@ryuenn3123/agentic-senior-core 1.9.0 → 1.9.1

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();

package/scripts/detection-benchmark.mjs

@@ -1,138 +1,138 @@
-#!/usr/bin/env node
-
-/**
- * detection-benchmark.mjs
- *
- * Lightweight benchmark for stack detection heuristics used by the CLI.
- * Publishes top-1 accuracy and manual-correction-rate proxy.
- */
-
-const BLUEPRINT_RECOMMENDATIONS = {
-  'typescript.md': 'api-nextjs.md',
-  'python.md': 'fastapi-service.md',
-  'java.md': 'spring-boot-api.md',
-  'php.md': 'laravel-api.md',
-  'go.md': 'go-service.md',
-  'csharp.md': 'aspnet-api.md',
-};
-
-const AMBIGUITY_THRESHOLD = 0.08;
-
-const BENCHMARK_FIXTURES = [
-  { fixtureName: 'typescript-basic', markers: ['package.json', 'tsconfig.json'], expectedStack: 'typescript.md' },
-  { fixtureName: 'typescript-next', markers: ['package.json', 'tsconfig.json', 'next.config.js'], expectedStack: 'typescript.md' },
-  { fixtureName: 'python-poetry', markers: ['pyproject.toml'], expectedStack: 'python.md' },
-  { fixtureName: 'python-requirements', markers: ['requirements.txt'], expectedStack: 'python.md' },
-  { fixtureName: 'java-maven', markers: ['pom.xml'], expectedStack: 'java.md' },
-  { fixtureName: 'java-gradle', markers: ['build.gradle'], expectedStack: 'java.md' },
-  { fixtureName: 'php-composer', markers: ['composer.json'], expectedStack: 'php.md' },
-  { fixtureName: 'go-module', markers: ['go.mod'], expectedStack: 'go.md' },
-  { fixtureName: 'dotnet-solution', markers: ['app.sln'], expectedStack: 'csharp.md' },
-  { fixtureName: 'rust-cargo', markers: ['Cargo.toml'], expectedStack: 'rust.md' },
-  { fixtureName: 'ruby-gemfile', markers: ['Gemfile'], expectedStack: 'ruby.md' },
-  { fixtureName: 'mixed-ts-python', markers: ['package.json', 'tsconfig.json', 'pyproject.toml'], expectedStack: 'typescript.md' },
-];
-
-function detectProjectFromMarkers(markers) {
-  const markerNames = new Set(markers);
-  const candidates = [];
-
-  if (markerNames.has('package.json') || markerNames.has('tsconfig.json') || markerNames.has('next.config.js') || markerNames.has('next.config.mjs')) {
-    let confidenceScore = 0.7;
-    if (markerNames.has('package.json')) confidenceScore += 0.12;
-    if (markerNames.has('tsconfig.json')) confidenceScore += 0.12;
-    if (markerNames.has('next.config.js') || markerNames.has('next.config.mjs')) confidenceScore += 0.05;
-    candidates.push({ stackFileName: 'typescript.md', confidenceScore: Math.min(confidenceScore, 0.97) });
-  }
-
-  if (markerNames.has('pyproject.toml') || markerNames.has('requirements.txt')) {
-    candidates.push({
-      stackFileName: 'python.md',
-      confidenceScore: markerNames.has('pyproject.toml') ? 0.96 : 0.78,
-    });
-  }
-
-  if (markerNames.has('pom.xml') || markerNames.has('build.gradle') || markerNames.has('build.gradle.kts')) {
-    candidates.push({
-      stackFileName: 'java.md',
-      confidenceScore: markerNames.has('pom.xml') ? 0.95 : 0.84,
-    });
-  }
-
-  if (markerNames.has('composer.json')) candidates.push({ stackFileName: 'php.md', confidenceScore: 0.95 });
-  if (markerNames.has('go.mod')) candidates.push({ stackFileName: 'go.md', confidenceScore: 0.96 });
-  if (markerNames.has('Cargo.toml')) candidates.push({ stackFileName: 'rust.md', confidenceScore: 0.96 });
-  if (markerNames.has('Gemfile')) candidates.push({ stackFileName: 'ruby.md', confidenceScore: 0.95 });
-
-  const hasDotNetMarker = Array.from(markerNames).some((markerName) => markerName.endsWith('.sln') || markerName.endsWith('.csproj'));
-  if (hasDotNetMarker) {
-    candidates.push({ stackFileName: 'csharp.md', confidenceScore: 0.95 });
-  }
-
-  if (candidates.length === 0) {
-    return {
-      recommendedStack: null,
-      recommendedBlueprint: null,
-      confidenceScore: 0,
-      confidenceGap: 0,
-      needsManualCorrection: true,
-    };
-  }
-
-  candidates.sort((leftCandidate, rightCandidate) => rightCandidate.confidenceScore - leftCandidate.confidenceScore);
-  const strongestCandidate = candidates[0];
-  const secondStrongestCandidate = candidates[1];
-  const confidenceGap = secondStrongestCandidate
-    ? Number((strongestCandidate.confidenceScore - secondStrongestCandidate.confidenceScore).toFixed(2))
-    : Number(strongestCandidate.confidenceScore.toFixed(2));
-
-  return {
-    recommendedStack: strongestCandidate.stackFileName,
-    recommendedBlueprint: BLUEPRINT_RECOMMENDATIONS[strongestCandidate.stackFileName] || null,
-    confidenceScore: strongestCandidate.confidenceScore,
-    confidenceGap,
-    needsManualCorrection: confidenceGap < AMBIGUITY_THRESHOLD,
-  };
-}
-
-function runBenchmark() {
-  let passingFixtureCount = 0;
-  let manualCorrectionCount = 0;
-
-  const fixtureResults = BENCHMARK_FIXTURES.map((benchmarkFixture) => {
-    const detectionResult = detectProjectFromMarkers(benchmarkFixture.markers);
-    const isCorrect = detectionResult.recommendedStack === benchmarkFixture.expectedStack;
-
-    if (isCorrect) {
-      passingFixtureCount += 1;
-    }
-
-    if (detectionResult.needsManualCorrection) {
-      manualCorrectionCount += 1;
-    }
-
-    return {
-      fixtureName: benchmarkFixture.fixtureName,
-      expectedStack: benchmarkFixture.expectedStack,
-      detectedStack: detectionResult.recommendedStack,
-      confidenceGap: detectionResult.confidenceGap,
-      needsManualCorrection: detectionResult.needsManualCorrection,
-      isCorrect,
-    };
-  });
-
-  const totalFixtureCount = BENCHMARK_FIXTURES.length;
-  const top1Accuracy = Number((passingFixtureCount / totalFixtureCount).toFixed(4));
-  const manualCorrectionRate = Number((manualCorrectionCount / totalFixtureCount).toFixed(4));
-
-  return {
-    generatedAt: new Date().toISOString(),
-    fixtureCount: totalFixtureCount,
-    top1Accuracy,
-    manualCorrectionRate,
-    fixtures: fixtureResults,
-  };
-}
-
-const benchmarkResult = runBenchmark();
-console.log(JSON.stringify(benchmarkResult, null, 2));
+#!/usr/bin/env node
+
+/**
+ * detection-benchmark.mjs
+ *
+ * Lightweight benchmark for stack detection heuristics used by the CLI.
+ * Publishes top-1 accuracy and manual-correction-rate proxy.
+ */
+
+const BLUEPRINT_RECOMMENDATIONS = {
+  'typescript.md': 'api-nextjs.md',
+  'python.md': 'fastapi-service.md',
+  'java.md': 'spring-boot-api.md',
+  'php.md': 'laravel-api.md',
+  'go.md': 'go-service.md',
+  'csharp.md': 'aspnet-api.md',
+};
+
+const AMBIGUITY_THRESHOLD = 0.08;
+
+const BENCHMARK_FIXTURES = [
+  { fixtureName: 'typescript-basic', markers: ['package.json', 'tsconfig.json'], expectedStack: 'typescript.md' },
+  { fixtureName: 'typescript-next', markers: ['package.json', 'tsconfig.json', 'next.config.js'], expectedStack: 'typescript.md' },
+  { fixtureName: 'python-poetry', markers: ['pyproject.toml'], expectedStack: 'python.md' },
+  { fixtureName: 'python-requirements', markers: ['requirements.txt'], expectedStack: 'python.md' },
+  { fixtureName: 'java-maven', markers: ['pom.xml'], expectedStack: 'java.md' },
+  { fixtureName: 'java-gradle', markers: ['build.gradle'], expectedStack: 'java.md' },
+  { fixtureName: 'php-composer', markers: ['composer.json'], expectedStack: 'php.md' },
+  { fixtureName: 'go-module', markers: ['go.mod'], expectedStack: 'go.md' },
+  { fixtureName: 'dotnet-solution', markers: ['app.sln'], expectedStack: 'csharp.md' },
+  { fixtureName: 'rust-cargo', markers: ['Cargo.toml'], expectedStack: 'rust.md' },
+  { fixtureName: 'ruby-gemfile', markers: ['Gemfile'], expectedStack: 'ruby.md' },
+  { fixtureName: 'mixed-ts-python', markers: ['package.json', 'tsconfig.json', 'pyproject.toml'], expectedStack: 'typescript.md' },
+];
+
+function detectProjectFromMarkers(markers) {
+  const markerNames = new Set(markers);
+  const candidates = [];
+
+  if (markerNames.has('package.json') || markerNames.has('tsconfig.json') || markerNames.has('next.config.js') || markerNames.has('next.config.mjs')) {
+    let confidenceScore = 0.7;
+    if (markerNames.has('package.json')) confidenceScore += 0.12;
+    if (markerNames.has('tsconfig.json')) confidenceScore += 0.12;
+    if (markerNames.has('next.config.js') || markerNames.has('next.config.mjs')) confidenceScore += 0.05;
+    candidates.push({ stackFileName: 'typescript.md', confidenceScore: Math.min(confidenceScore, 0.97) });
+  }
+
+  if (markerNames.has('pyproject.toml') || markerNames.has('requirements.txt')) {
+    candidates.push({
+      stackFileName: 'python.md',
+      confidenceScore: markerNames.has('pyproject.toml') ? 0.96 : 0.78,
+    });
+  }
+
+  if (markerNames.has('pom.xml') || markerNames.has('build.gradle') || markerNames.has('build.gradle.kts')) {
+    candidates.push({
+      stackFileName: 'java.md',
+      confidenceScore: markerNames.has('pom.xml') ? 0.95 : 0.84,
+    });
+  }
+
+  if (markerNames.has('composer.json')) candidates.push({ stackFileName: 'php.md', confidenceScore: 0.95 });
+  if (markerNames.has('go.mod')) candidates.push({ stackFileName: 'go.md', confidenceScore: 0.96 });
+  if (markerNames.has('Cargo.toml')) candidates.push({ stackFileName: 'rust.md', confidenceScore: 0.96 });
+  if (markerNames.has('Gemfile')) candidates.push({ stackFileName: 'ruby.md', confidenceScore: 0.95 });
+
+  const hasDotNetMarker = Array.from(markerNames).some((markerName) => markerName.endsWith('.sln') || markerName.endsWith('.csproj'));
+  if (hasDotNetMarker) {
+    candidates.push({ stackFileName: 'csharp.md', confidenceScore: 0.95 });
+  }
+
+  if (candidates.length === 0) {
+    return {
+      recommendedStack: null,
+      recommendedBlueprint: null,
+      confidenceScore: 0,
+      confidenceGap: 0,
+      needsManualCorrection: true,
+    };
+  }
+
+  candidates.sort((leftCandidate, rightCandidate) => rightCandidate.confidenceScore - leftCandidate.confidenceScore);
+  const strongestCandidate = candidates[0];
+  const secondStrongestCandidate = candidates[1];
+  const confidenceGap = secondStrongestCandidate
+    ? Number((strongestCandidate.confidenceScore - secondStrongestCandidate.confidenceScore).toFixed(2))
+    : Number(strongestCandidate.confidenceScore.toFixed(2));
+
+  return {
+    recommendedStack: strongestCandidate.stackFileName,
+    recommendedBlueprint: BLUEPRINT_RECOMMENDATIONS[strongestCandidate.stackFileName] || null,
+    confidenceScore: strongestCandidate.confidenceScore,
+    confidenceGap,
+    needsManualCorrection: confidenceGap < AMBIGUITY_THRESHOLD,
+  };
+}
+
+function runBenchmark() {
+  let passingFixtureCount = 0;
+  let manualCorrectionCount = 0;
+
+  const fixtureResults = BENCHMARK_FIXTURES.map((benchmarkFixture) => {
+    const detectionResult = detectProjectFromMarkers(benchmarkFixture.markers);
+    const isCorrect = detectionResult.recommendedStack === benchmarkFixture.expectedStack;
+
+    if (isCorrect) {
+      passingFixtureCount += 1;
+    }
+
+    if (detectionResult.needsManualCorrection) {
+      manualCorrectionCount += 1;
+    }
+
+    return {
+      fixtureName: benchmarkFixture.fixtureName,
+      expectedStack: benchmarkFixture.expectedStack,
+      detectedStack: detectionResult.recommendedStack,
+      confidenceGap: detectionResult.confidenceGap,
+      needsManualCorrection: detectionResult.needsManualCorrection,
+      isCorrect,
+    };
+  });
+
+  const totalFixtureCount = BENCHMARK_FIXTURES.length;
+  const top1Accuracy = Number((passingFixtureCount / totalFixtureCount).toFixed(4));
+  const manualCorrectionRate = Number((manualCorrectionCount / totalFixtureCount).toFixed(4));
+
+  return {
+    generatedAt: new Date().toISOString(),
+    fixtureCount: totalFixtureCount,
+    top1Accuracy,
+    manualCorrectionRate,
+    fixtures: fixtureResults,
+  };
+}
+
+const benchmarkResult = runBenchmark();
+console.log(JSON.stringify(benchmarkResult, null, 2));