What are Agents, Skills, and Instructions

Understand the primary customization primitives that extend GitHub Copilot for specific workflows.

7 minutes Updated: 2026-02-26

Building great experiences with GitHub Copilot starts with understanding the core primitives that shape how Copilot behaves in different contexts. This article clarifies what each artifact does, how it is packaged inside this repository, and when to use it.

Agents

Agents are configuration files (*.agent.md) that describe:

  • The tasks they specialize in (for example, “Terraform Expert” or “LaunchDarkly Flag Manager”).
  • Which tools or MCP servers they can invoke.
  • Optional instructions that guide the conversation style or guardrails.

When you assign an issue to Copilot or open the Agents panel in VS Code, these configurations let you swap in a specialized assistant. Each agent in this repo lives under agents/ and includes metadata about the tools it depends on.

When to reach for an agent

  • You have a recurring workflow that benefits from deep tooling integrations.
  • You want Copilot to proactively execute commands or fetch context via MCP.
  • You need persona-level guardrails that persist throughout a coding session.

Skills

Skills are self-contained folders that package reusable capabilities for GitHub Copilot. Each skill lives in its own directory and contains a SKILL.md file along with optional bundled assets such as reference documents, templates, and scripts.

A SKILL.md defines:

  • A name (used as a /command in VS Code Chat and for agent discovery).
  • A description that tells agents and users when the skill is relevant.
  • Detailed instructions for how the skill should be executed.
  • References to any bundled assets the skill needs.

Skills follow the open Agent Skills specification, making them portable across coding agent systems beyond GitHub Copilot.

Why skills over prompts

Skills replace the earlier prompt file (*.prompt.md) pattern and offer several advantages:

  • Agent discovery: Skills include extended frontmatter that lets agents find and invoke them automatically—prompts could only be triggered manually via a slash command.
  • Richer context: Skills can bundle reference files, scripts, templates, and other assets alongside their instructions, giving the AI much more to work with.
  • Cross-platform portability: The Agent Skills specification is supported across multiple coding agent systems, so your investment travels with you.
  • Slash command support: Like prompts, skills can still be invoked via /command in VS Code Chat.

When to reach for a skill

  • You want to standardize how Copilot responds to a recurring task.
  • You need bundled resources (templates, schemas, scripts) to complete the task.
  • You want agents to discover and invoke the capability automatically.
  • You prefer to drive the conversation, but with guardrails and rich context.

Instructions

Instructions (*.instructions.md) provide background context that Copilot reads whenever it works on matching files. They often contain:

  • Coding standards or style guides (naming conventions, testing strategy).
  • Framework-specific hints (Angular best practices, .NET analyzers to suppress).
  • Repository-specific rules (“never commit secrets”, “feature flags must live in flags/”).

Instructions sit under instructions/ and can be scoped globally, per language, or per directory using glob patterns. They help Copilot align with your engineering playbook automatically.

When to reach for instructions

  • You need persistent guidance that applies across many sessions.
  • You are codifying architecture decisions or compliance requirements.
  • You want Copilot to understand patterns without manually pasting context.

How the artifacts work together

Think of these artifacts as complementary layers:

  1. Instructions lay the groundwork with long-lived guardrails.
  2. Skills let you trigger rich, reusable workflows on demand—and let agents discover those workflows automatically.
  3. Agents bring the most opinionated behavior, bundling tools and instructions into a single persona.

By combining all three, teams can achieve:

  • Consistent onboarding for new developers.
  • Repeatable operations tasks with reduced context switching.
  • Tailored experiences for specialized domains (security, infrastructure, data science, etc.).

Next steps

  • Explore the rest of the Fundamentals track for deeper dives on chat modes, collections, and MCP servers.
  • Browse the Awesome Agents, Skills, and Instructions directories for inspiration.
  • Try generating your own artifacts, then add them to the repo to keep the Learning Hub evolving.