Manage Agent Templates
Agents are specialized AI personas with scoped tools and permissions that run in isolated contexts.
Important: Agents are supported by Claude and Copilot only. Cursor does not support agents — consider using a skill with forkContext: true as an alternative for Cursor.
Finding Existing Agents
List files in .universal-ai-config/agents/ to discover existing agent templates. Read their frontmatter to understand each agent's purpose and capabilities.
Deciding What to Do
- •Create new: when you need a new specialized persona with distinct capabilities
- •Update existing: when an agent needs adjusted tools, permissions, or system prompt
- •Delete: when an agent is no longer needed
Creating a New Agent
- •Create a
.mdfile in.universal-ai-config/agents/with a descriptive name (e.g.code-reviewer.md) - •Add YAML frontmatter with at minimum
nameanddescription - •Write the agent's system prompt as the body
Frontmatter Fields
| Field | Description |
|---|---|
name | Agent identifier (lowercase with hyphens) |
description | When to delegate to this agent (helps AI decide when to use it) |
tools | Tools this agent can use (omit to inherit all) |
disallowedTools | Tools to explicitly deny |
permissionMode | Permission level: default, acceptEdits, dontAsk, bypassPermissions, plan |
skills | Skills to preload into the agent's context |
memory | Persistent memory scope: user, project, or local |
model | AI model to use (e.g. sonnet, opus, haiku) |
target | Copilot-specific: scope to vscode or github-copilot |
mcpServers | MCP server configurations |
handoffs | Copilot-specific: other agents to hand off to |
Example
markdown
--- name: security-reviewer description: Reviews code for security vulnerabilities and best practices. Use proactively after code changes. tools: ["Read", "Grep", "Glob", "Bash"] model: sonnet --- You are a security-focused code reviewer. When invoked: 1. Identify the changed files 2. Check for common vulnerabilities (injection, XSS, auth issues) 3. Review dependency usage for known CVEs 4. Report findings by severity (critical, warning, info)
After Changes
Run uac generate to regenerate target-specific config files and verify the output.