What problem does it solve? Building a custom GitHub Copilot agent is easy to get wrong: vague descriptions make agents invisible in the picker, and the tools: field silently restricts capabilities when engineers think it declares them. This Skill walks a learner through creating, testing, and iterating a focused .agent.md file so the agent is discoverable, correctly scoped, and behaves as intended. ## Core Features & Use Cases - Agent authoring workflow: Step-by-step phases for writing .github/agents/<name>.agent.md with valid YAML frontmatter, a behavioral persona, explicit scope guardrails, and anti-patterns. - The tools: field trap: Teaches that tools: is restrictive, not declarative, and that omitting it inherits the full default Copilot toolset unless a security or scope reason exists. - Fresh-session testing: Verifies the agent appears in the picker, follows its persona on test prompts, and has the tools it needs, with results captured in a test log. - Use Case: A learner builds an api-design-reviewer agent that critiques REST endpoints against RFC 7231 and REST conventions, tests it in a new chat, then refines the persona based on real output. ## Quick Start Ask the Mentor to start the ghc-custom-agents project and build an api-design-reviewer custom agent in a fresh Copilot chat session.
