Creating Collections

A collection groups prompts, instructions, chat modes, and agents around a role or skill set.

Quick Start

Create a Source or an additional Bundle

Project Structure

my-collection/
├── collections/
│   └── my-collection.collection.yml    # Manifest
├── prompts/
│   └── *.prompt.md
├── instructions/
│   └── *.instructions.md
├── agents/
│   └── *.agent.md
└── README.md

Collection Manifest

id: my-collection                    # lowercase, numbers, hyphens
name: My Collection
description: What this collection does
tags: [productivity, coding]
items:
  - path: prompts/task-helper.prompt.md
    kind: prompt                     # prompt | instruction | chat-mode | agent
  - path: instructions/standards.instructions.md
    kind: instruction

See Collection Schema for full reference.

Resource Types

Type	Extension	Purpose
Prompt	.prompt.md	Reusable prompt templates
Instruction	.instructions.md	System guidelines
Agent	.agent.md	Autonomous task patterns

Scaffolding a Project

Ctrl+Shift+P → "AI Primitives Hub: Scaffold Project"

1. Select project type

Option	Description
GitHub	GitHub-based prompt library with CI/CD workflows
APM Package	Distributable prompt package (apm.yml)
Agent Skill	Single skill with SKILL.md

2. Select target directory

A folder picker opens. The scaffolded files are written into this directory.

3. Enter project details

All types ask for:

Prompt	Default
Project name	example
GitHub Actions runner type	ubuntu-latest, self-hosted, or custom label

Then type-specific questions follow:

GitHub — organization details:

Prompt	Example
Author name	Your Name or Team Name
GitHub organization/username	your-org
Organization name (for LICENSE)	Your Organization
Security contact email	security@yourorg.com
Legal contact email	legal@yourorg.com
Organization policy URL (optional)	https://yourorg.com/policies

APM Package / Agent Skill — project metadata:

Prompt	Default
Description	(none)
Author name	$USER env variable
Tags (comma separated)	APM: apm, prompts / Skill: skill, prompts

4. What gets created

GitHub — a full repository structure including collections/, prompts/, instructions/, agents/, skills/, GitHub Actions workflows, community files (LICENSE, CONTRIBUTING, SECURITY, CODE_OF_CONDUCT), and VS Code workspace settings.

APM Package — apm.yml, package.json, example prompts/instructions under .apm/, and a validation workflow.

Agent Skill — a <project-name>/ subdirectory containing SKILL.md, README.md, example scripts, and references//assets/ directories.

After scaffolding, npm install runs automatically and you're offered to open the new folder.

Skill wizard (existing projects)

If you pick Agent Skill and the workspace already contains a collections/ or skills/ directory, a simplified wizard runs instead:

1. Enter skill name (lowercase, numbers, hyphens)
2. Enter skill description
3. Optionally select a collection to add the skill to
4. The skill is created under skills/<name>/ and optionally linked in the collection manifest

Testing Locally

1. Add local source: Ctrl+Shift+P → "Add Source" → local-awesome-copilot
2. Enter path to your collection directory
3. Bundles appear in Registry Explorer

Validating

• Validation

See Also

• Collection Schema — Full YAML reference
• Publishing — Distribute your collections