Agent Customization for Copilot CLI

Decision Flow

Primitive	When to Use
Workspace Instructions	Always-on defaults for the repository or your personal CLI environment
File Instructions	Explicit via applyTo, scoped by excludeAgent, or discovered on demand from description
MCP	Connect external systems, APIs, or data sources; use MCP-specific docs for server setup
Hooks	Deterministic shell commands at lifecycle points like preToolUse or postToolUse
Custom Agents	Specialized personas, tool restrictions, or CLI orchestration workflows
Commands / Plugins	Reusable terminal shortcuts, distributed command packs, or bundled CLI customizations
Skills	On-demand workflows with bundled references, scripts, and reusable operational context

Quick Reference

Consult the reference docs for templates, path rules, CLI-only frontmatter, hook schema details, plugin packaging, and troubleshooting steps. If the references are not enough, load the official GitHub Copilot CLI documentation for the relevant primitive.

Type	File	Location	Reference
Workspace Instructions	copilot-instructions.md, AGENTS.md, CLAUDE.md, GEMINI.md	.github/, repo root, or CLI instruction roots	Link
File Instructions	*.instructions.md	.github/instructions/ or roots listed in COPILOT_CUSTOM_INSTRUCTIONS_DIRS	Link
Commands / Plugins	plugin.json, commands/*.md	plugins/cli/<name>/	Link
Hooks	*.json	.github/hooks/	Link
Custom Agents	*.agent.md	.github/agents/ or ~/.copilot/agents/	Link
Skills	SKILL.md	.github/skills/<name>/ or ~/.copilot/skills/<name>/	Link

User-level CLI: ~/.copilot/copilot-instructions.md, ~/.copilot/agents/, and ~/.copilot/skills/ are the main personal discovery locations. COPILOT_CUSTOM_INSTRUCTIONS_DIRS can add extra instruction roots. Hooks are repo-scoped from the current working directory.

Creation Process

If you need to explore or validate existing patterns in the codebase, use a read-only subagent. If the ask-questions tool is available, use it to lock naming, scope, and packaging choices before editing multiple files.

Follow these steps when creating any Copilot CLI customization file.

1. Determine Scope

Ask where the customization belongs:

• Repository: Team-shared defaults and reusable assets -> .github/ or repo-managed plugin directories
• Home directory: Personal, cross-repo CLI customizations -> ~/.copilot/
• Plugin bundle: Shareable packaged commands, skills, hooks, or agents -> plugins/cli/<name>/

2. Choose the Right Primitive

Use the Decision Flow above to pick the narrowest CLI surface that fits the request.

3. Create the File

Create the file directly at the appropriate path:

• Use the location tables in each reference file
• Include the required YAML or JSON structure
• Prefer plugin commands or skills instead of .prompt.md
• Keep SKILL.md lean and push detailed material into references/

4. Validate

After creating:

• Confirm the file is in the correct CLI discovery path
• Verify YAML or JSON syntax
• Check that description is present and keyword-rich
• Confirm CLI-only keys (disable-model-invocation, excludeAgent, hook event names, plugin fields) are valid for the chosen primitive
• Re-publish or reinstall when plugin-backed files change

Edge Cases

Instructions vs Skill? If it should affect most work or all matching files, use instructions. If it is an on-demand workflow, use a skill.

Skill vs Command / Plugin? Use a command or plugin command for a lightweight terminal shortcut. Use a skill when the workflow needs bundled references, scripts, or a larger reusable capability.

Command / Plugin vs Custom Agent? Use a command when the user starts a focused workflow directly. Use a custom agent when you need a persistent persona, isolated tool restrictions, or a reusable specialist that other agents can delegate to.

Skill vs Custom Agent? Use a skill when one workflow can run with the same capabilities throughout. Use a custom agent when you need context isolation or a specialist identity with specific CLI tool access.

Hooks vs Instructions? Instructions guide the model. Hooks enforce behavior with deterministic shell commands. If the behavior must always happen, use a hook.

AGENTS.md vs copilot-instructions.md? In Copilot CLI they are additive, not mutually exclusive. Use both only when their responsibilities are clearly separated.

Guardrails

• Use this skill only for Copilot CLI and terminal-first customization surfaces.
• If the request mentions VS Code prompt files, Chat Customizations UI, Settings Sync, editor prompt recommendations, agents:, or argument-hint:, stop and use copilot-vscode-agent-customization instead.
• Do not teach .prompt.md, VS Code custom-agent schema, or editor-only UX from this skill; keep this skill focused on CLI paths, CLI hooks, CLI agents, and commands/plugins.

Common Pitfalls

Description is the discovery surface. description is how the agent decides whether to load a skill, instruction, or agent. Include trigger phrases and "Use when..." wording.

.prompt.md does not port to CLI. Prompt files are IDE-only. For terminal-first reuse, prefer plugin commands or skills.

Do not copy VS Code-only schema into CLI. agents: and argument-hint: belong to VS Code custom agents, not CLI agents. CLI hooks also use a different schema and lowercase event names.

CLI instruction loading is additive. Copilot CLI can load .github/copilot-instructions.md and AGENTS.md together. Do not teach a fake "choose one" rule as a product constraint.

Plugin changes are not live. After editing plugin files, rebuild or reinstall the plugin so Copilot CLI picks up the new bundle contents.

Avoid broad applyTo defaults. applyTo: "**" loads everywhere and burns context. Use focused globs unless the instruction truly belongs in every request.