agentsys 5.0.3 → 5.1.0

package/plugins/debate/lib/enhance/auto-suppression.js

@@ -0,0 +1,622 @@
+/**
+ * Auto-Learning Suppression System for /enhance
+ *
+ * Automatically detects obvious false positives and stores them
+ * for future runs, making the enhance tool smarter over time.
+ *
+ * Key features:
+ * - Pattern-specific heuristics for common false positives
+ * - Cross-platform storage using getSuppressionPath()
+ * - 0.90+ confidence threshold for auto-suppression
+ * - Backward compatible with existing suppression.js
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { execFileSync } = require('child_process');
+
+// Try to import cross-platform helpers
+let getSuppressionPath;
+try {
+  const crossPlatform = require('../cross-platform');
+  getSuppressionPath = crossPlatform.getSuppressionPath;
+} catch {
+  // Fallback for when running from plugin directory
+  const os = require('os');
+  getSuppressionPath = () => path.join(os.homedir(), '.claude', 'enhance', 'suppressions.json');
+}
+
+/**
+ * Minimum confidence threshold for auto-suppression
+ */
+const CONFIDENCE_THRESHOLD = 0.90;
+
+/**
+ * Maximum suppressions per project (prevents file bloat)
+ */
+const MAX_SUPPRESSIONS_PER_PROJECT = 100;
+
+/**
+ * Suppression expiry in milliseconds (6 months)
+ */
+const SUPPRESSION_EXPIRY_MS = 6 * 30 * 24 * 60 * 60 * 1000;
+
+/**
+ * Pattern-specific heuristics for detecting false positives
+ * Each heuristic returns { reason, confidence } or null
+ */
+const PATTERN_HEURISTICS = {
+  /**
+   * vague_instructions: Detects when vague terms appear in pattern documentation
+   * False positive when content describes the pattern itself
+   */
+  vague_instructions: (finding, content, context) => {
+    const contentLower = content.toLowerCase();
+
+    // Check if file is pattern documentation describing vague language detection
+    const isPatternDoc =
+      /pattern.*detect.*usually|example.*vague|fuzzy.*language.*like/i.test(content) ||
+      /vague.*terms.*like|"usually".*"sometimes"/i.test(content);
+
+    if (isPatternDoc) {
+      return {
+        reason: 'Pattern documentation self-reference (describes vague language detection)',
+        confidence: 0.98
+      };
+    }
+
+    // Check if in table describing patterns
+    const line = finding.line || 0;
+    const lines = content.split('\n');
+    const surroundingLines = lines.slice(Math.max(0, line - 5), line + 5).join('\n');
+    if (/\|.*vague.*\||\|.*usually.*sometimes.*\|/i.test(surroundingLines)) {
+      return {
+        reason: 'Pattern table documentation',
+        confidence: 0.95
+      };
+    }
+
+    return null;
+  },
+
+  /**
+   * aggressive_emphasis: Detects legitimate workflow enforcement usage
+   * False positive in workflow gates and critical agent constraints
+   */
+  aggressive_emphasis: (finding, content, context) => {
+    const line = finding.line || 0;
+    const lines = content.split('\n');
+    const surroundingLines = lines.slice(Math.max(0, line - 20), line + 20).join('\n');
+
+    // Check for workflow gate context
+    const isWorkflowGate =
+      /WORKFLOW\s+GATES?/i.test(surroundingLines) ||
+      /\[CRITICAL\]\s*NO\s+AGENT\s+may/i.test(surroundingLines) ||
+      /MUST\s+NOT\s+DO|NEVER\s+skip|DO\s+NOT\s+proceed/i.test(surroundingLines) ||
+      /SubagentStop\s+hook|phase\s+9\s+review/i.test(surroundingLines);
+
+    if (isWorkflowGate) {
+      return {
+        reason: 'Workflow enforcement requires emphasis for gates',
+        confidence: 0.95
+      };
+    }
+
+    // Check for critical rules section
+    const isCriticalRules =
+      /critical-rules|Critical\s+Rules.*Priority/i.test(surroundingLines) ||
+      /<critical-rules>/i.test(surroundingLines);
+
+    if (isCriticalRules) {
+      return {
+        reason: 'Critical rules section requires emphasis',
+        confidence: 0.93
+      };
+    }
+
+    return null;
+  },
+
+  /**
+   * missing_examples: Detects orchestrator/workflow files that delegate to subagents
+   * False positive when file is orchestrator that spawns other agents
+   */
+  missing_examples: (finding, content, context) => {
+    const filePath = finding.file || context?.file || '';
+    const fileNameLower = path.basename(filePath).toLowerCase();
+
+    // Check if file is an orchestrator
+    const isOrchestrator =
+      fileNameLower.includes('orchestrator') ||
+      fileNameLower.includes('coordinator') ||
+      /Task\s*\(\s*\{[\s\S]*subagent_type/i.test(content);
+
+    if (isOrchestrator) {
+      return {
+        reason: 'Orchestrator file delegates to subagents (examples in subagents)',
+        confidence: 0.92
+      };
+    }
+
+    // Check if workflow command that invokes agents
+    const isWorkflowCommand =
+      /spawn.*agent|invoke.*agent|Task\s*\(\s*\{/i.test(content) &&
+      fileNameLower.endsWith('.md');
+
+    if (isWorkflowCommand) {
+      return {
+        reason: 'Workflow command invokes agents with examples',
+        confidence: 0.90
+      };
+    }
+
+    return null;
+  },
+
+  /**
+   * missing_output_format: Detects files that spawn subagents with output specs
+   * False positive when subagent is responsible for output format
+   */
+  missing_output_format: (finding, content, context) => {
+    // Check if content spawns subagents with their own output specs
+    const spawnsSubagent =
+      /subagent_type|spawn.*agent|Task\s*\(\s*\{/i.test(content) ||
+      /enhance:.*-enhancer|enhance:.*-reporter/i.test(content);
+
+    if (spawnsSubagent) {
+      return {
+        reason: 'Delegates output to subagent (subagent defines format)',
+        confidence: 0.91
+      };
+    }
+
+    return null;
+  },
+
+  /**
+   * missing_constraints: Detects files that already have constraint sections
+   * False positive when "## What Agent MUST NOT Do" or similar exists
+   */
+  missing_constraints: (finding, content, context) => {
+    // Check for constraint section presence
+    const hasConstraintSection =
+      /##\s*What\s+.*MUST\s+NOT\s+Do/i.test(content) ||
+      /##\s*Constraints/i.test(content) ||
+      /<constraints>/i.test(content) ||
+      /##\s*Critical\s+Constraints/i.test(content) ||
+      /WORKFLOW\s+GATES/i.test(content);
+
+    if (hasConstraintSection) {
+      return {
+        reason: 'File has constraint section (different heading format)',
+        confidence: 0.94
+      };
+    }
+
+    return null;
+  },
+
+  /**
+   * redundant_cot: Detects legitimate step-by-step for complex workflows
+   * False positive in multi-phase workflow prompts
+   */
+  redundant_cot: (finding, content, context) => {
+    // Check if multi-phase workflow
+    const isMultiPhase =
+      /Phase\s+\d+:|Step\s+\d+:|###\s+Phase/i.test(content) &&
+      /Phase\s+[2-9]:|Step\s+[2-9]:/i.test(content);
+
+    if (isMultiPhase) {
+      return {
+        reason: 'Multi-phase workflow requires step guidance',
+        confidence: 0.91
+      };
+    }
+
+    return null;
+  }
+};
+
+/**
+ * Check if a finding is likely a false positive
+ *
+ * @param {Object} finding - The finding to check
+ * @param {string} content - File content
+ * @param {Object} context - Additional context { file, analyzer, suppressions }
+ * @returns {{ reason: string, confidence: number } | null}
+ */
+function isLikelyFalsePositive(finding, content, context = {}) {
+  const patternId = (finding.patternId || finding.id || '').toLowerCase();
+
+  // Skip if no content
+  if (!content || typeof content !== 'string') {
+    return null;
+  }
+
+  // Check pattern-specific heuristic
+  const heuristic = PATTERN_HEURISTICS[patternId];
+  if (heuristic) {
+    const result = heuristic(finding, content, context);
+    if (result && result.confidence >= CONFIDENCE_THRESHOLD) {
+      return result;
+    }
+  }
+
+  // Generic self-reference detection: pattern names in pattern docs
+  if (finding.file && isPatternDocumentation(finding.file, content, patternId)) {
+    return {
+      reason: 'Pattern self-reference in documentation',
+      confidence: 0.96
+    };
+  }
+
+  return null;
+}
+
+/**
+ * Check if file is pattern documentation that mentions the pattern
+ *
+ * @param {string} filePath - File path
+ * @param {string} content - File content
+ * @param {string} patternId - Pattern being checked
+ * @returns {boolean}
+ */
+function isPatternDocumentation(filePath, content, patternId) {
+  const fileName = path.basename(filePath).toLowerCase();
+
+  // Pattern documentation file names
+  const isPatternFile =
+    fileName.includes('pattern') ||
+    fileName.includes('enhance.md') ||
+    fileName.includes('enhancer');
+
+  if (!isPatternFile) return false;
+
+  // Check if content documents patterns in table format
+  const patternIdReadable = patternId.replace(/_/g, ' ');
+  const describesPattern =
+    new RegExp(`\\|[^|]*${patternId}[^|]*\\|`, 'i').test(content) ||
+    new RegExp(`\\|[^|]*${patternIdReadable}[^|]*\\|`, 'i').test(content);
+
+  return describesPattern;
+}
+
+/**
+ * Get project identifier from git remote or directory hash
+ *
+ * @param {string} projectRoot - Project root directory
+ * @returns {string} Project identifier
+ */
+function getProjectId(projectRoot = process.cwd()) {
+  try {
+    // Try to get git remote
+    const remote = execFileSync('git', ['remote', 'get-url', 'origin'], {
+      cwd: projectRoot,
+      encoding: 'utf8',
+      stdio: ['pipe', 'pipe', 'pipe']
+    }).trim();
+
+    if (remote) {
+      // Normalize remote URL to identifier
+      // https://github.com/user/repo.git -> github.com/user/repo
+      // git@github.com:user/repo.git -> github.com/user/repo
+      return remote
+        .replace(/^https?:\/\//, '')
+        .replace(/^git@/, '')
+        .replace(/\.git$/, '')
+        .replace(':', '/');
+    }
+  } catch {
+    // Git not available or not a git repo
+  }
+
+  // Fallback to directory name hash
+  const absPath = path.resolve(projectRoot);
+  return `local:${path.basename(absPath)}`;
+}
+
+/**
+ * Load auto-learned suppressions for a project
+ *
+ * @param {string} suppressionPath - Path to suppressions.json
+ * @param {string} projectId - Project identifier
+ * @returns {Object} Auto-learned suppressions { patterns: {}, stats: {} }
+ */
+function loadAutoSuppressions(suppressionPath, projectId) {
+  const defaultResult = { patterns: {}, stats: { totalSuppressed: 0 } };
+
+  try {
+    if (!fs.existsSync(suppressionPath)) {
+      return defaultResult;
+    }
+
+    const data = JSON.parse(fs.readFileSync(suppressionPath, 'utf8'));
+    const projectData = data.projects?.[projectId];
+
+    if (!projectData?.auto_learned) {
+      return defaultResult;
+    }
+
+    // Prune expired suppressions
+    const autoLearned = projectData.auto_learned;
+    const now = Date.now();
+    const prunedPatterns = {};
+
+    for (const [patternId, suppression] of Object.entries(autoLearned.patterns || {})) {
+      const learnedAt = new Date(suppression.learnedAt).getTime();
+      if (now - learnedAt < SUPPRESSION_EXPIRY_MS) {
+        prunedPatterns[patternId] = suppression;
+      }
+    }
+
+    return {
+      patterns: prunedPatterns,
+      stats: autoLearned.stats || { totalSuppressed: 0 }
+    };
+  } catch {
+    return defaultResult;
+  }
+}
+
+/**
+ * Save auto-learned suppressions
+ *
+ * @param {string} suppressionPath - Path to suppressions.json
+ * @param {string} projectId - Project identifier
+ * @param {Array} findings - Findings to save as suppressions
+ */
+function saveAutoSuppressions(suppressionPath, projectId, findings) {
+  if (!findings || findings.length === 0) return;
+
+  // Ensure directory exists
+  const dir = path.dirname(suppressionPath);
+  if (!fs.existsSync(dir)) {
+    fs.mkdirSync(dir, { recursive: true });
+  }
+
+  // Load existing data
+  let data = { version: '2.0', projects: {} };
+  try {
+    if (fs.existsSync(suppressionPath)) {
+      data = JSON.parse(fs.readFileSync(suppressionPath, 'utf8'));
+    }
+  } catch {
+    // Start fresh on error
+  }
+
+  // Ensure structure
+  if (!data.projects) data.projects = {};
+  if (!data.projects[projectId]) data.projects[projectId] = {};
+  if (!data.projects[projectId].auto_learned) {
+    data.projects[projectId].auto_learned = {
+      patterns: {},
+      stats: { totalSuppressed: 0, lastAnalysis: null }
+    };
+  }
+
+  const autoLearned = data.projects[projectId].auto_learned;
+  const now = new Date().toISOString();
+
+  // Group findings by pattern
+  const byPattern = {};
+  for (const finding of findings) {
+    const patternId = (finding.patternId || finding.id || '').toLowerCase();
+    if (!patternId) continue;
+
+    if (!byPattern[patternId]) {
+      byPattern[patternId] = {
+        files: [],
+        reason: finding.suppressionReason || 'Auto-detected false positive',
+        confidence: finding.confidence || CONFIDENCE_THRESHOLD,
+        learnedAt: now,
+        occurrences: 0
+      };
+    }
+
+    if (finding.file && !byPattern[patternId].files.includes(finding.file)) {
+      byPattern[patternId].files.push(finding.file);
+    }
+    byPattern[patternId].occurrences++;
+
+    // Keep highest confidence
+    if (finding.confidence > byPattern[patternId].confidence) {
+      byPattern[patternId].confidence = finding.confidence;
+      byPattern[patternId].reason = finding.suppressionReason;
+    }
+  }
+
+  // Merge with existing patterns
+  for (const [patternId, newSuppression] of Object.entries(byPattern)) {
+    const existing = autoLearned.patterns[patternId];
+    if (existing) {
+      // Merge files
+      const allFiles = [...new Set([...existing.files, ...newSuppression.files])];
+      existing.files = allFiles.slice(0, 50); // Cap files per pattern
+      existing.occurrences = (existing.occurrences || 0) + newSuppression.occurrences;
+      existing.lastSeen = now;
+      // Update confidence if higher
+      if (newSuppression.confidence > existing.confidence) {
+        existing.confidence = newSuppression.confidence;
+        existing.reason = newSuppression.reason;
+      }
+    } else {
+      autoLearned.patterns[patternId] = newSuppression;
+    }
+  }
+
+  // Enforce max suppressions per project
+  const patternIds = Object.keys(autoLearned.patterns);
+  if (patternIds.length > MAX_SUPPRESSIONS_PER_PROJECT) {
+    // Remove oldest patterns
+    const sorted = patternIds.sort((a, b) => {
+      const aDate = new Date(autoLearned.patterns[a].learnedAt);
+      const bDate = new Date(autoLearned.patterns[b].learnedAt);
+      return aDate - bDate;
+    });
+
+    const toRemove = sorted.slice(0, patternIds.length - MAX_SUPPRESSIONS_PER_PROJECT);
+    for (const id of toRemove) {
+      delete autoLearned.patterns[id];
+    }
+  }
+
+  // Update stats
+  autoLearned.stats.totalSuppressed = Object.keys(autoLearned.patterns).length;
+  autoLearned.stats.lastAnalysis = now;
+
+  // Write file
+  fs.writeFileSync(suppressionPath, JSON.stringify(data, null, 2));
+}
+
+/**
+ * Clear auto-learned suppressions for a project
+ *
+ * @param {string} suppressionPath - Path to suppressions.json
+ * @param {string} projectId - Project identifier
+ */
+function clearAutoSuppressions(suppressionPath, projectId) {
+  try {
+    if (!fs.existsSync(suppressionPath)) return;
+
+    const data = JSON.parse(fs.readFileSync(suppressionPath, 'utf8'));
+
+    if (data.projects?.[projectId]?.auto_learned) {
+      data.projects[projectId].auto_learned = {
+        patterns: {},
+        stats: { totalSuppressed: 0, lastAnalysis: new Date().toISOString() }
+      };
+      fs.writeFileSync(suppressionPath, JSON.stringify(data, null, 2));
+    }
+  } catch {
+    // Ignore errors
+  }
+}
+
+/**
+ * Merge auto-learned suppressions with manual suppressions
+ *
+ * @param {Object} autoLearned - Auto-learned suppressions
+ * @param {Object} manual - Manual suppressions (from config)
+ * @returns {Object} Merged suppressions
+ */
+function mergeSuppressions(autoLearned, manual) {
+  return {
+    ignore: {
+      patterns: [...(manual.ignore?.patterns || [])],
+      files: [...(manual.ignore?.files || [])],
+      rules: { ...(manual.ignore?.rules || {}) }
+    },
+    severity: { ...(manual.severity || {}) },
+    auto_learned: autoLearned
+  };
+}
+
+/**
+ * Export learned suppressions for team sharing
+ *
+ * @param {string} suppressionPath - Path to suppressions.json
+ * @param {string} projectId - Project identifier
+ * @returns {Object} Exportable suppression data
+ */
+function exportAutoSuppressions(suppressionPath, projectId) {
+  const autoLearned = loadAutoSuppressions(suppressionPath, projectId);
+
+  return {
+    exportedAt: new Date().toISOString(),
+    projectId,
+    suppressions: autoLearned.patterns,
+    stats: autoLearned.stats
+  };
+}
+
+/**
+ * Import shared suppressions
+ *
+ * @param {string} suppressionPath - Path to suppressions.json
+ * @param {string} projectId - Project identifier
+ * @param {Object} importData - Data from exportAutoSuppressions
+ */
+function importAutoSuppressions(suppressionPath, projectId, importData) {
+  if (!importData?.suppressions) return;
+
+  // Convert imported suppressions to findings format
+  const findings = [];
+  for (const [patternId, suppression] of Object.entries(importData.suppressions)) {
+    for (const file of suppression.files || []) {
+      findings.push({
+        patternId,
+        file,
+        suppressionReason: suppression.reason,
+        confidence: suppression.confidence
+      });
+    }
+  }
+
+  saveAutoSuppressions(suppressionPath, projectId, findings);
+}
+
+/**
+ * Analyze findings for potential auto-suppression
+ * Returns findings that should be auto-suppressed
+ *
+ * @param {Array} findings - All findings
+ * @param {Map<string, string>} fileContents - Map of file path to content
+ * @param {Object} options - Options { noLearn, projectRoot }
+ * @returns {Array} Findings to be auto-suppressed
+ */
+function analyzeForAutoSuppression(findings, fileContents, options = {}) {
+  if (options.noLearn) return [];
+
+  const toSuppress = [];
+
+  for (const finding of findings) {
+    const filePath = finding.file || finding.filePath;
+    const content = fileContents.get(filePath);
+
+    if (!content) continue;
+
+    const fpCheck = isLikelyFalsePositive(finding, content, {
+      file: filePath,
+      projectRoot: options.projectRoot
+    });
+
+    if (fpCheck) {
+      toSuppress.push({
+        ...finding,
+        suppressed: true,
+        suppressionReason: fpCheck.reason,
+        confidence: fpCheck.confidence
+      });
+    }
+  }
+
+  return toSuppress;
+}
+
+module.exports = {
+  // Constants
+  CONFIDENCE_THRESHOLD,
+  MAX_SUPPRESSIONS_PER_PROJECT,
+  SUPPRESSION_EXPIRY_MS,
+
+  // Core functions
+  isLikelyFalsePositive,
+  getProjectId,
+
+  // Storage functions
+  loadAutoSuppressions,
+  saveAutoSuppressions,
+  clearAutoSuppressions,
+  mergeSuppressions,
+
+  // Import/export
+  exportAutoSuppressions,
+  importAutoSuppressions,
+
+  // Analysis helper
+  analyzeForAutoSuppression,
+
+  // For testing
+  PATTERN_HEURISTICS,
+  isPatternDocumentation
+};