Autonomous TDD Debugger & Self-Healing Agent

English | Bahasa Indonesia

---

<a name="english"></a>

English

Orchestration & Integration

• Test Frameworks & E2E: e2e-testing-expert, property-mutation-testing-expert, accessibility-testing-expert
• Quality Gates & Anti-Slop: anti-slop, ai-code-review-autonomous, zero-tech-debt-auditor
• Formal Verification & Contracts: formal-spec-z3-verifier, spec-driven-development-expert
• Error Resilience & Healing: error-resilience-expert, self-healing-cloud-orchestrator, living-codebase-ast-graph
• Lifecycle & Governance: zero-to-prod-orchestrator, session-memory-manager

Description

Transforms the AI agent from a passive code generator into an active, autonomous Test-Driven Development (TDD) engineer. When invoked, the agent is mandated to execute tests, parse terminal stack traces, analyze failing assertions, and autonomously modify code in an iterative self-healing loop until all tests pass (exit code 0)—without prompting the user to manually run commands or debug errors.

Trigger Conditions

• User asks to fix failing tests or debug runtime errors without providing full error logs.
• Directives such as "Write the feature and ensure all tests pass", "Fix this bug autonomously", or "Refactor and verify".
• Phase 6 (Automated Testing & Quality Gate) execution in zero-to-prod-orchestrator.
• Post-migration regression validation in legacy-code-translator.

Autonomous Self-Healing Loop (5-Step Protocol)

┌────────────────────────────────────────────────────────┐
│ 1. EXECUTE TEST RUNNER                                 │
│    Run `vitest run`, `cargo test`, `pytest`, or `go test`│
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│ 2. PARSE STACK TRACE & FILTER NOISE                   │
│    Extract failing file, line number, expected vs actual│
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│ 3. AST BLAST-RADIUS CHECK                              │
│    Locate caller functions; isolate logic from tests   │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│ 4. APPLY PRECISION PATCH                               │
│    Fix production implementation; NEVER mock fakes     │
└───────────────────────────┬────────────────────────────┘
                            │
                            ▼
┌────────────────────────────────────────────────────────┐
│ 5. RE-RUN & VERIFY (Max 5 Iterations)                  │
│    Exit Code 0? ──► [YES] Complete & Report Artifact   │
│                 └──► [NO]  Loop back to Step 2         │
└────────────────────────────────────────────────────────┘

Self-Healing Script Example (TypeScript / Vitest)

// scripts/tdd-runner.mjs
import { execSync } from 'node:child_process';
import path from 'node:path';

export function runTDDLoop(testPattern = '', maxAttempts = 5): boolean {
  console.log(`[TDD-Debugger] Starting autonomous test cycle (Limit: ${maxAttempts} attempts)...`);

  for (let attempt = 1; attempt <= maxAttempts; attempt++) {
    console.log(`\n--- Attempt ${attempt} of ${maxAttempts} ---`);
    try {
      const output = execSync(`npx vitest run ${testPattern} --reporter=verbose`, {
        encoding: 'utf8',
        stdio: 'pipe',
      });
      console.log('✅ All tests passed successfully!');
      return true;
    } catch (err: any) {
      const stdout = err.stdout?.toString() || '';
      const stderr = err.stderr?.toString() || '';
      const combined = stdout + '\n' + stderr;

      // Extract failing test location
      const frameMatch = combined.match(/(?:FAIL|Error:)\s+(.*?):(\d+):(\d+)/) ||
                         combined.match(/at\s+(.*?):(\d+):(\d+)/);

      if (frameMatch) {
        const [, file, line] = frameMatch;
        console.warn(`[TDD-Debugger] Failure detected at ${path.basename(file)}:${line}`);
      } else {
        console.warn('[TDD-Debugger] Test failure detected. Analyzing raw output...');
      }

      if (attempt === maxAttempts) {
        console.error('❌ Reached maximum self-healing attempts. Circuit breaker tripped.');
        return false;
      }
    }
  }
  return false;
}

Non-Negotiable Agent Constraints

1. Zero-Human Intervention: NEVER ask the user "Please run this and tell me the error". You possess run_command—execute it directly in the terminal, read stdout/stderr, and iterate until green.
2. Respect Test Integrity (Anti-Slop Pillar 2):
   • Do NOT comment out or skip failing tests (test.skip) to fake a passing suite.
   • Do NOT replace failing assertions with trivial tautologies (expect(true).toBe(true)).
   • Fix the production source code first. Only modify test files if the test specification itself is demonstrably obsolete or contradictory.
3. Circuit Breaker: Capped at 5 self-healing iterations. If an issue remains unresolved after 5 cycles, pause to re-verify fundamental architectural assumptions.

---

<a name="bahasa-indonesia"></a>

Bahasa Indonesia

Integrasi Orkestrasi

• Framework Pengujian & E2E: e2e-testing-expert, property-mutation-testing-expert, accessibility-testing-expert
• Gerbang Kualitas & Anti-Slop: anti-slop, ai-code-review-autonomous, zero-tech-debt-auditor
• Verifikasi Formal & Kontrak: formal-spec-z3-verifier, spec-driven-development-expert
• Ketahanan Error & Pemulihan: error-resilience-expert, self-healing-cloud-orchestrator, living-codebase-ast-graph
• Siklus Hidup & Tata Kelola: zero-to-prod-orchestrator, session-memory-manager

Deskripsi

Mengubah agen AI dari sekadar pembuat kode pasif menjadi insinyur Test-Driven Development (TDD) yang aktif dan mandiri. Saat diaktifkan, agen diwajibkan menjalankan pengujian otomatis, mem-parsing stack trace di terminal, menganalisis kegagalan asersi, dan memodifikasi kode secara otonom dalam siklus perbaikan (self-healing loop) berulang hingga seluruh pengujian lulus (exit code 0)—tanpa meminta pengguna untuk melakukan pengujian manual.

Kondisi Pemicu

• Pengguna meminta untuk memperbaiki test yang gagal atau mencari bug tanpa menyertakan log error lengkap.
• Instruksi seperti "Tulis fitur ini dan pastikan semua test lulus", "Debug kode ini secara mandiri", atau "Refaktor dan buktikan dengan test".
• Eksekusi Fase 6 (Pengujian Otomatis & Gerbang Kualitas) pada zero-to-prod-orchestrator.
• Verifikasi regresi setelah migrasi kode warisan pada legacy-code-translator.

Siklus Mandiri 5 Langkah

1. JALANKAN TEST RUNNER (via run_command)
   npx vitest run, cargo test, pytest, atau go test
        │
        ▼
2. FILTER STACK TRACE & NOISE TERMINAL
   Ekstrak file yang gagal, nomor baris, nilai aktual vs ekspektasi
        │
        ▼
3. ANALISIS RADIUS DAMPAK (AST CHECK)
   Cari fungsi implementasi terkait; pisahkan logika dari file test
        │
        ▼
4. TERAPKAN PERBAIKAN PRESISI
   Perbaiki implementasi di kode produksi; DILARANG membuat mock palsu
        │
        ▼
5. VERIFIKASI ULANG (Maksimal 5 Iterasi)
   Exit code 0? ──► [YA] Selesai & Laporkan ke pengguna
                └──► [TIDAK] Ulangi ke Langkah 2

Panduan Ketat Tindakan Agen

1. Dilarang Meminta Bantuan User (Zero-Human Intervention): Jangan pernah berkata "Tolong jalankan perintah ini dan beritahu saya error-nya". Anda memiliki tool run_command untuk menjalankannya sendiri, membaca output terminal, dan memperbaikinya.
2. Kepatuhan Anti-Slop (Pilar 2):
   • Dilarang keras menonaktifkan test gagal (test.skip) demi memalsukan exit code 0.
   • Dilarang mengubah asersi menjadi bernilai selalu benar (expect(true).toBe(true)).
   • Selalu perbaiki kode implementasi produksi terlebih dahulu.
3. Circuit Breaker 5 Siklus: Jika kode belum berhasil diperbaiki setelah 5 kali iterasi, hentikan loop dan evaluasi ulang arsitektur logika dasar agar tidak menghabiskan konteks.