agent-design

name	agent-design
description	Use when designing Copilot agents, authoring agent definitions, creating skill folders, or scaffolding instruction files. Provides templates and conventions for the agent ecosystem.
compatibility	["VS Code","Cursor","Windsurf","Claude Code"]
metadata	{"category":"Uncategorized","tags":["uncategorized"],"maturity":"beta","audience":["developers"]}
allowed-tools	["bash","git","grep","find"]

Agent Design Skill

Use this skill when the goal is to design, scaffold, or author Copilot agent definitions, skill folders, or instruction files for a shared customization repository.

Template Index

Template	Purpose	Path
Agent definition	Scaffold a new .agent.md file with frontmatter and standard sections	agent-template.md
Skill folder	Scaffold a new skill with SKILL.md and supporting structure	skill-template.md
Instruction file	Scaffold a reusable instruction file	instruction-template.md

When to Use Each Template

• Agent template — you need a new autonomous agent with its own workflow, tool grants, and domain guidance.
• Skill template — you need a reusable knowledge module that agents can reference but that does not run on its own.
• Instruction template — you need a scoped directive that applies to a file, folder, or repository without requiring a full agent or skill.

Agent Pairing

This skill is designed to work with:

• agent-designer agent (agents/agent-designer.agent.md) — the primary consumer of these templates. The agent-designer invokes this skill to scaffold new agents and skills.
• prompt-engineer agent (agents/prompt-engineer.agent.md) — optimizes the instruction text produced by these templates.

Conventions

• Agent filenames: <kebab-case-name>.agent.md
• Skill folders: skills/<kebab-case-name>/SKILL.md
• Instruction files: instructions/<kebab-case-name>.instructions.md
• All files use YAML frontmatter fenced by --- as the first content.
• Descriptions must include trigger phrases for discovery.

Guardrails

• Do not create an agent when a skill or instruction would suffice.
• Do not create a skill that duplicates an existing one — check the inventory first.
• Keep each primitive (agent, skill, instruction) scoped to a single responsibility.
• Validate all frontmatter with a YAML linter before committing.