name: creating-agents description: Step-by-step guidance for creating custom Copilot agents in VS Code, including YAML frontmatter, file structure, handoffs, and best practices. Use when you want to define a new persona, workflow, or specialized agent for your workspace or team. argument-hint: "[agent purpose] [desired tools or workflow]" user-invokable: true disable-model-invocation: false

Creating Custom Copilot Agents in VS Code

When to Use This Skill

• When you want to create a new custom agent (persona, workflow, or specialized role) for Copilot Chat in VS Code.
• When you need to define agent instructions, available tools, handoffs, or agent-specific workflows.
• When you want to share or maintain agents for your team or workspace.

Core Concepts

• Custom agents let you tailor Copilot's behavior for specific tasks, roles, or workflows.
• Agents are defined in .agent.md files with YAML frontmatter and Markdown instructions.
• You can control available tools, handoffs, and model selection per agent.
• Agents can be workspace-specific, user-profile, or organization-wide.

How-To: Create a Custom Agent

1. Create the Agent File

• Place your agent file in .github/agents/ (workspace), your user profile, or organization folder.
• Name it with a .agent.md extension (e.g., my-reviewer.agent.md).

2. Add YAML Frontmatter

Example:

---
name: my-reviewer
# Optional: If omitted, file name is used
model: GPT-4.1 (copilot)
description: Code review agent for security and quality. Suggests improvements and flags issues.
tools: [search, fetch, read, apply_patch, runTests]
argument-hint: "[file or code] [review focus]"
user-invokable: true
disable-model-invocation: false
target: vscode
handoffs:
  - label: Handoff to Implementation
    agent: implementation
    prompt: Implement the changes suggested above.
    send: false
    model: GPT-4.1 (copilot)
---

• Fields:
  • name: Unique agent name (lowercase, hyphens, max 64 chars)
  • description: What the agent does and when to use it (max 1024 chars)
  • tools: List of allowed tools (e.g., search, fetch, read, apply_patch, runTests)
  • argument-hint: Input hint for chat
  • model: (Optional) Preferred model(s)
  • user-invokable: Show in agent picker (default true)
  • disable-model-invocation: Prevent as subagent (default false)
  • target: vscode or github-copilot
  • handoffs: (Optional) List of next-step actions

3. Write Agent Instructions (Markdown Body)

• Below the YAML, add persona, workflow, and usage instructions.
• Use sections for principles, decision-making, communication style, etc.
• Reference tools with #tool:<tool-name> if needed.

4. (Optional) Add Handoffs

• Use handoffs in YAML to guide users to the next agent or workflow step.
• Each handoff includes a label, target agent, prompt, and options.

5. Test and Share

• In VS Code, use the agents dropdown or /agents to manage and test agents.
• Ensure your agent appears and loads as expected.
• Share agent files in your repo, user profile, or organization for team use.

Patterns & Templates

• Use clear, specific descriptions and argument hints.
• Limit tool access to only what's needed for the agent's purpose.
• Use handoffs for multi-step workflows (e.g., Plan → Implement → Review).
• Keep instructions concise and actionable.

Anti-Patterns / When NOT to Use

• Don't use agents for one-off instructions (use custom instructions instead).
• Avoid vague descriptions or tool lists (be explicit for best results).
• Don't overload a single agent with unrelated roles or workflows.

References

• VS Code Copilot Custom Agents
• VS Code Agent Skills Docs
• agentskills.io
• github/awesome-copilot
• anthropics/skills