Skill: copilot-cli

Guidance for installing the GitHub Copilot CLI on Debian/Ubuntu and using it with custom agents via command-line options.

When to Use

• When executing programmatic requests against GitHub Copilot via the terminal.
• To orchestrate specialized subagents using the --agent flag.
• When installing or authenticating the Copilot CLI in a CI/CD or local environment.

When Not to Use

• When trying to use standard gh commands for repo management (Copilot CLI is an extension specifically for AI assistance).
• Inside highly restricted environments without internet access or where external API calls to api.github.com are blocked.
• For generating extremely large, multi-file architectural scaffolds where a dedicated repository generation tool is more appropriate.

Common Pitfalls

• Interactive Prompts in Scripts: Using the raw copilot command in a CI script without the -s (silent) or programmatic flags, causing the pipeline to hang waiting for user input.
• Agent Name Clashes: Creating a custom agent that shares a name with a built-in alias, leading to the wrong prompt instructions being loaded.
• Untrusted Directories: Failing to pre-approve or trust the working directory, causing Copilot CLI to block execution and demand a manual terminal confirmation.

Core Process

1. Install CLI: Use npm install -g @github/copilot (recommended), curl -fsSL https://gh.io/copilot-install | bash (install script), or snap install copilot-cli on Debian/Ubuntu.
2. Authentication: Use copilot login or set COPILOT_GITHUB_TOKEN. Fine-grained PATs require the Copilot Requests permission.
3. Discover Usage: Run copilot --help for standard command usage options. Alternatively, load the copilot-docs skill for detailed overview.
4. Agent Selection: Use the --agent flag to target specialized .agent.md files. For details, load the copilot-docs skill.
5. Command Execution: Provide the explicit instruction string via the --prompt or -p flag. Use -s (silent) in scripts to capture output. Manage tools via tool flags.
6. Task Steering: Control outputs dynamically or remotely via workflow configuration. For detailed guides on steering, load the copilot-docs skill.
7. Session Management: Validate and trace logic using session commands. For comprehensive session management, load the copilot-docs skill.
8. Pull Requests: For collaborative tasks, leverage PR-focused commands described in the copilot-docs skill.

Core Principles

• Programmatic Execution: Avoid interactive slash commands (like /agent) in scripts. Always use explicit programmatic flags (--agent and --prompt).
• Context Management: Utilize custom subagents to offload specific tasks, ensuring the main agent's context window remains uncluttered. In scripts, keep prompts narrowly scoped, pass context explicitly with --prompt, and start a new copilot invocation when you need a fresh or reduced context window.
• Agent Resolution: If custom agents share a name, the resolution order is: User (~/.copilot/agents/) > Project (.github/agents/) > Organization (.github-private/agents/).
• Trusted Directories: Copilot CLI requires confirmation to trust the working directory. Permanent trust is stored in ~/.copilot/config.json.

Commands / Usage Patterns

Installation & Authentication

# Recommended (requires Node.js 22+)
npm install -g @github/copilot

# Install script (macOS/Linux)
curl -fsSL https://gh.io/copilot-install | bash

# Authenticate
copilot login

Security & Permissions

# Preferred: grant only the specific permissions needed for the task
copilot --allow-tool='shell(git)' --deny-tool='shell(rm)' --prompt "Clean repo"

# Only use YOLO mode after the user has explicitly confirmed unrestricted access
copilot --yolo --prompt "Perform complex task"

Programmatic Custom Agent Execution Specify the custom agent file name (excluding the .agent.md extension) and the exact instruction prompt.

copilot --agent security-auditor --prompt "Check <target-file>"

Related Skills

• copilot-docs: Detailed references on automating, customizing, and advanced usage patterns for Copilot CLI.
• gh-agent-task: Use for managing preview agent tasks on repositories and pull requests.