What problem does it solve? Designing custom Copilot agents often produces vague descriptions, bloated tool lists, and missing constraints, which leads to agents that trigger incorrectly or underperform. This Skill provides a structured procedure for defining agent scope, selecting minimal tools, assigning skills, and validating the resulting .agent.md file. ## Core Features & Use Cases - Scoped Agent Design: Guides definition of a single role, explicit exclusions, delegation rules, and output format before writing any file. - Minimal Tool Selection: Provides decision tables for choosing only the tools (read, edit, search, execute, web, agent) the agent's role requires, with Swiss-army anti-pattern detection. - Validation & Sync: Includes shell-based validation of frontmatter fields, filename/name matching, skill path resolution, and a bootstrap step to symlink agents into ~/.copilot/ and VS Code prompt directories. - Use Case: You need a read-only research subagent for a multi-agent workflow. Use this Skill to define its persona, restrict tools to [read, search, web], write the .agent.md, validate it, and sync it into your active Copilot environment. ## Quick Start Ask the agent to create a new .agent.md for a specialized subagent, providing its name, purpose, required tools, and the skills it should load.
