@ryuenn3123/agentic-senior-core 1.9.0 → 1.9.2

package/scripts/benchmark-gate.mjs

@@ -1,121 +1,121 @@
-#!/usr/bin/env node
-
-/**
- * benchmark-gate.mjs
- *
- * Anti-regression gate for benchmark quality signals.
- * Fails when benchmark metrics drop below configured thresholds.
- */
-
-import { existsSync, readFileSync } from 'node:fs';
-import { dirname, join, resolve } from 'node:path';
-import { fileURLToPath } from 'node:url';
-import { execFileSync } from 'node:child_process';
-
-const SCRIPT_FILE_PATH = fileURLToPath(import.meta.url);
-const SCRIPT_DIR = dirname(SCRIPT_FILE_PATH);
-const REPOSITORY_ROOT = resolve(SCRIPT_DIR, '..');
-const BENCHMARK_THRESHOLD_PATH = join(REPOSITORY_ROOT, '.agent-context', 'state', 'benchmark-thresholds.json');
-const DETECTION_BENCHMARK_PATH = join(REPOSITORY_ROOT, 'scripts', 'detection-benchmark.mjs');
-
-function readThresholdConfiguration() {
-  if (!existsSync(BENCHMARK_THRESHOLD_PATH)) {
-    return {
-      minimumTop1Accuracy: 0.9,
-      maximumManualCorrectionRate: 0.12,
-      maximumTop1AccuracyDrop: 0.02,
-      maximumManualCorrectionIncrease: 0.03,
-      previousReleaseBaseline: {
-        top1Accuracy: 0.9167,
-        manualCorrectionRate: 0.0833,
-      },
-    };
-  }
-
-  return JSON.parse(readFileSync(BENCHMARK_THRESHOLD_PATH, 'utf8'));
-}
-
-function runDetectionBenchmark() {
-  const benchmarkRawOutput = execFileSync('node', [DETECTION_BENCHMARK_PATH], {
-    cwd: REPOSITORY_ROOT,
-    encoding: 'utf8',
-  });
-
-  return JSON.parse(benchmarkRawOutput);
-}
-
-function buildCheckResult(checkName, passed, details) {
-  return {
-    checkName,
-    passed,
-    details,
-  };
-}
-
-function runBenchmarkGate() {
-  const thresholdConfiguration = readThresholdConfiguration();
-  const benchmarkResult = runDetectionBenchmark();
-  const benchmarkChecks = [];
-
-  const top1AccuracyPassed = benchmarkResult.top1Accuracy >= thresholdConfiguration.minimumTop1Accuracy;
-  benchmarkChecks.push(
-    buildCheckResult(
-      'minimum-top1-accuracy',
-      top1AccuracyPassed,
-      `top1Accuracy=${benchmarkResult.top1Accuracy} minimum=${thresholdConfiguration.minimumTop1Accuracy}`,
-    ),
-  );
-
-  const manualCorrectionPassed = benchmarkResult.manualCorrectionRate <= thresholdConfiguration.maximumManualCorrectionRate;
-  benchmarkChecks.push(
-    buildCheckResult(
-      'maximum-manual-correction-rate',
-      manualCorrectionPassed,
-      `manualCorrectionRate=${benchmarkResult.manualCorrectionRate} maximum=${thresholdConfiguration.maximumManualCorrectionRate}`,
-    ),
-  );
-
-  const previousReleaseBaseline = thresholdConfiguration.previousReleaseBaseline;
-  if (previousReleaseBaseline && typeof previousReleaseBaseline === 'object') {
-    const top1AccuracyDrop = Number((previousReleaseBaseline.top1Accuracy - benchmarkResult.top1Accuracy).toFixed(4));
-    const manualCorrectionIncrease = Number((benchmarkResult.manualCorrectionRate - previousReleaseBaseline.manualCorrectionRate).toFixed(4));
-
-    const top1AccuracyDropPassed = top1AccuracyDrop <= thresholdConfiguration.maximumTop1AccuracyDrop;
-    benchmarkChecks.push(
-      buildCheckResult(
-        'maximum-top1-accuracy-drop',
-        top1AccuracyDropPassed,
-        `drop=${top1AccuracyDrop} maximum=${thresholdConfiguration.maximumTop1AccuracyDrop}`,
-      ),
-    );
-
-    const manualCorrectionIncreasePassed = manualCorrectionIncrease <= thresholdConfiguration.maximumManualCorrectionIncrease;
-    benchmarkChecks.push(
-      buildCheckResult(
-        'maximum-manual-correction-increase',
-        manualCorrectionIncreasePassed,
-        `increase=${manualCorrectionIncrease} maximum=${thresholdConfiguration.maximumManualCorrectionIncrease}`,
-      ),
-    );
-  }
-
-  const failedCheckCount = benchmarkChecks.filter((benchmarkCheck) => !benchmarkCheck.passed).length;
-  const benchmarkGateReport = {
-    generatedAt: new Date().toISOString(),
-    gateName: 'benchmark-gate',
-    passed: failedCheckCount === 0,
-    failureCount: failedCheckCount,
-    benchmarkResult: {
-      fixtureCount: benchmarkResult.fixtureCount,
-      top1Accuracy: benchmarkResult.top1Accuracy,
-      manualCorrectionRate: benchmarkResult.manualCorrectionRate,
-    },
-    thresholds: thresholdConfiguration,
-    results: benchmarkChecks,
-  };
-
-  console.log(JSON.stringify(benchmarkGateReport, null, 2));
-  process.exit(benchmarkGateReport.passed ? 0 : 1);
-}
-
-runBenchmarkGate();
+#!/usr/bin/env node
+
+/**
+ * benchmark-gate.mjs
+ *
+ * Anti-regression gate for benchmark quality signals.
+ * Fails when benchmark metrics drop below configured thresholds.
+ */
+
+import { existsSync, readFileSync } from 'node:fs';
+import { dirname, join, resolve } from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { execFileSync } from 'node:child_process';
+
+const SCRIPT_FILE_PATH = fileURLToPath(import.meta.url);
+const SCRIPT_DIR = dirname(SCRIPT_FILE_PATH);
+const REPOSITORY_ROOT = resolve(SCRIPT_DIR, '..');
+const BENCHMARK_THRESHOLD_PATH = join(REPOSITORY_ROOT, '.agent-context', 'state', 'benchmark-thresholds.json');
+const DETECTION_BENCHMARK_PATH = join(REPOSITORY_ROOT, 'scripts', 'detection-benchmark.mjs');
+
+function readThresholdConfiguration() {
+  if (!existsSync(BENCHMARK_THRESHOLD_PATH)) {
+    return {
+      minimumTop1Accuracy: 0.9,
+      maximumManualCorrectionRate: 0.12,
+      maximumTop1AccuracyDrop: 0.02,
+      maximumManualCorrectionIncrease: 0.03,
+      previousReleaseBaseline: {
+        top1Accuracy: 0.9167,
+        manualCorrectionRate: 0.0833,
+      },
+    };
+  }
+
+  return JSON.parse(readFileSync(BENCHMARK_THRESHOLD_PATH, 'utf8'));
+}
+
+function runDetectionBenchmark() {
+  const benchmarkRawOutput = execFileSync('node', [DETECTION_BENCHMARK_PATH], {
+    cwd: REPOSITORY_ROOT,
+    encoding: 'utf8',
+  });
+
+  return JSON.parse(benchmarkRawOutput);
+}
+
+function buildCheckResult(checkName, passed, details) {
+  return {
+    checkName,
+    passed,
+    details,
+  };
+}
+
+function runBenchmarkGate() {
+  const thresholdConfiguration = readThresholdConfiguration();
+  const benchmarkResult = runDetectionBenchmark();
+  const benchmarkChecks = [];
+
+  const top1AccuracyPassed = benchmarkResult.top1Accuracy >= thresholdConfiguration.minimumTop1Accuracy;
+  benchmarkChecks.push(
+    buildCheckResult(
+      'minimum-top1-accuracy',
+      top1AccuracyPassed,
+      `top1Accuracy=${benchmarkResult.top1Accuracy} minimum=${thresholdConfiguration.minimumTop1Accuracy}`,
+    ),
+  );
+
+  const manualCorrectionPassed = benchmarkResult.manualCorrectionRate <= thresholdConfiguration.maximumManualCorrectionRate;
+  benchmarkChecks.push(
+    buildCheckResult(
+      'maximum-manual-correction-rate',
+      manualCorrectionPassed,
+      `manualCorrectionRate=${benchmarkResult.manualCorrectionRate} maximum=${thresholdConfiguration.maximumManualCorrectionRate}`,
+    ),
+  );
+
+  const previousReleaseBaseline = thresholdConfiguration.previousReleaseBaseline;
+  if (previousReleaseBaseline && typeof previousReleaseBaseline === 'object') {
+    const top1AccuracyDrop = Number((previousReleaseBaseline.top1Accuracy - benchmarkResult.top1Accuracy).toFixed(4));
+    const manualCorrectionIncrease = Number((benchmarkResult.manualCorrectionRate - previousReleaseBaseline.manualCorrectionRate).toFixed(4));
+
+    const top1AccuracyDropPassed = top1AccuracyDrop <= thresholdConfiguration.maximumTop1AccuracyDrop;
+    benchmarkChecks.push(
+      buildCheckResult(
+        'maximum-top1-accuracy-drop',
+        top1AccuracyDropPassed,
+        `drop=${top1AccuracyDrop} maximum=${thresholdConfiguration.maximumTop1AccuracyDrop}`,
+      ),
+    );
+
+    const manualCorrectionIncreasePassed = manualCorrectionIncrease <= thresholdConfiguration.maximumManualCorrectionIncrease;
+    benchmarkChecks.push(
+      buildCheckResult(
+        'maximum-manual-correction-increase',
+        manualCorrectionIncreasePassed,
+        `increase=${manualCorrectionIncrease} maximum=${thresholdConfiguration.maximumManualCorrectionIncrease}`,
+      ),
+    );
+  }
+
+  const failedCheckCount = benchmarkChecks.filter((benchmarkCheck) => !benchmarkCheck.passed).length;
+  const benchmarkGateReport = {
+    generatedAt: new Date().toISOString(),
+    gateName: 'benchmark-gate',
+    passed: failedCheckCount === 0,
+    failureCount: failedCheckCount,
+    benchmarkResult: {
+      fixtureCount: benchmarkResult.fixtureCount,
+      top1Accuracy: benchmarkResult.top1Accuracy,
+      manualCorrectionRate: benchmarkResult.manualCorrectionRate,
+    },
+    thresholds: thresholdConfiguration,
+    results: benchmarkChecks,
+  };
+
+  console.log(JSON.stringify(benchmarkGateReport, null, 2));
+  process.exit(benchmarkGateReport.passed ? 0 : 1);
+}
+
+runBenchmarkGate();

package/scripts/benchmark-intelligence.mjs

@@ -1,140 +1,140 @@
-#!/usr/bin/env node
-
-/**
- * benchmark-intelligence.mjs
- *
- * Competitive intelligence cadence validator.
- * Ensures benchmark watchlist coverage and review freshness.
- */
-
-import { existsSync, readFileSync } from 'node:fs';
-import { dirname, join, resolve } from 'node:path';
-import { fileURLToPath } from 'node:url';
-
-const SCRIPT_FILE_PATH = fileURLToPath(import.meta.url);
-const SCRIPT_DIR = dirname(SCRIPT_FILE_PATH);
-const REPOSITORY_ROOT = resolve(SCRIPT_DIR, '..');
-const WATCHLIST_PATH = join(REPOSITORY_ROOT, '.agent-context', 'state', 'benchmark-watchlist.json');
-const REVIEW_SLA_DAYS = 14;
-const REQUIRED_BENCHMARK_REPOSITORIES = new Set([
-  'sickn33/antigravity-awesome-skills',
-  'github/awesome-copilot',
-  'MiniMax-AI/skills',
-]);
-
-function parseDateOrNull(rawDateValue) {
-  if (typeof rawDateValue !== 'string') {
-    return null;
-  }
-
-  if (!/^\d{4}-\d{2}-\d{2}$/.test(rawDateValue)) {
-    return null;
-  }
-
-  const parsedDate = new Date(`${rawDateValue}T00:00:00.000Z`);
-  return Number.isNaN(parsedDate.getTime()) ? null : parsedDate;
-}
-
-function calculateAgeInDays(referenceDate, targetDate) {
-  const ageInMilliseconds = referenceDate.getTime() - targetDate.getTime();
-  return Math.floor(ageInMilliseconds / (1000 * 60 * 60 * 24));
-}
-
-function loadWatchlistConfiguration() {
-  if (!existsSync(WATCHLIST_PATH)) {
-    return { repositories: [] };
-  }
-
-  return JSON.parse(readFileSync(WATCHLIST_PATH, 'utf8'));
-}
-
-function runIntelligenceValidation() {
-  const watchlistConfiguration = loadWatchlistConfiguration();
-  const watchlistEntries = Array.isArray(watchlistConfiguration.repositories)
-    ? watchlistConfiguration.repositories
-    : [];
-  const validationResults = [];
-  const currentDate = new Date();
-
-  const trackedRepositoryNames = new Set();
-  for (const watchlistEntry of watchlistEntries) {
-    trackedRepositoryNames.add(watchlistEntry.repository);
-  }
-
-  for (const requiredRepositoryName of REQUIRED_BENCHMARK_REPOSITORIES) {
-    const hasRequiredRepository = trackedRepositoryNames.has(requiredRepositoryName);
-    validationResults.push({
-      checkName: 'required-benchmark-repository',
-      repository: requiredRepositoryName,
-      passed: hasRequiredRepository,
-      details: hasRequiredRepository
-        ? `${requiredRepositoryName} is present in watchlist`
-        : `${requiredRepositoryName} is missing from watchlist`,
-    });
-  }
-
-  const watchlistReport = watchlistEntries.map((watchlistEntry) => {
-    const repositoryName = watchlistEntry.repository;
-    const repositoryOwner = watchlistEntry.owner;
-    const lastReviewedDate = parseDateOrNull(watchlistEntry.lastReviewedAt);
-
-    const hasOwner = typeof repositoryOwner === 'string' && repositoryOwner.trim().length > 0;
-    validationResults.push({
-      checkName: 'watchlist-owner-defined',
-      repository: repositoryName,
-      passed: hasOwner,
-      details: hasOwner ? `Owner ${repositoryOwner} is defined` : 'Owner is missing',
-    });
-
-    if (!lastReviewedDate) {
-      validationResults.push({
-        checkName: 'review-date-format',
-        repository: repositoryName,
-        passed: false,
-        details: `Invalid or missing lastReviewedAt: ${String(watchlistEntry.lastReviewedAt)}`,
-      });
-
-      return {
-        repository: repositoryName,
-        owner: repositoryOwner,
-        lastReviewedAt: watchlistEntry.lastReviewedAt,
-        ageInDays: null,
-        stale: true,
-      };
-    }
-
-    const reviewAgeInDays = calculateAgeInDays(currentDate, lastReviewedDate);
-    const reviewWithinSla = reviewAgeInDays <= REVIEW_SLA_DAYS;
-
-    validationResults.push({
-      checkName: 'review-sla-compliance',
-      repository: repositoryName,
-      passed: reviewWithinSla,
-      details: `ageInDays=${reviewAgeInDays} slaDays=${REVIEW_SLA_DAYS}`,
-    });
-
-    return {
-      repository: repositoryName,
-      owner: repositoryOwner,
-      lastReviewedAt: watchlistEntry.lastReviewedAt,
-      ageInDays: reviewAgeInDays,
-      stale: !reviewWithinSla,
-    };
-  });
-
-  const failedCheckCount = validationResults.filter((validationResult) => !validationResult.passed).length;
-  const intelligenceReport = {
-    generatedAt: new Date().toISOString(),
-    reportName: 'benchmark-intelligence',
-    passed: failedCheckCount === 0,
-    failureCount: failedCheckCount,
-    reviewSlaDays: REVIEW_SLA_DAYS,
-    watchlist: watchlistReport,
-    results: validationResults,
-  };
-
-  console.log(JSON.stringify(intelligenceReport, null, 2));
-  process.exit(intelligenceReport.passed ? 0 : 1);
-}
-
-runIntelligenceValidation();
+#!/usr/bin/env node
+
+/**
+ * benchmark-intelligence.mjs
+ *
+ * Competitive intelligence cadence validator.
+ * Ensures benchmark watchlist coverage and review freshness.
+ */
+
+import { existsSync, readFileSync } from 'node:fs';
+import { dirname, join, resolve } from 'node:path';
+import { fileURLToPath } from 'node:url';
+
+const SCRIPT_FILE_PATH = fileURLToPath(import.meta.url);
+const SCRIPT_DIR = dirname(SCRIPT_FILE_PATH);
+const REPOSITORY_ROOT = resolve(SCRIPT_DIR, '..');
+const WATCHLIST_PATH = join(REPOSITORY_ROOT, '.agent-context', 'state', 'benchmark-watchlist.json');
+const REVIEW_SLA_DAYS = 14;
+const REQUIRED_BENCHMARK_REPOSITORIES = new Set([
+  'sickn33/antigravity-awesome-skills',
+  'github/awesome-copilot',
+  'MiniMax-AI/skills',
+]);
+
+function parseDateOrNull(rawDateValue) {
+  if (typeof rawDateValue !== 'string') {
+    return null;
+  }
+
+  if (!/^\d{4}-\d{2}-\d{2}$/.test(rawDateValue)) {
+    return null;
+  }
+
+  const parsedDate = new Date(`${rawDateValue}T00:00:00.000Z`);
+  return Number.isNaN(parsedDate.getTime()) ? null : parsedDate;
+}
+
+function calculateAgeInDays(referenceDate, targetDate) {
+  const ageInMilliseconds = referenceDate.getTime() - targetDate.getTime();
+  return Math.floor(ageInMilliseconds / (1000 * 60 * 60 * 24));
+}
+
+function loadWatchlistConfiguration() {
+  if (!existsSync(WATCHLIST_PATH)) {
+    return { repositories: [] };
+  }
+
+  return JSON.parse(readFileSync(WATCHLIST_PATH, 'utf8'));
+}
+
+function runIntelligenceValidation() {
+  const watchlistConfiguration = loadWatchlistConfiguration();
+  const watchlistEntries = Array.isArray(watchlistConfiguration.repositories)
+    ? watchlistConfiguration.repositories
+    : [];
+  const validationResults = [];
+  const currentDate = new Date();
+
+  const trackedRepositoryNames = new Set();
+  for (const watchlistEntry of watchlistEntries) {
+    trackedRepositoryNames.add(watchlistEntry.repository);
+  }
+
+  for (const requiredRepositoryName of REQUIRED_BENCHMARK_REPOSITORIES) {
+    const hasRequiredRepository = trackedRepositoryNames.has(requiredRepositoryName);
+    validationResults.push({
+      checkName: 'required-benchmark-repository',
+      repository: requiredRepositoryName,
+      passed: hasRequiredRepository,
+      details: hasRequiredRepository
+        ? `${requiredRepositoryName} is present in watchlist`
+        : `${requiredRepositoryName} is missing from watchlist`,
+    });
+  }
+
+  const watchlistReport = watchlistEntries.map((watchlistEntry) => {
+    const repositoryName = watchlistEntry.repository;
+    const repositoryOwner = watchlistEntry.owner;
+    const lastReviewedDate = parseDateOrNull(watchlistEntry.lastReviewedAt);
+
+    const hasOwner = typeof repositoryOwner === 'string' && repositoryOwner.trim().length > 0;
+    validationResults.push({
+      checkName: 'watchlist-owner-defined',
+      repository: repositoryName,
+      passed: hasOwner,
+      details: hasOwner ? `Owner ${repositoryOwner} is defined` : 'Owner is missing',
+    });
+
+    if (!lastReviewedDate) {
+      validationResults.push({
+        checkName: 'review-date-format',
+        repository: repositoryName,
+        passed: false,
+        details: `Invalid or missing lastReviewedAt: ${String(watchlistEntry.lastReviewedAt)}`,
+      });
+
+      return {
+        repository: repositoryName,
+        owner: repositoryOwner,
+        lastReviewedAt: watchlistEntry.lastReviewedAt,
+        ageInDays: null,
+        stale: true,
+      };
+    }
+
+    const reviewAgeInDays = calculateAgeInDays(currentDate, lastReviewedDate);
+    const reviewWithinSla = reviewAgeInDays <= REVIEW_SLA_DAYS;
+
+    validationResults.push({
+      checkName: 'review-sla-compliance',
+      repository: repositoryName,
+      passed: reviewWithinSla,
+      details: `ageInDays=${reviewAgeInDays} slaDays=${REVIEW_SLA_DAYS}`,
+    });
+
+    return {
+      repository: repositoryName,
+      owner: repositoryOwner,
+      lastReviewedAt: watchlistEntry.lastReviewedAt,
+      ageInDays: reviewAgeInDays,
+      stale: !reviewWithinSla,
+    };
+  });
+
+  const failedCheckCount = validationResults.filter((validationResult) => !validationResult.passed).length;
+  const intelligenceReport = {
+    generatedAt: new Date().toISOString(),
+    reportName: 'benchmark-intelligence',
+    passed: failedCheckCount === 0,
+    failureCount: failedCheckCount,
+    reviewSlaDays: REVIEW_SLA_DAYS,
+    watchlist: watchlistReport,
+    results: validationResults,
+  };
+
+  console.log(JSON.stringify(intelligenceReport, null, 2));
+  process.exit(intelligenceReport.passed ? 0 : 1);
+}
+
+runIntelligenceValidation();