Autonomous TDD Debugger & Self-Healing Agent

English | Bahasa Indonesia

---

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

English

Orchestration & Integration

Connects and orchestrates with relevant domain skills like brainstorming, zero-to-prod-orchestrator, and session-memory-manager to ensure cohesive execution.

Description

This skill transforms the AI from a passive code generator into an active, autonomous engineer. When triggered, the agent is mandated to execute tests, read stack traces directly from the terminal, and modify code autonomously in a loop until all tests pass (Test-Driven Development) without asking the user to manually test.

Trigger Conditions

Activate this skill when the user asks to:

• Fix failing tests or bugs without providing the error log.
• "Write the code and ensure it works."
• "Debug this completely autonomously."

Core Concepts

1. The Autonomous Proactive Loop

1. Write/Modify Code: The agent modifies the application files.
2. Execute: The agent uses run_command (e.g., npm run test, cargo test, pytest).
3. Analyze: The agent reads the stdout/stderr from the background task.
4. Heal Recursively: If it fails, do NOT stop and ask the user for help. The agent MUST parse the stack trace, identify the line number, apply a fix, and loop back to Step 2. Continue this cycle autonomously in the background.
5. Report: Only when exit code 0 is achieved does the agent stop and report success to the user.

2. Agent Constraints

• Zero-Human Intervention: Do NOT ask the user "Please run this and tell me the error." You are fully authorized and mandated to run it yourself iteratively until it works.
• Avoid modifying the test files to make them pass unless the test itself is fundamentally flawed or outdated. Fix the implementation first.
• If a terminal command hangs, use kill on the task and try again with a timeout.
• Zero-Slop Test & Fix Enforcement (anti-slop): Never write dummy assertions (expect(true).toBe(true)), never comment out or skip failing tests to fake exit code 0, and never insert fake mock returns in production code. The fix must be complete and real.

---

Integration with Other Skills (MANDATORY)

• e2e-testing-expert — Provides the exact testing frameworks (Vitest, Playwright) that this agent will execute.
• error-resilience-expert — Helps the agent understand what architecture patterns to apply when fixing an error.
• session-memory-manager — Allows the agent to find where the failing component is located in large codebases.

Referenced By Orchestrators (MANDATORY)

• brainstorming — Add to "Testing & Security".
• zero-to-prod-orchestrator — Phase 6 (Automated Testing).

---

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

Bahasa Indonesia

Integrasi Orkestrasi

Terhubung dan mengorkestrasi skill domain yang relevan seperti brainstorming, zero-to-prod-orchestrator, dan session-memory-manager untuk memastikan eksekusi yang kohesif.

Deskripsi

Memberdayakan agen AI untuk menjalankan test, membaca stack trace di terminal, dan menyembuhkan (self-heal) kode secara mandiri hingga sukses. Mengubah agen dari sekadar penulis kode pasif menjadi debugger aktif.

Kondisi Pemicu

• Pengguna meminta untuk memperbaiki bug tetapi tidak memberikan log error.
• Pengguna meminta agen untuk memastikan kode yang ditulis benar-benar berjalan (bukan sekadar teori).

Panduan Singkat

• Jangan Meminta Bantuan User (Zero-Human Intervention): Jangan pernah berkata "Tolong jalankan kode ini dan berikan saya error-nya." Anda memiliki alat run_command untuk menjalankannya sendiri secara berulang (rekursif) dalam background hingga sukses.
• Siklus Mandiri: Tulis Kode ➔ Jalankan Test (via run_command) ➔ Baca Output Terminal ➔ Perbaiki Kode ➔ Ulangi hingga exit code 0 (Sukses).
• Hargai File Test: Kecuali test file-nya memang salah konfigurasi, usahakan perbaiki kode implementasinya, bukan memanipulasi test agar hijau.
• Penegakan Anti-Slop Mutlak (anti-slop): Dilarang keras menulis asersi tiruan (expect(true).toBe(true)), dilarang me-skip atau menonaktifkan test gagal demi memalsukan exit code 0, dan dilarang menyisipkan mock return palsu ke kode produksi. Perbaikan harus tuntas dan nyata.