closed-loop-cli 1.0.3 → 1.0.4

package/src/orchestrator/dgm-archive.ts

@@ -1,257 +0,0 @@
-import * as fs from 'fs';
-import * as path from 'path';
-
-/**
- * Darwin Gödel Machine (DGM) — Archive Entry
- * แต่ละ entry คือ snapshot ของ agent ณ จุดหนึ่งใน evolutionary lineage
- */
-export interface ArchiveEntry {
-  /** Unique identifier สำหรับ snapshot นี้ */
-  id: string;
-  /** Git commit hash ที่ snapshot นี้ถูก commit ไว้ */
-  commitHash: string;
-  /** ID ของ parent entry ที่สร้าง snapshot นี้ขึ้นมา (null = origin) */
-  parentId: string | null;
-  /** Empirical fitness score (0–1) วัดจาก test pass rate */
-  fitness: number;
-  /** Mutation strategy ที่ถูกใช้เพื่อสร้าง snapshot นี้ */
-  mutationStrategy: string;
-  /** Task description ที่ mutation นี้ถูก apply สำหรับ */
-  task: string;
-  /** Timestamp ของการสร้าง snapshot */
-  timestamp: string;
-  /** Metadata เพิ่มเติม (test counts, metrics, etc.) */
-  metadata: {
-    passCount?: number;
-    failCount?: number;
-    totalTests?: number;
-    passRate?: number;
-    goalDriftIndex?: number;
-    constraintPreservationScore?: number;
-    [key: string]: unknown;
-  };
-}
-
-/**
- * Summary statistics ของ Archive ปัจจุบัน
- */
-export interface ArchiveStats {
-  totalEntries: number;
-  bestFitness: number;
-  averageFitness: number;
-  bestEntryId: string | null;
-  latestEntryId: string | null;
-  generationDepth: number;
-}
-
-const ARCHIVE_FILENAME = 'dgm-archive.json';
-const DEFAULT_MAX_SIZE = 20;
-
-/**
- * DGM Archive Manager
- *
- * เก็บ population ของ agent snapshots และจัดการ selection สำหรับ
- * Darwin Gödel Machine evolution loop ตาม paper:
- * "Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents"
- */
-export class DGMArchive {
-  private archivePath: string;
-  private entries: ArchiveEntry[];
-
-  constructor(workspaceRoot?: string) {
-    const root = workspaceRoot || process.cwd();
-    this.archivePath = path.join(root, ARCHIVE_FILENAME);
-    this.entries = this.load();
-  }
-
-  // ─────────────────────────────────────────
-  // Persistence
-  // ─────────────────────────────────────────
-
-  private load(): ArchiveEntry[] {
-    if (!fs.existsSync(this.archivePath)) {
-      return [];
-    }
-    try {
-      const raw = fs.readFileSync(this.archivePath, 'utf-8');
-      const parsed = JSON.parse(raw);
-      return Array.isArray(parsed.entries) ? parsed.entries : [];
-    } catch {
-      return [];
-    }
-  }
-
-  private save(): void {
-    const data = {
-      version: '1.0.0',
-      lastUpdated: new Date().toISOString(),
-      entries: this.entries
-    };
-    try {
-      fs.writeFileSync(this.archivePath, JSON.stringify(data, null, 2), 'utf-8');
-    } catch (err) {
-      console.error('[DGMArchive] Failed to save archive:', err);
-    }
-  }
-
-  // ─────────────────────────────────────────
-  // Core Operations
-  // ─────────────────────────────────────────
-
-  /**
-   * เพิ่ม snapshot ใหม่เข้า archive
-   * @returns entry ที่เพิ่งเพิ่มเข้าไป
-   */
-  addEntry(entry: Omit<ArchiveEntry, 'id' | 'timestamp'>): ArchiveEntry {
-    const newEntry: ArchiveEntry = {
-      ...entry,
-      id: this.generateId(),
-      timestamp: new Date().toISOString()
-    };
-    this.entries.push(newEntry);
-    this.save();
-    return newEntry;
-  }
-
-  /**
-   * เลือก parent entry สำหรับ mutation รอบต่อไป
-   * ใช้ fitness-weighted sampling (DGM parent selection strategy)
-   *
-   * @param strategy 'fitness_weighted' | 'best' | 'random'
-   */
-  getBestParent(strategy: 'fitness_weighted' | 'best' | 'random' = 'fitness_weighted'): ArchiveEntry | null {
-    if (this.entries.length === 0) return null;
-
-    if (strategy === 'best') {
-      return [...this.entries].sort((a, b) => b.fitness - a.fitness)[0];
-    }
-
-    if (strategy === 'random') {
-      return this.entries[Math.floor(Math.random() * this.entries.length)];
-    }
-
-    // fitness_weighted: สุ่มโดยน้ำหนักตาม fitness score (DGM default)
-    return this.fitnessWeightedSample();
-  }
-
-  /**
-   * Trace lineage จาก entry ไปถึง origin
-   * @returns ordered array ตั้งแต่ origin จนถึง entry นี้
-   */
-  getLineage(entryId: string): ArchiveEntry[] {
-    const lineage: ArchiveEntry[] = [];
-    let current = this.findById(entryId);
-
-    while (current) {
-      lineage.unshift(current);
-      if (!current.parentId) break;
-      current = this.findById(current.parentId);
-    }
-
-    return lineage;
-  }
-
-  /**
-   * ตัด entries เก่าเพื่อป้องกัน archive ใหญ่เกินไป
-   * จะเก็บ entries ที่มี fitness สูงสุดไว้ก่อน
-   */
-  pruneOldEntries(maxSize: number = DEFAULT_MAX_SIZE): number {
-    if (this.entries.length <= maxSize) return 0;
-
-    // เรียง fitness สูงสุดก่อน แล้วตัดส่วนที่เกิน
-    const sorted = [...this.entries].sort((a, b) => b.fitness - a.fitness);
-    const pruned = this.entries.length - maxSize;
-    this.entries = sorted.slice(0, maxSize);
-    this.save();
-    return pruned;
-  }
-
-  // ─────────────────────────────────────────
-  // Query / Stats
-  // ─────────────────────────────────────────
-
-  getAll(): ArchiveEntry[] {
-    return [...this.entries];
-  }
-
-  findById(id: string): ArchiveEntry | undefined {
-    return this.entries.find(e => e.id === id);
-  }
-
-  getBestEntry(): ArchiveEntry | null {
-    if (this.entries.length === 0) return null;
-    return [...this.entries].sort((a, b) => b.fitness - a.fitness)[0];
-  }
-
-  getStats(): ArchiveStats {
-    if (this.entries.length === 0) {
-      return {
-        totalEntries: 0,
-        bestFitness: 0,
-        averageFitness: 0,
-        bestEntryId: null,
-        latestEntryId: null,
-        generationDepth: 0
-      };
-    }
-
-    const sorted = [...this.entries].sort((a, b) => b.fitness - a.fitness);
-    const best = sorted[0];
-    const latest = [...this.entries].sort(
-      (a, b) => new Date(b.timestamp).getTime() - new Date(a.timestamp).getTime()
-    )[0];
-
-    const avgFitness =
-      this.entries.reduce((sum, e) => sum + e.fitness, 0) / this.entries.length;
-
-    const depth = latest ? this.getLineage(latest.id).length : 0;
-
-    return {
-      totalEntries: this.entries.length,
-      bestFitness: best.fitness,
-      averageFitness: Math.round(avgFitness * 1000) / 1000,
-      bestEntryId: best.id,
-      latestEntryId: latest?.id ?? null,
-      generationDepth: depth
-    };
-  }
-
-  /**
-   * ดึง history ล่าสุด N entries (เรียงตามเวลา)
-   */
-  getRecentHistory(n: number = 10): ArchiveEntry[] {
-    return [...this.entries]
-      .sort((a, b) => new Date(b.timestamp).getTime() - new Date(a.timestamp).getTime())
-      .slice(0, n);
-  }
-
-  // ─────────────────────────────────────────
-  // Private Helpers
-  // ─────────────────────────────────────────
-
-  /**
-   * Fitness-weighted random sampling (Roulette Wheel Selection)
-   * สูตรจาก DGM paper: เลือก parent ตาม probability ∝ fitness
-   */
-  private fitnessWeightedSample(): ArchiveEntry {
-    const totalFitness = this.entries.reduce((sum, e) => sum + Math.max(e.fitness, 0.01), 0);
-    let r = Math.random() * totalFitness;
-
-    for (const entry of this.entries) {
-      r -= Math.max(entry.fitness, 0.01);
-      if (r <= 0) return entry;
-    }
-
-    // fallback: คืน entry สุดท้าย
-    return this.entries[this.entries.length - 1];
-  }
-
-  /**
-   * สร้าง unique ID แบบ timestamp + random
-   */
-  private generateId(): string {
-    const ts = Date.now().toString(36);
-    const rand = Math.random().toString(36).substring(2, 6);
-    return `dgm_${ts}_${rand}`;
-  }
-}

package/src/orchestrator/fitness-evaluator.ts

@@ -1,154 +0,0 @@
-import { runCommand } from '../tools/shell-tools';
-
-/**
- * Fitness Report จากการ evaluate โดย unit tests
- * Fitness = empirical test pass rate (0–1) ตาม DGM paper concept
- */
-export interface FitnessReport {
-  /** Pass rate (0–1) = passCount / totalTests */
-  passRate: number;
-  /** จำนวน tests ที่ผ่าน */
-  passCount: number;
-  /** จำนวน tests ที่ fail */
-  failCount: number;
-  /** จำนวน tests ทั้งหมด */
-  totalTests: number;
-  /** Composite fitness score (0–1) รวม compile status */
-  score: number;
-  /** Compilation ผ่านหรือไม่ */
-  compileSuccess: boolean;
-  /** สรุปผล */
-  summary: string;
-  /** Raw output จาก test runner */
-  rawOutput?: string;
-}
-
-/**
- * ผลเปรียบเทียบ fitness ระหว่าง baseline กับ candidate
- */
-export interface FitnessComparison {
-  improved: boolean;
-  delta: number;
-  baseline: FitnessReport;
-  candidate: FitnessReport;
-  message: string;
-}
-
-/**
- * วัด empirical fitness จาก test suite ปัจจุบัน
- *
- * DGM approach: fitness วัดจากผล unit test จริง ไม่ใช่ heuristic เพียงอย่างเดียว
- * "Each new agent version is evaluated empirically using coding benchmarks"
- */
-export async function evaluateFitness(): Promise<FitnessReport> {
-  // Step 1: ตรวจ compile
-  const compileRes = await runCommand('npm run build');
-  const compileSuccess = compileRes.exitCode === 0;
-
-  if (!compileSuccess) {
-    return {
-      passRate: 0,
-      passCount: 0,
-      failCount: 0,
-      totalTests: 0,
-      score: 0,
-      compileSuccess: false,
-      summary: `Compilation failed — fitness = 0`,
-      rawOutput: compileRes.stdout + compileRes.stderr
-    };
-  }
-
-  // Step 2: run tests
-  const testRes = await runCommand('npm test');
-  const rawOutput = (testRes.stdout || '') + (testRes.stderr || '');
-
-  // Step 3: parse test output
-  const parsed = parseTestOutput(rawOutput);
-
-  // Step 4: คำนวณ composite score
-  // score = pass_rate ถ้า compile ผ่าน, 0 ถ้า compile fail
-  const score = compileSuccess ? parsed.passRate : 0;
-
-  const summary =
-    `Compile: ${compileSuccess ? 'PASS' : 'FAIL'} | ` +
-    `Tests: ${parsed.passCount}/${parsed.totalTests} passed ` +
-    `(${(parsed.passRate * 100).toFixed(1)}%) | Score: ${score.toFixed(3)}`;
-
-  return {
-    ...parsed,
-    score,
-    compileSuccess,
-    summary,
-    rawOutput
-  };
-}
-
-/**
- * เปรียบเทียบ fitness ระหว่าง baseline snapshot กับ candidate snapshot
- */
-export function compareToBaseline(
-  candidate: FitnessReport,
-  baseline: FitnessReport
-): FitnessComparison {
-  const delta = candidate.score - baseline.score;
-  const improved = delta > 0;
-
-  const message = improved
-    ? `✔ Fitness improved: ${baseline.score.toFixed(3)} → ${candidate.score.toFixed(3)} (+${delta.toFixed(3)})`
-    : delta === 0
-    ? `= Fitness unchanged: ${candidate.score.toFixed(3)}`
-    : `✘ Fitness regressed: ${baseline.score.toFixed(3)} → ${candidate.score.toFixed(3)} (${delta.toFixed(3)})`;
-
-  return { improved, delta, baseline, candidate, message };
-}
-
-/**
- * Parse test runner output เพื่อนับ pass/fail counts
- * รองรับ format ของ test runner ปัจจุบันในโปรเจกต์
- */
-export function parseTestOutput(output: string): Pick<
-  FitnessReport,
-  'passRate' | 'passCount' | 'failCount' | 'totalTests'
-> {
-  let passCount = 0;
-  let failCount = 0;
-
-  // Pattern 1: "[Pass] <test name>" ซึ่งเป็น format ของ run-tests.ts ปัจจุบัน
-  const passMatches = output.match(/\[Pass\]/g);
-  const failMatches = output.match(/\[Fail\]/g);
-
-  if (passMatches || failMatches) {
-    passCount = passMatches ? passMatches.length : 0;
-    failCount = failMatches ? failMatches.length : 0;
-  }
-
-  // Pattern 2: Mocha format — "X passing" / "X failing"
-  const mochaPassing = output.match(/(\d+)\s+passing/i);
-  const mochaFailing = output.match(/(\d+)\s+failing/i);
-
-  if (mochaPassing || mochaFailing) {
-    passCount = mochaPassing ? parseInt(mochaPassing[1], 10) : passCount;
-    failCount = mochaFailing ? parseInt(mochaFailing[1], 10) : failCount;
-  }
-
-  // Pattern 3: Jest format — "Tests: X passed, Y failed"
-  const jestLine = output.match(/Tests:\s+(\d+)\s+passed(?:,\s+(\d+)\s+failed)?/i);
-  if (jestLine) {
-    passCount = parseInt(jestLine[1], 10) || passCount;
-    failCount = jestLine[2] ? parseInt(jestLine[2], 10) : failCount;
-  }
-
-  // Pattern 4: "ALL BINARY ASSERTION TESTS PASSED" — หมายถึง 100% ผ่าน
-  if (output.includes('ALL BINARY ASSERTION TESTS PASSED')) {
-    // ถ้าตรวจจับ pass ไม่ได้ชัดเจนให้ fallback = ผ่านทั้งหมด
-    if (passCount === 0 && failCount === 0) {
-      passCount = 11; // จำนวน tests ใน run-tests.ts (Test 1–11)
-      failCount = 0;
-    }
-  }
-
-  const totalTests = passCount + failCount;
-  const passRate = totalTests > 0 ? passCount / totalTests : (failCount === 0 ? 1 : 0);
-
-  return { passRate, passCount, failCount, totalTests };
-}

package/src/orchestrator/mutation-strategies.ts

@@ -1,214 +0,0 @@
-import { DGMArchive, ArchiveEntry } from './dgm-archive';
-
-/**
- * DGM Mutation Strategies
- *
- * จาก DGM paper: open-ended evolution ต้องการ diversity ของ mutation types
- * ไม่ใช่แค่ refactor อย่างเดียว — เพื่อสำรวจ search space ที่กว้างขึ้น
- */
-export enum MutationStrategy {
-  /** เพิ่ม feature ใหม่เข้าไปในระบบ */
-  ADD_FEATURE = 'add_feature',
-  /** Refactor code เดิมให้อ่านง่ายขึ้น / ลด duplication */
-  REFACTOR = 'refactor',
-  /** Optimize performance หรือ token efficiency */
-  OPTIMIZE = 'optimize',
-  /** เพิ่ม unit tests เพื่อเพิ่ม coverage */
-  ADD_TESTS = 'add_tests',
-  /** แก้ bug ที่ตรวจพบจาก test failures */
-  FIX_BUG = 'fix_bug',
-  /** ปรับปรุง system prompt / agent prompt */
-  IMPROVE_PROMPT = 'improve_prompt',
-}
-
-/**
- * ผลการเลือก mutation strategy
- */
-export interface MutationSelection {
-  strategy: MutationStrategy;
-  rationale: string;
-  targetHint?: string;
-}
-
-/**
- * เลือก mutation strategy ที่เหมาะสมจาก archive history + task context
- *
- * DGM approach: strategy selection ควร adaptive ตาม population history
- * — ถ้า fitness ต่ำ → ลอง ADD_TESTS หรือ FIX_BUG ก่อน
- * — ถ้า fitness สูงแล้ว → ลอง ADD_FEATURE หรือ OPTIMIZE
- * — ถ้า task พูดถึง bug → เลือก FIX_BUG ก่อน
- */
-export function selectMutationStrategy(
-  archive: DGMArchive,
-  task: string,
-  currentFitness: number = 0
-): MutationSelection {
-  const taskLower = task.toLowerCase();
-
-  // 1. Task-based override: ถ้า task บ่งชี้ strategy ชัดเจน
-  if (taskLower.includes('fix') || taskLower.includes('bug') || taskLower.includes('error')) {
-    return {
-      strategy: MutationStrategy.FIX_BUG,
-      rationale: 'Task description indicates a bug fix is needed'
-    };
-  }
-
-  if (taskLower.includes('test') || taskLower.includes('coverage')) {
-    return {
-      strategy: MutationStrategy.ADD_TESTS,
-      rationale: 'Task description requests test improvements'
-    };
-  }
-
-  if (taskLower.includes('prompt') || taskLower.includes('instruction')) {
-    return {
-      strategy: MutationStrategy.IMPROVE_PROMPT,
-      rationale: 'Task description targets prompt engineering'
-    };
-  }
-
-  if (taskLower.includes('refactor') || taskLower.includes('clean') || taskLower.includes('reformat')) {
-    return {
-      strategy: MutationStrategy.REFACTOR,
-      rationale: 'Task description requests code refactoring'
-    };
-  }
-
-  if (taskLower.includes('optim') || taskLower.includes('speed') || taskLower.includes('performance') || taskLower.includes('token')) {
-    return {
-      strategy: MutationStrategy.OPTIMIZE,
-      rationale: 'Task description requests optimization'
-    };
-  }
-
-  if (taskLower.includes('add') || taskLower.includes('implement') || taskLower.includes('create') || taskLower.includes('new feature')) {
-    return {
-      strategy: MutationStrategy.ADD_FEATURE,
-      rationale: 'Task description requests adding a new feature'
-    };
-  }
-
-  // 2. Fitness-based heuristic: ถ้า fitness ต่ำ → focus on fixing
-  if (currentFitness < 0.7) {
-    return {
-      strategy: MutationStrategy.FIX_BUG,
-      rationale: `Low fitness (${(currentFitness * 100).toFixed(1)}%) — prioritizing stability fixes`
-    };
-  }
-
-  // 3. Archive diversity: ตรวจสอบว่า strategy ไหนถูกใช้ไปน้อยสุดใน archive
-  const recentHistory = archive.getRecentHistory(8);
-  if (recentHistory.length > 0) {
-    const strategyCounts: Record<string, number> = {};
-    for (const s of Object.values(MutationStrategy)) {
-      strategyCounts[s] = 0;
-    }
-    for (const entry of recentHistory) {
-      if (entry.mutationStrategy && strategyCounts[entry.mutationStrategy] !== undefined) {
-        strategyCounts[entry.mutationStrategy]++;
-      }
-    }
-
-    // เลือก strategy ที่ถูกใช้น้อยสุด (diversity promotion)
-    const leastUsed = Object.entries(strategyCounts)
-      .sort((a, b) => a[1] - b[1])[0];
-
-    if (leastUsed && leastUsed[1] < 2) {
-      return {
-        strategy: leastUsed[0] as MutationStrategy,
-        rationale: `Diversity promotion: "${leastUsed[0]}" has been used least (${leastUsed[1]}x in recent history)`
-      };
-    }
-  }
-
-  // 4. Default: ADD_FEATURE (DGM paper prefers expansive mutations)
-  return {
-    strategy: MutationStrategy.ADD_FEATURE,
-    rationale: 'Default open-ended evolution strategy: expanding capabilities'
-  };
-}
-
-/**
- * สร้าง task prompt เฉพาะ mutation strategy
- * ให้ Agent ทราบว่ากำลังทำ mutation ประเภทใด
- */
-export function generateMutationPrompt(
-  strategy: MutationStrategy,
-  originalTask: string,
-  context?: {
-    targetFile?: string;
-    parentEntry?: ArchiveEntry | null;
-    currentFitness?: number;
-  }
-): string {
-  const parentInfo = context?.parentEntry
-    ? `\n[DGM Context] Building upon parent snapshot: ${context.parentEntry.id} (fitness: ${(context.parentEntry.fitness * 100).toFixed(1)}%)`
-    : '';
-
-  const fitnessInfo = context?.currentFitness !== undefined
-    ? `\n[DGM Context] Current system fitness: ${(context.currentFitness * 100).toFixed(1)}% (test pass rate)`
-    : '';
-
-  const fileInfo = context?.targetFile
-    ? `\n[DGM Context] Primary target file: ${context.targetFile}`
-    : '';
-
-  const dgmHeader = `[DGM Mutation: ${strategy.toUpperCase()}]${parentInfo}${fitnessInfo}${fileInfo}\n\n`;
-
-  switch (strategy) {
-    case MutationStrategy.ADD_FEATURE:
-      return dgmHeader +
-        `Your mutation goal is to ADD A NEW FEATURE. Implement the following capability:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Create new files or add new exported functions/classes as needed\n` +
-        `- Do not break existing functionality\n` +
-        `- Ensure all new code compiles and existing tests still pass\n` +
-        `- The feature should integrate cleanly with the existing codebase`;
-
-    case MutationStrategy.REFACTOR:
-      return dgmHeader +
-        `Your mutation goal is to REFACTOR existing code for better quality:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Improve readability, reduce duplication, or strengthen type safety\n` +
-        `- Preserve all existing behavior exactly (zero functional change)\n` +
-        `- All existing tests must still pass after refactoring`;
-
-    case MutationStrategy.OPTIMIZE:
-      return dgmHeader +
-        `Your mutation goal is to OPTIMIZE for performance or efficiency:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Focus on reducing token usage, execution time, or memory\n` +
-        `- Preserve correctness — all tests must still pass\n` +
-        `- Measure and report the optimization impact if possible`;
-
-    case MutationStrategy.ADD_TESTS:
-      return dgmHeader +
-        `Your mutation goal is to ADD UNIT TESTS to improve coverage:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Add tests to src/tests/dynamic/ directory as .ts files\n` +
-        `- Tests must export a default function or a run() function\n` +
-        `- Prioritize testing edge cases and untested public functions\n` +
-        `- Do not modify existing test files (protected by Campbell Regime)`;
-
-    case MutationStrategy.FIX_BUG:
-      return dgmHeader +
-        `Your mutation goal is to FIX A BUG or stability issue:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Diagnose the root cause carefully before making changes\n` +
-        `- Make the minimal change required to fix the issue\n` +
-        `- Add a regression test if appropriate\n` +
-        `- All tests must pass after the fix`;
-
-    case MutationStrategy.IMPROVE_PROMPT:
-      return dgmHeader +
-        `Your mutation goal is to IMPROVE AGENT PROMPTS for better performance:\n${originalTask}\n\n` +
-        `Guidelines:\n` +
-        `- Edit src/orchestrator/system-prompt.txt or src/orchestrator/agent-prompts.ts\n` +
-        `- Make prompts more concise, clear, and effective\n` +
-        `- Preserve all existing agent capabilities\n` +
-        `- Build compiles and tests pass after changes`;
-
-    default:
-      return dgmHeader + originalTask;
-  }
-}