HCMD — Human‑Centered Meta‑Development (2026 Edition)

HCMD is a deterministic, semantic, and reproducible workflow for AI‑assisted software development.It separates meaning, structure, and implementation into explicit human‑authored layers.HCMD provides a stable alternative to prompt engineering by externalizing reasoning into a structured pipeline:

MeaningMeaning‑StabilizationAMLSTSPSCImplementation

HCMD is engine‑agnostic.Any AI system that satisfies HCMD’s semantic and structural constraints can implement the pipeline.Today, the Copilot pair is the first practical practitioner. Purpose

---

Modern AI coding tools lack:

• semantic grounding

• explicit task decomposition

• deterministic structure

• reproducibility

• regeneration stability

• separation of meaning from mechanics

HCMD addresses these gaps by defining a pipeline where humans control meaning and structure, and AI performs mechanical execution. The HCMD Pipeline

---

MeaningHuman intent expressed naturally.

Meaning‑StabilizationClarifies intent, resolves ambiguity, and prepares semantic ground.

AMLSemantic meaning expressed in a structured, language‑agnostic form.

STSProcedural meaning. Defines what must be done without implementation mechanics.

PSCDeterministic structure. Defines how the work is organized.

ImplementationMechanical code generation performed by an AI system.

The pipeline is one‑directional and regeneration‑safe. Engine‑Agnostic Architecture

---

HCMD does not depend on any specific AI system.It requires two roles:

Semantic AI

• meaning refinement

• meaning‑stabilization

• AML generation

• STS generation

Mechanical AI

• PSC generation

• implementation generation

• deterministic regeneration

Any future AI system that can perform these roles can implement HCMD. Copilot Pair as Reference Implementation

---

Although HCMD is engine‑agnostic, we already have a working implementation using two Copilot environments.

Copilot App

• meaning refinement

• meaning‑stabilization

• AML generation

• STS generation

GitHub Copilot

• PSC generation

• implementation generation

• deterministic regeneration

This pair satisfies HCMD’s constraints and serves as the first practical practitioner.The door remains open to any future AI system that can perform the same roles. PSC‑First Training

---

PSC replaces local training.PSC is the training.

PSC acts as the structural bytecode of the domain.It ensures deterministic behavior across cloud AI systems, local LLMs, and offline engines.

PSC is portable, regeneration‑safe, and language‑agnostic. Determinism Doctrine

---

HCMD guarantees:

• deterministic regeneration

• deterministic structure

• deterministic execution

• deterministic offline behavior

Determinism is achieved through Meaning‑Stabilization, AML semantics, STS decomposition, PSC structure, and offline execution models such as InjectML. Domain Narrowing

---

Domain narrowing ensures:

• semantic focus

• structural clarity

• deterministic behavior

• reproducible implementation

Domain narrowing is applied at Meaning‑Stabilization, AML, STS, and PSC. InjectML as Canonical Example

---

InjectML is the reference implementation of HCMD.It demonstrates:

• deterministic offline execution

• PSC‑driven pack loading

• STS‑1, STS‑2, STS‑3, STS‑4 operational pipeline

• embedded runtime behavior

• PairWise as canonical example

InjectML expresses the full HCMD pipeline in a compact, reproducible form. PairWise as Canonical STS Example

---

PairWise demonstrates:

• STS normalization

• STS tokenization

• knowledge pack construction

• deterministic execution

• PSC‑driven runtime behavior

PairWise is included in InjectML and referenced throughout HCMD‑Tutorial. Repository Structure

---

HCMD‑Dev is organized into three layers:

concept → environment → implementationmeaning → operation → execution

concept

Contains the conceptual foundation:

• HCMD Overview

• HCMD Principles

• HCMD Manifesto

• Meaning, AML, STS, PSC, HMD concepts

• HCMD Core Loop

• Domain Narrowing

• Determinism Doctrine

• Conceptual Workflow

• Conceptual Example

environment

Explains how HCMD is executed using:

• Copilot App

• GitHub Copilot

• Windows

Includes operational workflow, example sessions, regeneration guidelines, and PSC‑first training integration.

implementation

Contains HMD schemas, PSC templates, implementation guidelines, and tooling notes.

examples

Contains HCMD examples, including:

• hello‑hcmd (legacy)

• InjectML (planned integration)

• PairWise (planned integration)

Status Update — August 2026

The concept layer has been fully modernized:

• updated HCMD Overview

• updated HCMD Principles

• updated HCMD Manifesto

• updated Meaning, AML, STS, PSC, HMD concepts

• updated Conceptual Workflow

• updated index.md

These documents reflect the 2026 HCMD architecture. Next Steps

---

Planned work:

• update the environment layer

• align the implementation layer

• rewrite HCMD‑Tutorial using InjectML and PairWise

• add PSC‑first training examples

• add regeneration‑safe implementation templates

• integrate InjectML as the canonical HCMD example

About

HCMD is my personal research project, developed independently.I remain affiliated with LBNL and can be reached at H_Nishimura@lbl.gov.