| name | vscode-customization |
| description | VS Code customization: agent configuration, skills, hooks, prompts, workspace instructions, file patterns, and applyTo rules. Use when creating or editing .instructions.md, .prompt.md, .agent.md, SKILL.md, AGENTS.md, copilot-instructions.md, or configuring VS Code agent behavior. Triggers: customize copilot, agent instructions, slash commands, context patterns. |
| user-invocable | true |
VS Code Agent Customization
When to Use This Skill
- Creating or editing agent skills (SKILL.md)
- Setting up workspace instructions (copilot-instructions.md)
- Creating custom agents (.agent.md)
- Configuring file-specific instructions (*.instructions.md)
- Setting up hooks for automation
- Creating prompt templates (*.prompt.md)
- Configuring MCP servers
- Debugging why instructions aren't being applied
Skill Locations
| Scope | Path |
|---|
| Project skills | .github/skills/<name>/ |
| Project agents | .github/agents/<name>/ |
| User prompts | {{VSCODE_USER_PROMPTS_FOLDER}}/ |
File Types
SKILL.md (Skills)
Bundled on-demand workflows with scripts and assets.
---
name: skill-name
description: 'Clear description of when to use'
user-invocable: true
disable-model-invocation: false
---
.instructions.md (File Instructions)
Apply to specific files via applyTo patterns.
---
name: file-instructions
applyTo:
- "**/*.cs"
- "**/*.json"
---
.prompt.md (Prompts)
Parameterized templates for specific tasks.
.agent.md (Custom Agents)
Subagents with specialized tool restrictions and prompts.
copilot-instructions.md (Workspace Instructions)
Always-on instructions for the entire workspace.
applyTo Patterns
Glob patterns that control when file instructions activate:
| Pattern | Matches |
|---|
**/*.cs | All C# files |
**/*.csproj | Project files only |
src/**/*.ts | TS files in src folder |
**/test/** | Anything in test folders |
Creating Custom Agents
---
name: code-reviewer
description: 'Specialized code review agent'
tools:
allow:
- read_file
- glob
- grep
deny:
- bash
- write_file
---
Hooks
Automation at agent lifecycle points (stored in .github/hooks/).
Best Practices
- Progressive loading: Keep SKILL.md under 500 lines
- Keyword-rich descriptions: Enable discovery by the model
- Self-contained skills: Include all procedural knowledge
- Consistent naming: lowercase-with-hyphens
Anti-Patterns
- Vague descriptions ("A helpful skill")
- Monolithic files instead of references
- Name mismatch between folder and SKILL.md
- Missing step-by-step procedures
Resources