If your AI assistant keeps giving generic, unreliable answers, the problem is not the model. It is the missing structure around it. That is exactly what Agent Skills solve.
Modern LLMs are strong enough to handle complex tasks. The real issue is that they are working without stable context, rules, or domain-specific instructions that guide how they should behave in your environment.
This is where Agent Skills change the equation.
Agent skills are instructions and resources that a model can use for a given task to improve its output.
Agent skills are designed to work with modern AI coding assistants and LLM-based agents. While our current implementation leverages GitHub Copilot across tools like VS Code and CLI workflows, the architecture remains provider-agnostic and can be adapted to any compatible AI platform.
You can create your own skills so an agent can perform tasks in a specific way.
You can use both skills and custom instructions to teach Copilot how to work in your repository and how to perform specific tasks.
Agents are massively capable, but often they don't have the needed context for more reliable output. Using skills we can solve this by packaging user-specific context into a portable version an agent can load on demand so agents can be:
Domain expertise: Capture specialized knowledge as reusable instructions and resources.
Repeatable workflows: Turn multi-step tasks into consistent, auditable procedures.
Cross-product reuse: Build a skill once and use it across any skills-compatible agent.
Agents load skills in three stages:
Discovery: At startup, agents load only the name and description of each available skill, just enough to know when it might be relevant.
Activation: When a task matches a skill's description, the agent reads the full SKILL.md instructions into context.
Execution: The agent follows the instructions, optionally executing bundled code or referenced files as needed.
Full instructions load only when a task calls for them, so agents can keep many skills on hand with only a small context footprint.
A common pitfall in skill creation is asking an LLM to generate a skill without providing domain-specific context, relying solely on the LLM’s general training knowledge. The result is vague, generic procedures (“handle errors appropriately,” “follow best practices for authentication”) rather than the specific API patterns, edge cases, and project conventions that make a skill valuable.
Effective skills are grounded in real expertise. The key is feeding domain-specific context into the creation process.
Project skills, stored in your repository (.github/skills, .claude/skills, or .agents/skills)
Personal skills, stored in your home directory and shared across projects (~/.copilot/skills or ~/.agents/skills)
When performing tasks, Copilot will decide when to use your skills based on your prompt and the skill's description.
When Copilot chooses to use a skill, the SKILL.md file will be injected in the agent's context, giving the agent access to your instructions. It can then follow those instructions and use any scripts or examples you may have included in the skill's directory.
For Copilot code review on GitHub:
If you want to ensure Copilot code review uses a skill, place it in a review-focused directory like code-review
.
Skills in .github/skills
can also be picked up automatically when relevant.
To create an agent skill, you write a SKILL.md
file and optionally add supporting files like scripts or examples.
Create a skills directory in one of these locations:
Project skills:
.github/skills
.claude/skills
.agents/skills
Personal skills:
~/.copilot/skills
~/.agents/skills
Each skill must live in its own folder:
.github/skills/webapp-testing
Folder names must be lowercase and use hyphens.
Inside it, create SKILL.md
.
Important
Skill files must be named
SKILL.md
.It must include:
YAML frontmatter:
- name (required): lowercase identifier matching folder name
- description (required): when the skill should be used
- license (optional)
Markdown body:
- Instructions
- Examples
- Guidelines for the agent
---
name: resume-update
description: Use this skill when the user asks to update, rewrite, or tailor a resume for a job application.
---
## Instructions
- Rewrite resumes based on job description
- Improve clarity and structure
- Use concise bullet points
- Highlight relevant skills and achievements
- Keep it ATS-friendly
## Example
User: Update my resume for a .NET backend role
Assistant: Returns a rewritten resume emphasizing .NET, APIs, and backend experience.
Just ask normally:
Help me update my resume for a .NET backend role
If skill matches description → it auto-loads
/resume-update
/webapp-testing
/github-actions-debugging
You can also add extra context:
/resume-update for senior backend role
---
name: skill-name
description: When and why this skill should be used
---
Then markdown instructions:
### Instructions
- Step 1
- Step 2
- Step 3
We recommend using custom instructions for simple instructions relevant to almost every task (for example information about your repository's coding standards), and skills for more detailed instructions that Copilot should only access when relevant.