The Problem

Teams using GitHub Copilot get inconsistent results: prompts are ad-hoc, coding standards are not enforced, and there is no repeatable workflow for planning, implementing, and reviewing AI-assisted work. HVE Core addresses this by packaging agents, prompts, instructions, and skills into a structured workflow system that makes AI-assisted development repeatable and standards-aligned.

What This Does

This repo is a content library, not a runtime application. It contains 2,625 files, almost entirely Markdown (1,174 files), YAML (278), and Python (245). The .github/ directory holds the substance: specialized Copilot agents (.github/agents/), reusable prompts (.github/prompts/), and coding-standard instructions (.github/instructions/) covering security, privacy, accessibility, RAI planning, project planning, and design thinking. The evals/ directory (139 files) contains test harnesses, and docs/ (556 files) hosts a Docusaurus site.

The README explicitly warns this is an opinionated, rapidly evolving framework best treated as a source of patterns rather than a stable dependency. It is a fork of microsoft/hve-core (1,384 stars); this fork has 0 stars.

How It Is Wired

There is no central call graph or hub. Control flow is delegated to GitHub Copilot itself: Copilot loads .agent.md files and .github/copilot-instructions.md as context, then routes user requests through the selected agent. The primary entry point is the RPI Agent (.github/agents/hve-core/rpi-agent.agent.md), which the README instructs users to invoke via /rpi in Copilot Chat. That agent coordinates sub-agents (rpi-planner, rpi-researcher) and delegates to domain planners and reviewers (.github/agents/security/security-planner.agent.md, .github/agents/project-planning/backlog-manager.agent.md, etc.).

The wiring for this repository has not been mapped beyond these agent-to-subagent relationships. There is no executable main path; the only Python entry points (__main__.py files) live inside individual skills such as .github/skills/accessibility/accessibility/scripts/runtime_a11y/. Each skill is self-contained with its own pyproject.toml and, in the accessibility case, a package.json for a runtime checker. The extension/ directory (6 files) contains the VS Code marketplace extension packaging.

How To Use It

Setup — install the VS Code extension from the marketplace link in the README, or register the Copilot CLI plugin:

copilot plugin marketplace add microsoft/hve-core
copilot plugin install hve-core@hve-core

Configuration — .github/copilot-instructions.md is the root instruction file Copilot loads. Per-skill dependencies use pyproject.toml (Python) or package.json (Node). A .devcontainer/Dockerfile provides a development environment.

Running it — open Copilot Chat, select the RPI Agent from the agent picker, or run /rpi and describe the task. No CLI or server exists to start.

Real-World Use

Onboarding a new project: run /rpi describing the task. The RPI Agent plans, then routes to domain agents — security-planner for threat modeling, backlog-manager for work items, code-review for PR standards. The evals/ directory provides a harness to validate agent outputs against expected behavior.

Code Health & Issues

Deep static analysis has not run for this repo. Structural observations:

  • Low — Fork carries no stars or community signal; upstream microsoft/hve-core is the canonical source.
  • Low — README itself cautions against production dependency: interfaces and workflows change without backward compatibility.

CI (GitHub Actions), 820 test files, license, and lockfiles are present. No configuration secrets or obvious structural red flags were found.

The Bottom Line

A well-organized pattern library for Copilot-driven SDLC workflows, with real CI and test infrastructure. Use it as reference material or fork it to build your own agentic workflow — do not treat it as a stable runtime dependency. The upstream Microsoft repo is the better base if you want community updates.