Agent Engineering Method

The problem

Coding agents often start writing code before fully understanding the requirements, leading to hallucinations and broken builds. Without a structured plan, agents frequently waste tokens on circular debugging and plausible-looking but incorrect logic.

What it does

• Enforces a brainstorm phase to define observable behavior and constraints before any code is generated.
• Generates an executable, step-by-step implementation plan where every task includes a verification check.
• Implements a Red/Green TDD cycle, requiring failing tests before writing functional code.
• Manages subagent delegation by assigning specific file paths and success criteria to prevent file conflicts.
• Provides a systematic debugging framework focused on root-cause analysis rather than symptom patching.

Why this beats prompting it yourself

Generic prompts rely on the agent's internal logic to maintain context, which decays as the session grows. This skill provides a rigid engineering framework that forces documentation and verification at every stage. It stops the agent from "guessing" by requiring written specs and test-driven proofs, significantly reducing rework and drift.

Use cases

• Architecting complex new features where multiple files must be synchronized.
• Refactoring legacy modules where regression testing is critical.
• Orchestrating a team of subagents to work on independent components in parallel.
• Solving elusive bugs through disciplined hypothesis testing and reproduction steps.