Agentic Engineering

A minimal, vendor-neutral starter for using coding agents without outsourcing understanding, judgment, or responsibility.

Copy these files into any software project. They work with any capable coding agent.

Agents are code generators; engineers are responsible for system design, constraints, verification, and judgment.

The goal is not merely correct generated code.

The goal is a continuous human ↔ agent feedback loop so the engineer still owns the system: intent, behavior, logic, architecture, invariants, tradeoffs, and verification.

Human Intent
    ↓
Agent Interpretation
    ↓
Shared Mental Model
    ↓
Human Feedback
    ↓
Logic / Algorithm
    ↓
Implementation
    ↓
Agent Explanation
    ↓
Verification
    ↓
Human Understanding

If the human engineer can no longer explain the important behavior of the system after agent work, the loop has failed.

Contents

AGENTS.md
.agents/skills/engineering-feedback-loop/SKILL.md
.agents/skills/architecture-picture/SKILL.md
.agents/skills/architecture-picture/HTML-PAGE.md
.agents/skills/frontend-interview-drill/SKILL.md

• AGENTS.md is the philosophy. Drop it at the project root. Many coding agents read this file automatically. It points at engineering-feedback-loop and requires post-complex teach-back after G2/G3 / large implementations.
• engineering-feedback-loop is the practice. It tells the agent when and how to keep the feedback loop, scaled to small, medium, and large changes.
• architecture-picture produces a visual HTML page (big picture, architecture, flow, modules) for large or architectural work before implementation. Temp file, open in browser — nothing committed to the repo.
• frontend-interview-drill is the interviewer. Use it in a practice workspace (interviews) to assign TypeScript, React, or Next.js challenges, hint without spoiling, and verify by running tests.

This repository intentionally contains no vendor-specific tooling: no package manager, CI, MCP config, or Claude / Codex / Cursor setup.

Use in another project

1. Copy AGENTS.md to the other project's root.
2. Copy .agents/skills/engineering-feedback-loop/ into that project's .agents/skills/ directory (create the folders if needed).
3. Copy .agents/skills/architecture-picture/ into that project's .agents/skills/ directory.
4. For interview practice, also copy .agents/skills/frontend-interview-drill/ into the exam-room repo (interviews).
5. Point the agent at the copied files if it does not load them by default.

Keep project-specific conventions in the host repository. Do not add stack, vendor, or product rules to AGENTS.md, engineering-feedback-loop, or architecture-picture.

What this is not

This is not a framework, CLI, or agent runtime.

It does not make the engineer unnecessary. It makes the engineer more effective while keeping them informed and in control.