HCMD — Human‑Centered Meta‑Development

A structured, semantic, and deterministic workflow for AI‑assisted software development

HCMD is a methodology that separates meaning, tasks, and structure into explicit, human‑authored layers that guide AI tools during implementation.
It provides a stable, reproducible alternative to prompt engineering and ad‑hoc agent behavior by introducing a clear semantic protocol:

• AML — the meaning layer
• STS — the task layer
• PSC — the structural layer
• Code — generated by AI tools from PSC

HCMD is designed for developers and scientific professionals who want AI assistance without losing architectural control, determinism, or reproducibility.
It works naturally with Windows, Visual Studio 2026, VS Code, and the Copilot App.

---

Why HCMD?

Current AI coding tools are powerful, but they lack:

• semantic grounding
• explicit task decomposition
• implementation‑free architecture
• reproducibility across sessions
• deterministic behavior

HCMD addresses these gaps by externalizing the reasoning process into structured, human‑authored documents validated by XML/XSD.
This allows humans to define what the system means and how it is shaped, while AI handles the mechanical implementation.

---

Skills and Huts in HCMD

HCMD adopts the term Skill in a non‑agentic way.

In agentic systems, a Skill is an autonomous capability used for planning or orchestration. HCMD takes the opposite approach: a Skill is a Hut — a dynamic thinking protocol that constrains how Copilot reasons within a specific conceptual stance.

• A Skill/Hut contains no knowledge
• It defines how to think, not what to know
• It is dynamic and reconstructed on demand
• It enables co‑existence and co‑work between human and AI

This keeps HCMD lightweight, deterministic, and eco‑efficient, in contrast to resource‑intensive agentic architectures.

---

Repository Structure

HCMD is organized into three layers:

concept/        →        environment/        →        (future) implementation/
meaning                 operation                     execution

concept/ — The Conceptual Layer

This folder contains the complete conceptual foundation of HCMD:

• HCMD Manifesto
• HCMD overview and principles
• AML, PSC, and HMD conceptual definitions
• The Three‑Layer Model
• Conceptual workflow and example

These documents define the discipline before any tooling or implementation enters the picture.

---

environment/ — The Operational Layer (Reference Environment)

This folder explains how HCMD is actually executed using:

• Copilot App — the semantic environment
• GitHub Copilot in Visual Studio 2026 or VS Code — the implementation environment
• Windows — the stable execution substrate

Although this folder uses Windows, Copilot App, and .NET as the reference environment, HCMD itself is platform‑agnostic; this layer is specific only because the operational workflow must be concrete.

It includes:

• the role of the Copilot App
• the role of GitHub Copilot
• the operational workflow
• a complete example session

This layer is essential because AML, STS, and PSC are not written manually.
They are generated inside the environment through structured interaction with Copilot tools.

---

Future Layers

The following layers will be added as HCMD evolves:

• implementation/ — reference schemas, templates, generators, and HMD tooling
• examples/ — complete HCMD projects
• docs/ — public documentation for practitioners

These layers will build on the conceptual and environment foundations.

---

Current Status

This repository is the official home of HCMD.
It currently contains:

• the complete conceptual layer
• the complete environment layer (reference environment)
• initial structure for future implementation work

More materials — including AML/STS/PSC specifications, XSD schemas, and full example projects — will be added over time.

---

About the Author

HCMD is my personal research project, and you can reach me at H_Nishimura@lbl.gov, where I remain affiliated.

---