Skip to main content
Skills are packaged instructions that extend what your agents can do. Both the Claude Code and GitHub Copilot runtimes in Jinzo support skills — each with its own format, but the same idea: teach the agent domain-specific behavior it can invoke automatically.

​
Claude Code Skills

Claude Code skills are SKILL.md files with YAML frontmatter and Markdown instructions. Claude discovers them at startup and autonomously invokes the right skill based on context.

​
Skill Locations

LocationScope
.jinzo/skills/Project skills — shared with your team via git
~/.claude/skills/User skills — personal, available across all projects

​
Creating a Skill

Create a directory with a SKILL.md file inside: 
.jinzo/skills/code-review/
└── SKILL.md
A SKILL.md file has YAML frontmatter followed by Markdown instructions: 
---
description: Reviews code changes for security vulnerabilities and best practices
allowed-tools: ["Read", "Grep", "Glob"]
---
# Code Review
When reviewing code, check for:
- SQL injection and XSS vulnerabilities
- Hardcoded credentials or secrets
- Missing input validation
- Error handling gaps
Provide findings as a numbered list with severity ratings.
The description field is what Claude uses to decide when to invoke the skill. Make it specific.

​
Invoking Skills

Claude invokes skills automatically when a user request matches the skill’s description. You can also trigger them explicitly with slash commands: 
/code-review

​
Copilot Skills

Copilot skills are directories containing a skill.json manifest, prompt files, and optional tool definitions.

​
Structure

skills/
└── code-review/
    ├── skill.json
    └── prompts/
        └── system.md

​
skill.json

{
  "name": "code-review",
  "displayName": "Code Review Assistant",
  "description": "Specialized code review capabilities",
  "version": "1.0.0",
  "prompts": ["prompts/system.md"]
}

​
Loading Skills

Skills are loaded from configured directories when a Copilot session starts. You can disable specific skills by name if needed.

​
Tips

  • Write clear descriptions — the agent uses them to decide when to invoke a skill
  • Limit tool access — only grant the tools a skill actually needs
  • Keep skills focused — one skill per domain works better than a catch-all
  • Version project skills — commit .claude/skills/ so your whole team benefits