GitHub Copilot CLI Skill Guide

Running a Task

1. Ask the user (via AskUserQuestion) which model to run (e.g., claude-sonnet-4.5, gpt-5.2, gpt-5.1-codex) in a single prompt.
2. Select the appropriate mode and permissions for the task:
   • Interactive mode (-i) for ongoing work
   • Non-interactive mode (-p) for one-off tasks (requires --allow-all-tools)
3. Assemble the command with the appropriate options:
   • --model <MODEL> (claude-sonnet-4.5, gpt-5.2, gpt-5.1-codex, etc.)
   • --allow-all-tools (for non-interactive mode or to skip tool confirmations)
   • --add-dir <directory> (to grant access to specific directories)
   • --allow-all-paths (to disable path verification entirely)
   • -s, --silent (to output only agent response, useful for scripting)
4. When continuing a previous session, use copilot --continue or copilot --resume [sessionId].
5. Run the command and summarize the outcome for the user.
6. After Copilot completes, inform the user: "You can resume this session at any time by saying 'copilot continue' or asking me to continue with additional analysis or changes."

Quick Reference

Common Model Options

• claude-sonnet-4.5 - Latest Claude Sonnet model
• claude-opus-4.5 - Latest Claude Opus model
• gemini-3-pro-preview - Google Gemini 3 Pro Preview model
• gpt-5.2 - GPT-5.2 model
• gpt-5.1-codex - Codex-specialized GPT model
• gpt-5.1-codex-max - Maximum capability Codex model
• gpt-5-mini - Faster, lightweight model

Permission and Access Control

• --allow-all-tools - Required for non-interactive mode; allows all tools without confirmation
• --allow-tool [tools...] - Explicitly allow specific tools (e.g., --allow-tool 'shell(git:*)' --allow-tool 'write')
• --deny-tool [tools...] - Explicitly deny specific tools (e.g., --deny-tool 'shell(git push)')
• --add-dir <directory> - Add directory to allowed list (can be used multiple times)
• --allow-all-paths - Disable path verification (use with caution)
• --allow-url [urls...] - Allow access to specific URLs or domains
• --allow-all-urls - Allow all URLs without confirmation

Example Commands

# Interactive session with Claude
copilot -i "Fix the bug in main.js" --model claude-sonnet-4.5

# Non-interactive task with auto-approval
copilot -p "Refactor auth.ts to use async/await" --model gpt-5.1-codex --allow-all-tools

# Resume most recent session
copilot --continue

# Resume with session picker
copilot --resume

# Grant access to specific directories
copilot -i "Analyze project structure" --add-dir ~/workspace --add-dir /tmp

# Silent mode for scripting
copilot -p "Generate test file for utils.js" --allow-all-tools --silent

# Share session to file after completion
copilot -p "Add error handling" --allow-all-tools --share ./session-log.md

Following Up

• After every copilot command, immediately use AskUserQuestion to confirm next steps, collect clarifications, or decide whether to continue with copilot --continue.
• When resuming, use copilot --continue for the most recent session or copilot --resume to pick from previous sessions.
• The resumed session automatically uses the same model and settings from the original session.
• Restate the chosen model and mode when proposing follow-up actions.

Error Handling

• Stop and report failures whenever a copilot command exits non-zero; request direction before retrying.
• Before using high-impact flags (--allow-all-tools, --allow-all-paths), ask the user for permission using AskUserQuestion unless it was already given.
• When output includes warnings or partial results, summarize them and ask how to adjust using AskUserQuestion.
• If copilot is not installed, guide the user to install it from https://docs.github.com/en/copilot/using-github-copilot/using-github-copilot-in-the-command-line

Best Practices

1. Model Selection: Choose the appropriate model based on task complexity (codex models for code-heavy tasks, claude for analysis)
2. Permission Management: Use --add-dir to grant specific access instead of --allow-all-paths when possible
3. Tool Permissions: In interactive mode, let the user approve tools individually; use --allow-all-tools only for non-interactive automation
4. Session Continuity: Leverage --continue and --resume for multi-step workflows instead of starting fresh each time
5. Silent Mode: Use --silent when scripting to get clean output without stats

Additional Options

• --log-level <level> - Set logging verbosity (none, error, warning, info, debug, all)
• --stream <mode> - Enable or disable streaming (on/off)
• --no-custom-instructions - Disable loading custom instructions from AGENTS.md
• --agent <agent> - Specify a custom agent (only in prompt mode)
• --available-tools [tools...] - Limit available tools to specific set
• --disable-parallel-tools-execution - Execute tools sequentially instead of in parallel