Cli Mastery

Copilot CLI Mastery

UTILITY SKILL — interactive Copilot CLI trainer. INVOKES: ask_user, sql, view USE FOR: "cliexpert", "teach me the Copilot CLI", "quiz me on slash commands", "CLI cheat sheet", "copilot CLI final exam" DO NOT USE FOR: general coding, non-CLI questions, IDE-only features

Routing and Content

Specific CLI questions get direct answers without loading references. Reference files in references/ dir. Read on demand with view.

Behavior

On first interaction, initialize progress tracking:

CREATE TABLE IF NOT EXISTS mastery_progress (key TEXT PRIMARY KEY, value TEXT);
CREATE TABLE IF NOT EXISTS mastery_completed (module TEXT PRIMARY KEY, completed_at TEXT DEFAULT (datetime('now')));
INSERT OR IGNORE INTO mastery_progress (key,value) VALUES ('xp','0'),('level','Newcomer'),('module','0');

XP: lesson +20, correct +15, perfect quiz +50, scenario +30. Levels: 0=Newcomer 100=Apprentice 250=Navigator 400=Practitioner 550=Specialist 700=Expert 850=Virtuoso 1000=Architect 1150=Grandmaster 1500=Wizard. Max XP from all content: 1600 (8 modules × 145 + 8 scenarios × 30 + final exam 200).

When module counter exceeds 8 and user says "cliexpert", offer: scenarios, final exam, or review any module.

Rules: ask_user with choices for ALL quizzes/scenarios. Show XP after correct answers. One concept at a time; offer quiz or review after each lesson.

Final Exam

Present a 10-question comprehensive exam using ask_user with 4 choices each. Require 80%+ to pass. Vary the selection each time.

Question Bank

1. Which command initializes Copilot CLI in a new project? → /init 2. What shortcut cycles through modes? → Shift+Tab 3. Where are repo-level custom agents stored? → .github/agents/*.md 4. What does MCP stand for? → Model Context Protocol 5. Which agent is safe to run in parallel? → explore 6. How do you add a file to AI context? → @filename (e.g. @src/auth.ts) 7. What file has the highest instruction precedence? → CLAUDE.md / GEMINI.md / AGENTS.md (git root + cwd) 8. Which command compresses conversation history? → /compact 9. Where is MCP configured at project level? → .github/mcp-config.json 10. What does --yolo do? → Same as --allow-all (skip all confirmations) 11. What does /research do? → Run a deep research investigation with sources 12. Which shortcut opens input in $EDITOR? → Ctrl+G 13. What does /reset-allowed-tools do? → Re-enables confirmation prompts 14. Which command copies the last AI response to your clipboard? → /copy 15. What does /compact do? → Summarizes conversation to free context

On pass (80%+): Award "CLI Wizard" title, congratulate enthusiastically! On fail: Show which they got wrong, encourage retry.

Module 1: Slash Commands

Teach these categories one at a time, with examples and "when to use" guidance.

Getting Started

Models & Agents

Code & Review

Session & Context

Permissions & Directories

Configuration & Customization

Extensibility

Workflows & Research

Quiz (5+ questions, use ask_user with 4 choices each)

Ask "Which command would you use to [scenario]?" style questions.

Module 2: Keyboard Shortcuts

Navigation & Editing

Line Editing

Pro tips to teach

• @ is THE most important shortcut — it's how you give precise context
• !git status runs git directly without AI processing
• Shift+Tab into Plan mode BEFORE complex tasks
• Ctrl+G opens your $EDITOR for long prompts — game changer
• Ctrl+S lets you iterate on a prompt without retyping

Module 3: Interaction Modes

Interactive Mode (default)

• AI acts immediately on your prompts
• Asks permission for risky operations
• Best for: quick tasks, debugging, exploring code
• 80% of your time will be here

Plan Mode (Shift+Tab or /plan)

• AI creates a step-by-step plan FIRST
• You review and approve before execution
• Best for: complex refactoring, architecture changes, risky operations
• Key insight: Use this when mistakes are expensive

Autopilot Mode (experimental, /experimental)

• AI acts without asking for confirmation
• Best for: trusted environments, long-running tasks
• Use with caution — pair with /allow-all or --yolo

Mode Comparison

Teaching point: The right mode at the right time = 10x productivity.

Module 4: Agent System

Built-in Agents

Custom Agents — define your own in Markdown

Agent file anatomy

---
name: my-agent
description: What this agent does
tools:
  - bash
  - edit
  - view
---

# Agent Instructions
Your detailed behavior instructions here.

Agent orchestration patterns

1. Fan-out exploration — Launch multiple explore agents in parallel to answer different questions simultaneously 2. Pipeline — explore → understand → general-purpose → implement → code-review → verify 3. Specialist handoff — Identify task → /agent to pick specialist → review with /fleet or /tasks

Key insight: The AI automatically delegates to subagents when appropriate.

Module 5: Skills System

What are skills?

• Specialized capability packages the AI can invoke
• Think of them as "expert modes" with domain-specific knowledge
• Managed via /skills command

Skill locations

Creating a custom skill

1. Create the directory: mkdir -p ~/.copilot/skills/my-skill/ (or mkdir -p ~/.agents/skills/my-skill/) 2. Create SKILL.md with YAML frontmatter (name, description, optional tools) 3. Write detailed instructions for the AI's behavior 4. Verify with /skills

Skill design best practices

• Clear description — helps the AI match tasks to your skill automatically
• Focused scope — each skill should do ONE thing well
• Include instructions — specify exactly how the skill should operate
• Test thoroughly — use /skills to verify, then invoke and check results

Auto-matching

When you describe a task, the AI checks if any skill matches and suggests using it.

Module 6: MCP Integration

What is MCP?

• Model Context Protocol — a standard for connecting AI to external tools
• Think of it as "USB ports for AI" — plug in any compatible tool
• The GitHub MCP server is built-in (search repos, issues, PRs, actions)

Key commands

Popular MCP servers

• @modelcontextprotocol/server-postgres — Query PostgreSQL databases
• @modelcontextprotocol/server-sqlite — Query SQLite databases
• @modelcontextprotocol/server-filesystem — Access local files with permissions
• @modelcontextprotocol/server-memory — Persistent knowledge graph
• @modelcontextprotocol/server-puppeteer — Browser automation

Configuration

Config file format

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["@modelcontextprotocol/server-postgres", "{{env.DATABASE_URL}}"],
      "env": { "NODE_ENV": "development" }
    }
  }
}

Security best practices

• Never put credentials directly in config files
• Use environment variable references: {{env.SECRET}}
• Review MCP server source before using
• Only connect servers you actually need

Module 7: Advanced Techniques

1. `@` file mentions — Always give precise context, don't rely on the AI finding files

• @src/auth.ts — single file
• @src/components/ — directory listing
• "Fix @src/auth.ts to match @tests/auth.test.ts" — multi-file context

2. `! shell bypass` — !git log --oneline -5 runs instantly, no AI overhead

3. `/research` — Run a deep research investigation using GitHub search and web sources

4. `/resume` + `--continue` — Session continuity across CLI launches

5. `/compact` — Compress history when context gets large (auto at 95%)

• Check with /context first
• Best used at natural task boundaries
• Warning signs: AI contradicting earlier statements, token usage >80%

6. `/context` — Visualize what's eating your token budget

7. Custom instructions precedence (highest to lowest):

• CLAUDE.md / GEMINI.md / AGENTS.md (git root + cwd)
• .github/instructions/**/*.instructions.md (path-specific!)
• .github/copilot-instructions.md
• ~/.copilot/copilot-instructions.md
• COPILOT_CUSTOM_INSTRUCTIONS_DIRS (additional directories via env var)

8. Path-specific instructions:

• .github/instructions/backend.instructions.md with applyTo: "src/api/**"
• Different coding standards for different parts of the codebase

9. LSP config — ~/.copilot/lsp-config.json or .github/lsp.json

10. `/review` — Get code review without leaving terminal

11. `--allow-all` / `--yolo` — Full trust mode (use responsibly!)

12. `Ctrl+T` — Watch the AI think (learn its reasoning patterns)

Module 8: Configuration

Key files

Environment variables

Permissions model

• Default: confirmation required for edits, creates, shell commands
• /allow-all or --yolo: skip all confirmations for the session
• /reset-allowed-tools: re-enable confirmations
• Directory allowlists, tool approval gates, MCP server trust

Logging levels

error, warn, info, debug, trace (COPILOT_LOG_LEVEL=debug copilot)

Use debug/trace for: MCP connection issues, tool failures, unexpected behavior, bug reports

Scenario Challenges

Present these as real-world situations. Ask the user what commands/shortcuts they'd use. Use ask_user with choices for each step.

Scenario 1: Hotfix Review Under Pressure

A production bug fix is ready. You need to inspect the diff, run code review, and keep sensitive data hidden because you're on a livestream.

Answer: /streamer-mode → /diff → /review @src/payment.ts

Scenario 2: Context Window Rescue

Your session is huge and model quality is dropping. Keep continuity while shrinking noise.

Answer: /context → /compact → /resume (or restart with --continue)

Scenario 3: Autonomous Refactor Sprint

You want an agent to execute a refactor with minimal prompts, but only after reviewing a plan and setting permissions.

Answer: Shift+Tab (Plan mode) → validate plan → /allow-all → execute in Autopilot mode

Scenario 4: Enterprise Onboarding

Set up custom agents, repo instructions, and MCP integration for a new team repository.

Answer: Add agent profiles to .github/agents/, verify /instructions, then /mcp add

Scenario 5: Power Editing Session

You're crafting a long prompt and need to edit quickly without losing context.

Answer: Ctrl+G (open in editor), Ctrl+A (jump to start), Ctrl+K (trim)

Scenario 6: Agent Orchestration

You're leading a complex project: understand code, run tests, refactor, then review.

Answer: explore agent (understand) → task agent (tests) → general-purpose (refactor) → code-review (verify)

Scenario 7: New Project Setup

You cloned a new repo and need to set up Copilot CLI for max productivity.

Answer: /init → /model → /mcp add (if needed) → Shift+Tab to Plan mode for first task

Scenario 8: Production Safety

Switching from boilerplate work to production deployment scripts.

Answer: /reset-allowed-tools → Plan mode → /review before every commit