Skip to main content

In this article

Copilot CLI Plugin

Install the complete HVE Core component set as a Copilot CLI plugin for terminal-based AI-assisted development workflows.

Prerequisites

  • GitHub Copilot CLI installed and authenticated

Register hve-core as a Plugin Marketplace

Choose a registration that matches the content you need.

Register the ref-less development tip:

copilot plugin marketplace add microsoft/hve-core

Register a moving reviewed release channel:

copilot plugin marketplace add microsoft/hve-core#release/prerelease
copilot plugin marketplace add microsoft/hve-core#release/stable

Register an immutable channel tag:

copilot plugin marketplace add microsoft/hve-core#prerelease-v<version>
copilot plugin marketplace add microsoft/hve-core#v<version>

main is the development tip and its catalog entries omit source.ref. release/prerelease and release/stable are moving registrations that resolve their current reviewed branch catalog. Each branch catalog pins every entry to its corresponding exact channel tag. Exact-tag registrations freeze both the catalog selection and plugin source tag.

A published channel release provides release assurance for its exact tag, including release gates, SBOMs, attestations, provenance verification, and the configured publication path. The development tip does not provide that published-release assurance.

Browse Available Plugins

Type /plugin in a Copilot CLI chat session to browse available plugins.

Install a Plugin

Install hve-core from the registered marketplace through /plugin. The plugin includes the complete active HVE Core component set, including the Research, Plan, Implement, Review lifecycle.

copilot plugin install hve-core@hve-core

Update an Installed Plugin

Marketplace refresh and installed-plugin update are distinct actions. For a moving registration, refresh the catalog before requesting a plugin update:

copilot plugin marketplace update hve-core
copilot plugin update hve-core@hve-core

Switching registrations can require removing and re-adding the marketplace. Do not assume how the client handles duplicate same-name registrations; use the behavior supported by your Copilot CLI version.

Use the migration guide if you previously registered or installed a retired package identity.

Plugin Contents

Each plugin includes:

ComponentCLI DiscoveryDescription
AgentsYesCustom chat agents for specialized workflows
CommandsYesTask prompts accessible via the CLI
SkillsYesSelf-contained skill packages
InstructionsNoIncluded for #file: references, not auto-applied

Each plugin is a self-contained tree of regular files and real directories. Artifacts are copied from the source repository during generation, so a plugin installs the same way on every operating system and needs no symbolic link support.

Limitations

Instructions are not auto-applied from plugins

The Copilot CLI plugin spec recognizes agents, skills, commands, hooks, mcpServers, and lspServers as component types. There is no instructions component type.

The CLI loads path-specific instructions exclusively from .github/instructions/**/*.instructions.md in the project repo. Instruction files in plugin directories are not auto-applied via applyTo pattern matching.

Instruction files are still included in plugin output because agents and prompts reference them via #file: directives. Those cross-file references resolve correctly within the plugin directory tree. The difference is between explicit inclusion (an agent pulls in instruction content at execution time) and automatic application (the CLI matches applyTo patterns against the files you are editing).

For full path-specific instruction behavior, copy instruction files into your project's .github/instructions/ directory.

Other limitations

  • Skills require skill-compatible agent environments

Using Agents After Installation

After installing a plugin, agents and named commands are available in your CLI session.

Named Commands vs Agent Mode

CLI plugins provide two distinct interaction patterns:

ModeCommandBehavior
Named Command/git-commitExecutes a predefined workflow, then returns to default mode
Skill/rpi-researchActivates one reusable RPI phase capability
Agent Mode/agent RPI AgentSwitches to the coordinated RPI lifecycle

Named commands (prompts) run a specific workflow and produce structured output. Agent mode enables freeform conversation with a specialized agent until you exit.

IMPORTANT

The CLI does not switch to a custom agent on behalf of an agent-bound prompt. Select RPI Agent when you want lifecycle coordination, or invoke a direct phase skill such as /rpi-research:

/agent RPI Agent
Research API authentication patterns before deciding whether planning is ready.

Prompts that do not require an agent context (e.g., /git-commit, /git-merge) work directly from the default mode.

Example: Research Workflow

Invoke the Research phase skill directly:

> /rpi-research topic="API authentication patterns"
[Skill executes the research workflow and creates a research document]

Continue with follow-up questions in the same session:

> What are common API authentication patterns for REST APIs?
[Research conversation continues]
> How do OAuth2 and API keys compare for microservices?
[Follow-up within same agent context]

Available Agents

After installing the hve-core plugin, these agents are available via /agent <name>:

  • RPI Agent - coordinates Research, Plan, Implement, Review, and Follow-up
  • Documentation - audits, authors, and validates documentation

For the complete list, run /help in a CLI session to see all available commands and agents.

When to Use Each Mode

  • Use named commands (/git-commit-message, /git-merge) directly from default mode for workflows that do not require a custom agent.
  • Use direct skills (/rpi-research, /rpi-plan, /rpi-implement, /rpi-review) for one bounded RPI responsibility.
  • Use agent mode with /agent RPI Agent for lifecycle coordination.
  • Stay in agent mode for exploratory conversations, follow-up questions, or tasks that don't fit a predefined prompt.

🤖 Crafted with precision by ✨Copilot following brilliant human instruction, then carefully refined by our team of discerning human reviewers.