {
+      const verbosityViolations = analyzers.analyzeVerbosityRatio(content, {
+        minCodeLines: verbosityPattern.minCodeLines || 3,
+        maxCommentRatio: verbosityPattern.maxCommentRatio || 2.0,
+        filePath: file
+      });
+
+      for (const v of verbosityViolations) {
+        findings.push({
+          file,
+          line: v.line,
+          patternName: 'verbosity_ratio',
+          severity: verbosityPattern.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: `${verbosityPattern.description} (${v.commentLines} comment lines / ${v.codeLines} code lines = ${v.ratio}x)`,
+          autoFix: verbosityPattern.autoFix,
+          content: `Function at line ${v.line}`,
+          phase: 1,
+          details: { commentLines: v.commentLines, codeLines: v.codeLines, ratio: v.ratio }
+        });
+      }
+    }
+  }
+
+  // Project-level analyzers (run once, not per-file)
+  // Wrap in try-catch to prevent crashes
+  const overEngPattern = multiPassPatterns.over_engineering_metrics;
+  if (overEngPattern && !isLargeRepo) {
+    try {
+      // analyzeOverEngineering is async to avoid blocking I/O (PERF-007)
+      const overEngResult = await analyzers.analyzeOverEngineering(repoPath, {
+        fileRatioThreshold: overEngPattern.fileRatioThreshold || 20,
+        linesPerExportThreshold: overEngPattern.linesPerExportThreshold || 500,
+        depthThreshold: overEngPattern.depthThreshold || 4
+      });
+
+      for (const v of overEngResult.violations) {
+        findings.push({
+          file: 'project-level',
+          line: 0,
+          patternName: 'over_engineering_metrics',
+          severity: v.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: `Over-engineering: ${v.type} - ${v.value} (threshold: ${v.threshold})`,
+          autoFix: 'flag',
+          content: v.value,
+          phase: 1,
+          details: v.details
+        });
+      }
+    } catch (err) {
+      // Skip on error - don't crash the pipeline
+    }
+  }
+
+  // Buzzword inflation analysis - EXPENSIVE: reads files multiple times
+  // Skip for large repos to prevent memory exhaustion
+  const buzzwordPattern = multiPassPatterns.buzzword_inflation;
+  if (buzzwordPattern && !isLargeRepo) {
+    try {
+      const buzzwordResult = analyzers.analyzeBuzzwordInflation(repoPath, {
+        minEvidenceMatches: buzzwordPattern.minEvidenceMatches || 2
+      });
+
+      for (const v of buzzwordResult.violations) {
+        findings.push({
+          file: v.file,
+          line: v.line,
+          patternName: 'buzzword_inflation',
+          severity: v.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: v.message,
+          autoFix: 'flag',
+          content: v.claim,
+          phase: 1,
+          details: { buzzword: v.buzzword, category: v.category, evidenceCount: v.evidenceCount }
+        });
+      }
+    } catch (err) {
+      // Skip on error - don't crash the pipeline
+    }
+  }
+
+  // Infrastructure without implementation - EXPENSIVE: reads all files twice
+  // Skip for large repos to prevent memory exhaustion
+  const infraPattern = multiPassPatterns.infrastructure_without_implementation;
+  if (infraPattern && !isLargeRepo) {
+    try {
+      const infraResult = analyzers.analyzeInfrastructureWithoutImplementation(repoPath);
+
+      for (const v of infraResult.violations) {
+        findings.push({
+          file: v.file,
+          line: v.line,
+          patternName: 'infrastructure_without_implementation',
+          severity: v.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: v.message,
+          autoFix: 'flag',
+          content: v.content,
+          phase: 1,
+          details: { varName: v.varName, type: v.type }
+        });
+      }
+    } catch (err) {
+      // Skip on error - don't crash the pipeline
+    }
+  }
+
+  // Dead code analysis (per-file)
+  const deadCodePattern = multiPassPatterns.dead_code;
+  if (deadCodePattern) {
+    for (const file of targetFiles) {
+      // Skip test files
+      if (analyzers.isTestFile(file)) continue;
+      // Skip globally excluded files (pattern definition files)
+      if (slopPatterns.isFileExcluded(file, GLOBAL_EXCLUSIONS)) continue;
+
+      // Get pre-loaded content from map, fallback to synchronous read
+      const filePath = path.isAbsolute(file) ? file : path.join(repoPath, file);
+      const readResult = contentMap.get(file) || contentMap.get(filePath);
+
+      let content;
+      if (readResult && !readResult.error && readResult.content !== null) {
+        content = readResult.content;
+      } else {
+        // Fallback to synchronous read (for backward compatibility with direct calls)
+        try {
+          content = fs.readFileSync(filePath, 'utf8');
+        } catch {
+          continue;
+        }
+      }
+
+      const deadCodeViolations = analyzers.analyzeDeadCode(content, { filePath: file });
+
+      for (const v of deadCodeViolations) {
+        findings.push({
+          file,
+          line: v.line,
+          patternName: 'dead_code',
+          severity: deadCodePattern.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: `${deadCodePattern.description}: ${v.terminationType} at line ${v.terminationLine}`,
+          autoFix: deadCodePattern.autoFix,
+          content: v.content,
+          phase: 1,
+          details: { terminationType: v.terminationType, terminationLine: v.terminationLine }
+        });
+      }
+    }
+  }
+
+  // Stub function analysis (per-file, multi-language)
+  const stubPattern = multiPassPatterns.placeholder_stub_returns_js;
+  if (stubPattern) {
+    // Supported extensions for stub detection
+    const stubExtensions = /\.(js|jsx|ts|tsx|mjs|cjs|py|rs|java|go)$/i;
+
+    for (const file of targetFiles) {
+      if (analyzers.isTestFile(file)) continue;
+      if (!file.match(stubExtensions)) continue;
+      // Honor pattern exclude globs (e.g., *.config.*)
+      if (slopPatterns.isFileExcluded(file, stubPattern.exclude)) continue;
+
+      // Get pre-loaded content from map, fallback to synchronous read
+      const filePath = path.isAbsolute(file) ? file : path.join(repoPath, file);
+      const readResult = contentMap.get(file) || contentMap.get(filePath);
+
+      let content;
+      if (readResult && !readResult.error && readResult.content !== null) {
+        content = readResult.content;
+      } else {
+        // Fallback to synchronous read (for backward compatibility with direct calls)
+        try {
+          content = fs.readFileSync(filePath, 'utf8');
+        } catch {
+          continue;
+        }
+      }
+
+      const stubViolations = analyzers.analyzeStubFunctions(content, { filePath: file });
+
+      for (const v of stubViolations) {
+        findings.push({
+          file,
+          line: v.line,
+          patternName: 'placeholder_stub_returns',
+          severity: v.hasTodo ? 'high' : stubPattern.severity,
+          certainty: v.certainty,
+          description: `${stubPattern.description}: ${v.functionName}() returns ${v.returnValue}`,
+          autoFix: stubPattern.autoFix,
+          content: v.content,
+          phase: 1,
+          details: { functionName: v.functionName, returnValue: v.returnValue, hasTodo: v.hasTodo }
+        });
+      }
+    }
+  }
+
+  // Shotgun surgery analysis (git history) - can have large buffer issues
+  // Skip for large repos to prevent memory exhaustion
+  const shotgunPattern = multiPassPatterns.shotgun_surgery;
+  if (shotgunPattern && !isLargeRepo) {
+    try {
+      const shotgunResult = analyzers.analyzeShotgunSurgery(repoPath, {
+        commitLimit: Math.min(shotgunPattern.commitLimit || 100, 50), // Reduce commit limit
+        clusterThreshold: shotgunPattern.clusterThreshold || 5
+      });
+
+      for (const v of shotgunResult.violations) {
+        findings.push({
+          file: 'project-level',
+          line: 0,
+          patternName: 'shotgun_surgery',
+          severity: shotgunPattern.severity,
+          certainty: CERTAINTY.MEDIUM,
+          description: `${shotgunPattern.description}: ${v.files.length} files change together ${v.count} times`,
+          autoFix: shotgunPattern.autoFix,
+          content: v.files.join(', ').substring(0, 100),
+          phase: 1,
+          details: { files: v.files, changeCount: v.count }
+        });
+      }
+    } catch {
+      // Git not available or not a git repo - skip silently
+    }
+  }
+
+  return findings;
+}
+
+/**
+ * Phase 2: Run CLI tools (if available)
+ *
+ * @param {string} repoPath - Repository root
+ * @param {Object} cliTools - Available CLI tools { jscpd, madge, escomplex }
+ * @param {string[]} targetFiles - Files to analyze
+ * @returns {Array} Findings with LOW certainty
+ */
+function runPhase2(repoPath, cliTools, targetFiles) {
+  const findings = [];
+  const cliEnhancers = require('./cli-enhancers');
+
+  // Duplicate detection with jscpd
+  if (cliTools.jscpd) {
+    const duplicates = cliEnhancers.runDuplicateDetection(repoPath);
+    if (duplicates) {
+      for (const dup of duplicates) {
+        findings.push({
+          file: dup.firstFile,
+          line: dup.firstLine,
+          patternName: 'code_duplication',
+          severity: 'medium',
+          certainty: CERTAINTY.LOW,
+          description: `Code duplication: ${dup.lines} lines duplicated in ${dup.secondFile}:${dup.secondLine}`,
+          autoFix: 'flag',
+          content: `${dup.lines} lines duplicated`,
+          phase: 2,
+          details: dup
+        });
+      }
+    }
+  }
+
+  // Circular dependencies with madge
+  if (cliTools.madge) {
+    const circularDeps = cliEnhancers.runDependencyAnalysis(repoPath);
+    if (circularDeps) {
+      for (const cycle of circularDeps) {
+        findings.push({
+          file: cycle[0],
+          line: 0,
+          patternName: 'circular_dependency',
+          severity: 'high',
+          certainty: CERTAINTY.LOW,
+          description: `Circular dependency: ${cycle.join(' -> ')}`,
+          autoFix: 'flag',
+          content: cycle.join(' -> '),
+          phase: 2,
+          details: { cycle }
+        });
+      }
+    }
+  }
+
+  // Complexity analysis with escomplex
+  if (cliTools.escomplex) {
+    const complexityResults = cliEnhancers.runComplexityAnalysis(repoPath, targetFiles);
+    if (complexityResults) {
+      for (const result of complexityResults) {
+        if (result.complexity > 10) { // High cyclomatic complexity threshold
+          findings.push({
+            file: result.file,
+            line: result.line || 0,
+            patternName: 'high_complexity',
+            severity: result.complexity > 20 ? 'high' : 'medium',
+            certainty: CERTAINTY.LOW,
+            description: `High cyclomatic complexity: ${result.complexity} in ${result.name}`,
+            autoFix: 'flag',
+            content: `${result.name}: complexity ${result.complexity}`,
+            phase: 2,
+            details: result
+          });
+        }
+      }
+    }
+  }
+
+  return findings;
+}
+
+/**
+ * Build summary statistics from findings
+ *
+ * @param {Array} findings - All findings
+ * @returns {Object} Summary statistics
+ */
+function buildSummary(findings) {
+  const summary = {
+    total: findings.length,
+    bySeverity: { critical: 0, high: 0, medium: 0, low: 0 },
+    byCertainty: { HIGH: 0, MEDIUM: 0, LOW: 0 },
+    byPhase: { 1: 0, 2: 0 },
+    byAutoFix: { remove: 0, replace: 0, add_logging: 0, flag: 0, none: 0 },
+    topPatterns: {}
+  };
+
+  for (const f of findings) {
+    summary.bySeverity[f.severity] = (summary.bySeverity[f.severity] || 0) + 1;
+    summary.byCertainty[f.certainty] = (summary.byCertainty[f.certainty] || 0) + 1;
+    summary.byPhase[f.phase] = (summary.byPhase[f.phase] || 0) + 1;
+    summary.byAutoFix[f.autoFix] = (summary.byAutoFix[f.autoFix] || 0) + 1;
+    summary.topPatterns[f.patternName] = (summary.topPatterns[f.patternName] || 0) + 1;
+  }
+
+  return summary;
+}
+
+/**
+ * Format handoff prompt for LLM (Phase 3)
+ *
+ * Creates a token-efficient prompt for the agent to review findings.
+ * Groups by certainty level with action guidance:
+ * - HIGH: Apply directly (if apply mode)
+ * - MEDIUM: Verify context before applying
+ * - LOW: Use judgment, may be false positive
+ *
+ * @param {Array} findings - All findings
+ * @param {string} mode - report | apply
+ * @param {Object} options - Formatting options
+ * @param {boolean} options.compact - Use compact table format (60-70% fewer tokens)
+ * @param {number} options.maxFindings - Maximum findings to include (default: 50)
+ * @returns {string} Formatted prompt
+ */
+function formatHandoffPrompt(findings, mode, options = {}) {
+  const { compact = false, maxFindings = 50 } = options;
+
+  if (findings.length === 0) {
+    return '## Slop Detection Results\n\nNo issues detected.';
+  }
+
+  // Use compact format if requested
+  if (compact) {
+    return formatCompactPrompt(findings, mode, maxFindings);
+  }
+
+  // Group findings by certainty
+  const byGroup = {
+    HIGH: findings.filter(f => f.certainty === CERTAINTY.HIGH),
+    MEDIUM: findings.filter(f => f.certainty === CERTAINTY.MEDIUM),
+    LOW: findings.filter(f => f.certainty === CERTAINTY.LOW)
+  };
+
+  let prompt = '## Slop Detection Results\n\n';
+  prompt += `Mode: **${mode}** | Total: ${findings.length} findings\n\n`;
+
+  // HIGH certainty - definitive matches
+  if (byGroup.HIGH.length > 0) {
+    prompt += '### HIGH Certainty (Definitive - trust these)\n\n';
+    if (mode === 'apply') {
+      prompt += '_Action: Apply fixes directly for autoFix patterns._\n\n';
+    }
+    prompt += formatFindingsList(byGroup.HIGH);
+    prompt += '\n';
+  }
+
+  // MEDIUM certainty - needs context verification
+  if (byGroup.MEDIUM.length > 0) {
+    prompt += '### MEDIUM Certainty (Verify context)\n\n';
+    prompt += '_Action: Review surrounding code before applying._\n\n';
+    prompt += formatFindingsList(byGroup.MEDIUM);
+    prompt += '\n';
+  }
+
+  // LOW certainty - use judgment
+  if (byGroup.LOW.length > 0) {
+    prompt += '### LOW Certainty (Use judgment)\n\n';
+    prompt += '_Action: May be false positives. Investigate before acting._\n\n';
+    prompt += formatFindingsList(byGroup.LOW);
+    prompt += '\n';
+  }
+
+  // Action summary
+  prompt += '### Action Summary\n\n';
+  const autoFixable = findings.filter(f => f.autoFix && f.autoFix !== 'flag' && f.autoFix !== 'none');
+  const needsReview = findings.filter(f => f.autoFix === 'flag' || f.autoFix === 'none');
+
+  prompt += `- Auto-fixable: ${autoFixable.length}\n`;
+  prompt += `- Needs manual review: ${needsReview.length}\n`;
+
+  return prompt;
+}
+
+/**
+ * Format findings in compact table format for token efficiency
+ *
+ * Reduces token usage by ~60-70% compared to verbose format.
+ * Best for large finding sets where full descriptions aren't needed.
+ *
+ * @param {Array} findings - All findings
+ * @param {string} mode - report | apply
+ * @param {number} maxFindings - Maximum findings to include
+ * @returns {string} Compact formatted prompt
+ */
+function formatCompactPrompt(findings, mode, maxFindings) {
+  // Single pass to count certainty levels and auto-fixable findings
+  const { highCount, mediumCount, lowCount, autoFixableCount } = findings.reduce((acc, f) => {
+    switch (f.certainty) {
+      case CERTAINTY.HIGH: acc.highCount++; break;
+      case CERTAINTY.MEDIUM: acc.mediumCount++; break;
+      case CERTAINTY.LOW: acc.lowCount++; break;
+    }
+    if (f.autoFix && f.autoFix !== 'flag' && f.autoFix !== 'none') {
+      acc.autoFixableCount++;
+    }
+    return acc;
+  }, { highCount: 0, mediumCount: 0, lowCount: 0, autoFixableCount: 0 });
+
+  // Truncate if needed
+  const limited = findings.slice(0, maxFindings);
+  const truncated = findings.length > maxFindings;
+
+  // Summary header
+  let output = `## Slop: ${mode}|H:${highCount}|M:${mediumCount}|L:${lowCount}\n\n`;
+
+  // Table format
+  output += '|File|L|Pattern|Cert|Fix|\n';
+  output += '|---|---|---|---|---|\n';
+
+  for (const f of limited) {
+    const fix = f.autoFix && f.autoFix !== 'flag' && f.autoFix !== 'none' ? f.autoFix : '-';
+    const cert = f.certainty.charAt(0); // H, M, or L
+    output += `|${f.file}|${f.line}|${f.patternName}|${cert}|${fix}|\n`;
+  }
+
+  if (truncated) {
+    output += `\n_+${findings.length - maxFindings} more findings (truncated)_\n`;
+  }
+
+  // Auto-fix summary
+  output += `\n**Auto-fixable: ${autoFixableCount}** | Manual: ${findings.length - autoFixableCount}`;
+
+  return output;
+}
+
+/**
+ * Format a list of findings for the prompt
+ *
+ * @param {Array} findings - Findings to format
+ * @returns {string} Formatted list
+ */
+function formatFindingsList(findings) {
+  // Group by file for compact output
+  const byFile = {};
+  for (const f of findings) {
+    if (!byFile[f.file]) byFile[f.file] = [];
+    byFile[f.file].push(f);
+  }
+
+  let output = '';
+  for (const [file, fileFindings] of Object.entries(byFile)) {
+    output += `**${file}**\n`;
+    for (const f of fileFindings) {
+      const fixTag = f.autoFix && f.autoFix !== 'flag' && f.autoFix !== 'none'
+        ? ` [${f.autoFix}]`
+        : '';
+      output += `- L${f.line}: ${f.description}${fixTag}\n`;
+    }
+    output += '\n';
+  }
+
+  return output;
+}
+
+module.exports = {
+  runPipeline,
+  // Exported for testing
+  runPhase1,
+  runMultiPassAnalyzers,
+  runPhase2,
+  buildSummary,
+  formatHandoffPrompt,
+  formatCompactPrompt,
+  // Constants
+  CERTAINTY,
+  THOROUGHNESS
+};