tribunal-kit 9.2.6 → 9.2.7

package/src/system1/evidence_engine.js

@@ -0,0 +1,602 @@
+'use strict';
+
+/**
+ * evidence_engine.js — System-1 Capability 1: Evidence Intelligence
+ * =================================================================
+ * Deterministic, explainable, bounded evidence ranking layer.
+ * Determines what information is actually relevant to a decision.
+ *
+ * Evidence Hierarchy:
+ * 1. Changed files (directly modified)
+ * 2. Changed symbols/functions (extracted from diff/AST)
+ * 3. Direct imports/dependencies
+ * 4. Relevant tests (covering changed behavior)
+ * 5. Relevant configuration (controlling behavior)
+ * 6. Related security boundaries (auth, crypto, permissions)
+ * 7. Relevant historical memory (procedural & semantic)
+ * 8. Broader repository context (siblings, public entrypoints)
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { normalizePath, estimateTokens } = require('../context/ranker');
+const { HIGH_RISK_PATTERNS, HIGH_RISK_EXTENSIONS } = require('./constants');
+
+// Security boundary patterns
+const SECURITY_BOUNDARIES = [
+  /[\\/](?:auth|security|jwt|crypto|secrets?|tokens?|passwords?|sessions?)[\\/.]/i,
+  /[\\/](?:roles?|access-control|rbac|oauth|sso|permissions?)[\\/.]/i,
+  /[\\/]SECURITY\.md$/i,
+  /[\\/]\.env/i,
+];
+
+// Type hierarchy priority for deterministic ordering
+const TYPE_PRIORITY = {
+  changed_file: 1,
+  changed_symbol: 2,
+  direct_dependency: 3,
+  direct_importer: 3,
+  relevant_test: 4,
+  configuration: 5,
+  security_boundary: 6,
+  historical_memory: 7,
+  previous_finding: 8,
+  repository_context: 9,
+};
+
+/**
+ * Redact sensitive credentials and tokens from evidence snippets
+ */
+function redactSecrets(text) {
+  if (!text) return '';
+  return text
+    .replace(
+      /(?:-----BEGIN [A-Z ]+KEY-----[\s\S]*?-----END [A-Z ]+KEY-----)/g,
+      '[REDACTED_PRIVATE_KEY]',
+    )
+    .replace(/(eyJ[A-Za-z0-9-_=]+\.[A-Za-z0-9-_=]+\.[A-Za-z0-9-_=]+)/g, '[REDACTED_JWT_TOKEN]')
+    .replace(/(Bearer\s+)[A-Za-z0-9-_.~+/]+=*/gi, '$1[REDACTED_BEARER_TOKEN]')
+    .replace(
+      /((?:api[_-]?key|secret|password|passwd|token|auth[_-]?token|access[_-]?token|credentials?|private[_-]?key)\s*[:=]\s*['"]?)[^'"\s\n,;]+/gi,
+      '$1[REDACTED_SECRET]',
+    );
+}
+
+/**
+ * Extract symbols/functions modified in diff or file
+ * @param {string} diff
+ * @param {string} content
+ * @returns {Array<{ name: string, type: string, line?: number }>}
+ */
+function extractChangedSymbols(diff = '', content = '') {
+  const symbols = [];
+  const symbolNames = new Set();
+  const MAX_DIFF_LINES_TO_SCAN = 1000;
+  const MAX_SYMBOLS_PER_FILE = 16;
+
+  const functionRegex =
+    /(?:function\s+([a-zA-Z0-9_$]+)|(?:const|let|var)\s+([a-zA-Z0-9_$]+)\s*=\s*(?:async\s*)?\([^)]*\)\s*=>|(?:async\s+)?([a-zA-Z0-9_$]+)\s*\([^)]*\)\s*\{|class\s+([a-zA-Z0-9_$]+))/g;
+
+  // 1. Extract from added diff lines (bounded scan)
+  if (diff) {
+    const lines = diff.split('\n');
+    const scanLimit = Math.min(lines.length, MAX_DIFF_LINES_TO_SCAN);
+    for (let idx = 0; idx < scanLimit; idx++) {
+      if (symbols.length >= MAX_SYMBOLS_PER_FILE) break;
+      const line = lines[idx];
+      if (line.startsWith('+') && !line.startsWith('+++')) {
+        let match;
+        while ((match = functionRegex.exec(line)) !== null) {
+          const sym = match[1] || match[2] || match[3] || match[4];
+          if (sym && !symbolNames.has(sym) && sym.length > 1) {
+            symbolNames.add(sym);
+            symbols.push({
+              name: sym,
+              type: match[0].includes('class') ? 'class' : 'function',
+              diff_line: idx + 1,
+            });
+            if (symbols.length >= MAX_SYMBOLS_PER_FILE) break;
+          }
+        }
+      }
+    }
+  }
+
+  // 2. If content is available and symbols are scarce, extract top definitions
+  if (symbols.length === 0 && content) {
+    const scanContent = content.length > 50000 ? content.slice(0, 50000) : content;
+    let match;
+    while ((match = functionRegex.exec(scanContent)) !== null) {
+      const sym = match[1] || match[2] || match[3] || match[4];
+      if (sym && !symbolNames.has(sym) && sym.length > 1) {
+        symbolNames.add(sym);
+        symbols.push({ name: sym, type: match[0].includes('class') ? 'class' : 'function' });
+        if (symbols.length >= 8) break;
+      }
+    }
+  }
+
+  return symbols;
+}
+
+/**
+ * Extract bounded snippet from file content around specific line or symbol
+ */
+function extractBoundedSnippet(content, targetLine = null, targetSymbol = null, maxLines = 10) {
+  if (!content) return '';
+  const MAX_SNIPPET_CHARS = 2000;
+  const lines = content.split('\n');
+  let snippet = content;
+  if (lines.length > maxLines) {
+    let center = 0;
+    if (targetLine !== null && targetLine > 0 && targetLine <= lines.length) {
+      center = targetLine - 1;
+    } else if (targetSymbol) {
+      const foundIdx = lines.findIndex(l => l.includes(targetSymbol));
+      if (foundIdx !== -1) center = foundIdx;
+    }
+
+    const start = Math.max(0, center - Math.floor(maxLines / 2));
+    const end = Math.min(lines.length, start + maxLines);
+    snippet = lines.slice(start, end).join('\n');
+  }
+  if (snippet.length > MAX_SNIPPET_CHARS) {
+    snippet = snippet.slice(0, MAX_SNIPPET_CHARS) + '\n... [TRUNCATED]';
+  }
+  return redactSecrets(snippet);
+}
+
+/**
+ * Score and collect evidence items
+ * @param {Object} options
+ * @param {string} [options.task]
+ * @param {string} [options.diff]
+ * @param {string[]|string} [options.files]
+ * @param {string} [options.repoRoot]
+ * @param {Array} [options.memory]
+ * @param {Array} [options.previousFindings]
+ * @param {number} [options.maxItems]
+ * @param {number} [options.maxTokens]
+ * @returns {Object} Evidence Intelligence Result
+ */
+function collectAndRankEvidence(options = {}) {
+  const repoRoot = options.repoRoot ? path.resolve(options.repoRoot) : process.cwd();
+  const rawFiles = options.files;
+  const changedFiles = (
+    Array.isArray(rawFiles) ? rawFiles : typeof rawFiles === 'string' ? rawFiles.split(',') : []
+  )
+    .map(f => (typeof f === 'string' ? f.trim() : ''))
+    .filter(Boolean)
+    .map(normalizePath);
+
+  const task = typeof options.task === 'string' ? options.task : '';
+  const diff = typeof options.diff === 'string' ? options.diff : '';
+  const memoryItems = Array.isArray(options.memory) ? options.memory : [];
+  const prevFindings = Array.isArray(options.previousFindings) ? options.previousFindings : [];
+  const maxItems =
+    typeof options.maxItems === 'number' && options.maxItems > 0 ? options.maxItems : 20;
+  const maxTokens =
+    typeof options.maxTokens === 'number' && options.maxTokens > 0 ? options.maxTokens : 4000;
+
+  const rawEvidence = [];
+  const visitedPaths = new Set();
+
+  const MAX_FILE_SIZE = 1024 * 1024; // 1MB limit for bounded evidence extraction
+
+  function isPathInsideRepo(absPath, root) {
+    const rel = path.relative(root, absPath);
+    return !rel.startsWith('..') && !path.isAbsolute(rel);
+  }
+
+  // Helper to load file safely
+  const evidenceTimestamp = Date.now();
+  const fileCache = new Map();
+  function readFileSafe(relPath) {
+    let norm = normalizePath(relPath);
+    try {
+      norm = decodeURIComponent(norm);
+    } catch (_) {}
+    if (fileCache.has(norm)) return fileCache.get(norm);
+    const abs = path.resolve(repoRoot, norm);
+    try {
+      if (!isPathInsideRepo(abs, repoRoot)) {
+        fileCache.set(norm, null);
+        return null;
+      }
+      if (fs.existsSync(abs)) {
+        const stat = fs.statSync(abs);
+        if (stat.isFile() && stat.size <= MAX_FILE_SIZE) {
+          const text = fs.readFileSync(abs, 'utf8');
+          fileCache.set(norm, text);
+          return text;
+        }
+      }
+    } catch (_) {}
+    fileCache.set(norm, null);
+    return null;
+  }
+
+  // Helper to check if file exists
+  function fileExists(relPath) {
+    let norm = normalizePath(relPath);
+    try {
+      norm = decodeURIComponent(norm);
+    } catch (_) {}
+    const abs = path.resolve(repoRoot, norm);
+    try {
+      if (!isPathInsideRepo(abs, repoRoot)) return false;
+      return fs.existsSync(abs) && fs.statSync(abs).isFile();
+    } catch (_) {
+      return false;
+    }
+  }
+
+  // 1. Process Directly Changed Files with Priority & Budgeting
+  const MAX_DEEP_INSPECT_FILES = 50;
+
+  // Prioritize security boundaries and critical schema files first
+  changedFiles.sort((a, b) => {
+    const aRisk =
+      SECURITY_BOUNDARIES.some(p => p.test('/' + a)) || HIGH_RISK_EXTENSIONS.test(a) ? 1 : 0;
+    const bRisk =
+      SECURITY_BOUNDARIES.some(p => p.test('/' + b)) || HIGH_RISK_EXTENSIONS.test(b) ? 1 : 0;
+    return bRisk - aRisk;
+  });
+
+  for (let fIdx = 0; fIdx < changedFiles.length; fIdx++) {
+    const file = changedFiles[fIdx];
+    visitedPaths.add(file);
+    const isDeep = fIdx < MAX_DEEP_INSPECT_FILES;
+    const content = isDeep ? readFileSafe(file) : null;
+    const isSecurity =
+      SECURITY_BOUNDARIES.some(p => p.test('/' + file)) || HIGH_RISK_PATTERNS.test(file);
+    const isHighRiskExt = HIGH_RISK_EXTENSIONS.test(file);
+    const exists = isDeep ? fileExists(file) : true;
+
+    const reasons = ['directly modified file in change set'];
+    let score = 0.95;
+    const confidence = exists ? 'L1' : 'L2';
+
+    if (isSecurity || isHighRiskExt) {
+      score = 1.0;
+      reasons.push('touches critical security / schema boundary');
+    }
+
+    const snippet = content
+      ? extractBoundedSnippet(content, 1, null, 12)
+      : redactSecrets(diff.slice(0, 500) || '');
+
+    rawEvidence.push({
+      id: `file:${file}`,
+      type: 'changed_file',
+      path: file,
+      score,
+      reasons,
+      confidence,
+      provenance: 'OBSERVED',
+      timestamp: evidenceTimestamp,
+      is_mandatory: true,
+      estimated_tokens: estimateTokens(snippet || 100),
+      snippet,
+    });
+
+    if (!isDeep) continue; // For overflow files beyond capacity, skip deep AST symbol & dependency checks
+
+    // 2. Extract changed symbols/functions for this file
+    const symbols = extractChangedSymbols(diff, content);
+    for (const sym of symbols) {
+      const symSnippet = content ? extractBoundedSnippet(content, sym.diff_line, sym.name, 8) : '';
+      const symIsSecurity = HIGH_RISK_PATTERNS.test(sym.name);
+      const symReasons = [`modified ${sym.type} "${sym.name}" in ${file}`];
+      let symScore = 0.88;
+      if (symIsSecurity) {
+        symScore = 0.96;
+        symReasons.push('symbol name matches high-risk pattern');
+      }
+
+      rawEvidence.push({
+        id: `symbol:${file}:${sym.name}`,
+        type: 'changed_symbol',
+        path: file,
+        symbol: sym.name,
+        score: symScore,
+        reasons: symReasons,
+        confidence: exists ? 'L1' : 'L2',
+        provenance: 'INFERRED',
+        timestamp: evidenceTimestamp,
+        is_mandatory: false,
+        estimated_tokens: estimateTokens(symSnippet || 60),
+        snippet: symSnippet,
+      });
+    }
+
+    // 3. Direct imports & dependencies
+    if (content) {
+      const scanContent = content.length > 50000 ? content.slice(0, 50000) : content;
+      const importRegex =
+        /(?:import\s+(?:[^'"]*?\s+from\s+)?['"]([^'"]+)['"]|require\s*\(\s*['"]([^'"]+)['"]\s*\))/g;
+      let match;
+      while ((match = importRegex.exec(scanContent)) !== null) {
+        const importTarget = match[1] || match[2];
+        if (importTarget && (importTarget.startsWith('.') || importTarget.startsWith('/'))) {
+          // Resolve relative to file directory
+          const fileDir = path.dirname(file);
+          const resolvedRel = normalizePath(path.join(fileDir, importTarget));
+          // Check extensions (.js, .ts, /index.js, etc.)
+          const candidates = [
+            resolvedRel,
+            `${resolvedRel}.js`,
+            `${resolvedRel}.ts`,
+            `${resolvedRel}.json`,
+            `${resolvedRel}/index.js`,
+            `${resolvedRel}/index.ts`,
+          ];
+          const actualTarget = candidates.find(c => fileExists(c)) || resolvedRel;
+
+          if (!visitedPaths.has(actualTarget)) {
+            visitedPaths.add(actualTarget);
+            const depContent = readFileSafe(actualTarget);
+            const depIsSecurity = SECURITY_BOUNDARIES.some(p => p.test('/' + actualTarget));
+            const depReasons = [`direct dependency imported by ${file}`];
+            const depScore = depIsSecurity ? 0.9 : 0.8;
+            if (depIsSecurity) depReasons.push('dependency operates in security boundary');
+
+            rawEvidence.push({
+              id: `dep:${actualTarget}`,
+              type: 'direct_dependency',
+              path: actualTarget,
+              score: depScore,
+              reasons: depReasons,
+              confidence: fileExists(actualTarget) ? 'L2' : 'L3',
+              provenance: 'INFERRED',
+              timestamp: evidenceTimestamp,
+              is_mandatory: depIsSecurity,
+              estimated_tokens: estimateTokens(
+                depContent ? extractBoundedSnippet(depContent, 1, null, 8) : 80,
+              ),
+              snippet: depContent ? extractBoundedSnippet(depContent, 1, null, 8) : '',
+            });
+          }
+        }
+      }
+    }
+
+    // 4. Relevant tests for this changed file
+    const baseName = path.basename(file).replace(/\.[^.]+$/, '');
+    const possibleTestPaths = [
+      `test/unit/${baseName}.test.js`,
+      `test/unit/${baseName}.spec.js`,
+      `test/integration/${baseName}.test.js`,
+      `test/${baseName}.test.js`,
+      `tests/${baseName}.test.js`,
+      `${path.dirname(file)}/__tests__/${baseName}.test.js`,
+      `${path.dirname(file)}/${baseName}.test.js`,
+    ];
+
+    for (const testPath of possibleTestPaths) {
+      const normTest = normalizePath(testPath);
+      if (fileExists(normTest) && !visitedPaths.has(normTest)) {
+        visitedPaths.add(normTest);
+        const testContent = readFileSafe(normTest);
+        rawEvidence.push({
+          id: `test:${normTest}`,
+          type: 'relevant_test',
+          path: normTest,
+          score: 0.78,
+          reasons: [`test suite covering modified behavior in ${file}`],
+          confidence: 'L1',
+          provenance: 'OBSERVED',
+          timestamp: evidenceTimestamp,
+          is_mandatory: false,
+          estimated_tokens: estimateTokens(
+            testContent ? extractBoundedSnippet(testContent, 1, null, 10) : 100,
+          ),
+          snippet: testContent ? extractBoundedSnippet(testContent, 1, null, 10) : '',
+        });
+      }
+    }
+  }
+
+  // 5. Relevant configuration files (package.json, tsconfig.json, docker, etc.)
+  const commonConfigs = [
+    'package.json',
+    'tsconfig.json',
+    '.eslintrc.js',
+    'eslint.config.js',
+    'Dockerfile',
+    '.github/workflows/ci.yml',
+  ];
+  for (const cfg of commonConfigs) {
+    if (fileExists(cfg) && !visitedPaths.has(cfg)) {
+      const cfgContent = readFileSafe(cfg);
+      const isConfigRelevant =
+        changedFiles.some(cf => cf.endsWith('.json') || cf.endsWith('.js') || cf.endsWith('.ts')) ||
+        task.toLowerCase().includes('dependency') ||
+        task.toLowerCase().includes('package');
+
+      if (isConfigRelevant || cfg === 'package.json') {
+        visitedPaths.add(cfg);
+        rawEvidence.push({
+          id: `config:${cfg}`,
+          type: 'configuration',
+          path: cfg,
+          score: cfg === 'package.json' ? 0.72 : 0.65,
+          reasons: [`project configuration controlling runtime behavior (${cfg})`],
+          confidence: 'L1',
+          provenance: 'OBSERVED',
+          timestamp: evidenceTimestamp,
+          is_mandatory: false,
+          estimated_tokens: estimateTokens(
+            cfgContent ? extractBoundedSnippet(cfgContent, 1, null, 10) : 100,
+          ),
+          snippet: cfgContent ? extractBoundedSnippet(cfgContent, 1, null, 10) : '',
+        });
+      }
+    }
+  }
+
+  // 6. Security Boundaries (Root SECURITY.md, auth policies)
+  const rootSecurityFiles = ['SECURITY.md', '.env.example'];
+  for (const secFile of rootSecurityFiles) {
+    if (fileExists(secFile) && !visitedPaths.has(secFile)) {
+      visitedPaths.add(secFile);
+      const secContent = readFileSafe(secFile);
+      rawEvidence.push({
+        id: `security:${secFile}`,
+        type: 'security_boundary',
+        path: secFile,
+        score: 0.85,
+        reasons: [`repository root security and credential policy (${secFile})`],
+        confidence: 'L1',
+        provenance: 'OBSERVED',
+        timestamp: evidenceTimestamp,
+        is_mandatory: true,
+        estimated_tokens: estimateTokens(
+          secContent ? extractBoundedSnippet(secContent, 1, null, 10) : 120,
+        ),
+        snippet: secContent ? extractBoundedSnippet(secContent, 1, null, 10) : '',
+      });
+    }
+  }
+
+  // 7. Historical Memory from Memory Engine
+  for (const mem of memoryItems) {
+    const memContent = typeof mem === 'string' ? mem : mem.content || mem.summary || '';
+    if (!memContent) continue;
+    const memTags = Array.isArray(mem.tags) ? mem.tags : [];
+    const memIsRelevant =
+      changedFiles.some(f => {
+        const baseNoExt = path.basename(f, path.extname(f)).toLowerCase();
+        const baseWithExt = path.basename(f).toLowerCase();
+        return (
+          memContent.toLowerCase().includes(baseNoExt) ||
+          memContent.toLowerCase().includes(baseWithExt) ||
+          memTags.some(
+            t => baseNoExt.includes(t.toLowerCase()) || f.toLowerCase().includes(t.toLowerCase()),
+          )
+        );
+      }) ||
+      (task &&
+        (memContent.toLowerCase().includes(task.toLowerCase().slice(0, 15)) ||
+          memTags.some(t => task.toLowerCase().includes(t.toLowerCase()))));
+
+    if (memIsRelevant) {
+      rawEvidence.push({
+        id: `memory:${mem.id || Math.abs(memContent.length)}`,
+        type: 'historical_memory',
+        path: mem.tags ? mem.tags.join(',') : 'memory-store',
+        score: 0.68,
+        reasons: ['relevant historical memory from past sessions/decisions'],
+        confidence: 'L2',
+        provenance: 'REPORTED',
+        timestamp: evidenceTimestamp,
+        is_mandatory: false,
+        estimated_tokens: estimateTokens(memContent),
+        snippet: memContent.slice(0, 200),
+      });
+    }
+  }
+
+  // 8. Previous Findings
+  for (const f of prevFindings) {
+    const fDesc = f.desc || f.description || f.message || '';
+    if (!fDesc) continue;
+    rawEvidence.push({
+      id: `finding:${f.id || 'prev'}`,
+      type: 'previous_finding',
+      path: f.location || f.file || 'audit-result',
+      score: f.severity === 'Critical' ? 0.92 : f.severity === 'Important' ? 0.82 : 0.65,
+      reasons: [`previous finding: [${f.severity || 'Finding'}] ${fDesc}`],
+      confidence: 'L2',
+      provenance: 'REPORTED',
+      timestamp: evidenceTimestamp,
+      is_mandatory: f.severity === 'Critical',
+      estimated_tokens: estimateTokens(fDesc),
+      snippet: fDesc,
+    });
+  }
+
+  // Deterministic stable sorting:
+  // 1. score DESCENDING
+  // 2. TYPE_PRIORITY ASCENDING (1 to 9)
+  // 3. id / path ASCENDING (alphabetic)
+  rawEvidence.sort((a, b) => {
+    if (Math.abs(b.score - a.score) > 0.001) return b.score - a.score;
+    const prioA = TYPE_PRIORITY[a.type] || 99;
+    const prioB = TYPE_PRIORITY[b.type] || 99;
+    if (prioA !== prioB) return prioA - prioB;
+    return a.id.localeCompare(b.id);
+  });
+
+  // Apply budget constraints (maxItems & maxTokens)
+  const selectedEvidence = [];
+  const truncatedEvidence = [];
+  let tokenCount = 0;
+
+  // Step 1: Retain all mandatory items first
+  for (const item of rawEvidence) {
+    if (item.is_mandatory) {
+      selectedEvidence.push(item);
+      tokenCount += item.estimated_tokens;
+    }
+  }
+
+  // Step 2: Include non-mandatory items up to budget
+  for (const item of rawEvidence) {
+    if (item.is_mandatory) continue;
+    if (selectedEvidence.length < maxItems && tokenCount + item.estimated_tokens <= maxTokens) {
+      selectedEvidence.push(item);
+      tokenCount += item.estimated_tokens;
+    } else {
+      truncatedEvidence.push(item);
+    }
+  }
+
+  // Deterministic re-sort of selected items
+  selectedEvidence.sort((a, b) => {
+    if (Math.abs(b.score - a.score) > 0.001) return b.score - a.score;
+    const prioA = TYPE_PRIORITY[a.type] || 99;
+    const prioB = TYPE_PRIORITY[b.type] || 99;
+    if (prioA !== prioB) return prioA - prioB;
+    return a.id.localeCompare(b.id);
+  });
+
+  // Aggregate overall confidence and risk indicators
+  const touchesSecurityBoundary =
+    changedFiles.some(
+      f => SECURITY_BOUNDARIES.some(p => p.test('/' + f)) || HIGH_RISK_PATTERNS.test(f),
+    ) ||
+    (diff && HIGH_RISK_PATTERNS.test(diff));
+  const criticalCount = selectedEvidence.filter(e => e.is_mandatory).length;
+
+  let overallConfidence = 'L1';
+  if (selectedEvidence.length === 0) {
+    overallConfidence = 'L4';
+  } else if (selectedEvidence.some(e => e.confidence === 'L4')) {
+    overallConfidence = 'L3';
+  } else if (selectedEvidence.some(e => e.confidence === 'L3')) {
+    overallConfidence = 'L2';
+  }
+
+  return {
+    evidence: selectedEvidence,
+    truncated: truncatedEvidence,
+    summary: {
+      total_items_evaluated: rawEvidence.length,
+      total_items_selected: selectedEvidence.length,
+      total_tokens: tokenCount,
+      critical_evidence_count: criticalCount,
+      has_security_boundary: touchesSecurityBoundary,
+      confidence: overallConfidence,
+    },
+  };
+}
+
+module.exports = {
+  collectAndRankEvidence,
+  extractChangedSymbols,
+  extractBoundedSnippet,
+  redactSecrets,
+  TYPE_PRIORITY,
+  SECURITY_BOUNDARIES,
+};