03 · Development Workflows

What if the AI could find bugs you didn’t even know to ask about?

In this chapter, GitHub Copilot CLI becomes your daily driver. You’ll use it inside the workflows you already rely on every day: testing, refactoring, debugging, and Git.

🎯 Learning Objectives

By the end of this chapter, you’ll be able to:

• Run comprehensive code reviews with Copilot CLI
• Refactor legacy code safely
• Debug issues with AI assistance
• Generate tests automatically
• Integrate Copilot CLI with your git workflow

⏱️ Estimated Time: ~60 minutes (15 min reading + 45 min hands-on)

---

🧩 Real-World Analogy: A Carpenter’s Workflow

A carpenter doesn’t just know how to use tools, they have workflows for different jobs:

Similarly, developers have workflows for different tasks. GitHub Copilot CLI enhances each of these workflows, making you more efficient and effective in your daily coding tasks.

---

The Five Workflows

Each workflow below is self-contained. Pick the ones that match your current needs, or work through them all.

---

Choose Your Own Adventure

This chapter covers five workflows that developers typically use. However, you don’t need to read them all at once! Each workflow is self-contained in a collapsible section below. Pick the ones that match what you need and that fits best with your current project. You can always come back and explore the others later.

I want to…	Jump to
Review code before merging	Workflow 1: Code Review
Clean up messy or legacy code	Workflow 2: Refactoring
Track down and fix a bug	Workflow 3: Debugging
Generate tests for my code	Workflow 4: Test Generation
Write better commits and PRs	Workflow 5: Git Integration
Research before coding	Quick Tip: Research Before You Plan or Code
See a full bug-fix workflow end to end	Putting It All Together

Select a workflow below to expand it and see how GitHub Copilot CLI can enhance your development process in that area.

---

Workflow 1: Code Review - Review files, use the /review agent, create severity checklists

Basic Review

This example uses the @ symbol to reference a file, giving Copilot CLI direct access to its contents for review.

copilot

> Review @samples/book-app-project/book_app.py for code quality

---

🎬 See it in action!

Demo output varies. Your model, tools, and responses will differ from what’s shown here.

---

Input Validation Review

Ask Copilot CLI to focus its review on a specific concern (here, input validation) by listing the categories you care about in the prompt.

copilot

> Review @samples/book-app-project/utils.py for input validation issues. Check for: missing validation, error handling gaps, and edge cases

Cross-File Project Review

Reference an entire directory with @ to let Copilot CLI scan every file in the project at once.

copilot

> @samples/book-app-project/ Review this entire project. Create a markdown checklist of issues found, categorized by severity

Interactive Code Review

Use a multi-turn conversation to drill deeper. Start with a broad review, then ask follow-up questions without restarting.

copilot

> @samples/book-app-project/book_app.py Review this file for:
> - Input validation
> - Error handling
> - Code style and best practices

# Copilot CLI provides detailed review

> The user input handling - are there any edge cases I'm missing?

# Copilot CLI shows potential issues with empty strings, special characters

> Create a checklist of all issues found, prioritized by severity

# Copilot CLI generates prioritized action items

Review Checklist Template

Ask Copilot CLI to structure its output in a specific format (here, a severity-categorized markdown checklist you can paste into an issue).

copilot

> Review @samples/book-app-project/ and create a markdown checklist of issues found, categorized by:
> - Critical (data loss risks, crashes)
> - High (bugs, incorrect behavior)
> - Medium (performance, maintainability)
> - Low (style, minor improvements)

Understanding Git Changes (Important for /review)

Before using the /review command, you need to understand two types of changes in git:

Change Type	What It Means	How to See
Staged changes	Files you’ve marked for the next commit with git add	git diff --staged
Unstaged changes	Files you’ve modified but haven’t added yet	git diff

# Quick reference
git status           # Shows both staged and unstaged
git add file.py      # Stage a file for commit
git diff             # Shows unstaged changes
git diff --staged    # Shows staged changes

Using the /review Command

The /review command invokes the built-in code-review agent, which is optimized for analyzing staged and unstaged changes with high signal-to-noise output. Use a slash command to trigger a specialized built-in agent instead of writing a free-form prompt.

copilot

> /review
# Invokes the code-review agent on staged/unstaged changes
# Provides focused, actionable feedback

> /review Check for security issues in authentication
# Run review with specific focus area

💡 Tip: The code-review agent works best when you have pending changes. Stage your files with git add for more focused reviews.

---

Workflow 2: Refactoring - Restructure code, separate concerns, improve error handling

Simple Refactoring

Try this first: @samples/book-app-project/book_app.py The command handling uses if/elif chains. Refactor it to use a dictionary dispatch pattern.

Start with straightforward improvements. Try these on the book app. Each prompt uses an @ file reference paired with a specific refactoring instruction so Copilot CLI knows exactly what to change.

copilot

> @samples/book-app-project/book_app.py The command handling uses if/elif chains. Refactor it to use a dictionary dispatch pattern.

> @samples/book-app-project/utils.py Add type hints to all functions

> @samples/book-app-project/book_app.py Extract the book display logic into utils.py for better separation of concerns

💡 New to refactoring? Start with simple requests like adding type hints or improving variable names before tackling complex transformations.

---

🎬 See it in action!

Demo output varies. Your model, tools, and responses will differ from what’s shown here.

---

Separate Concerns

Reference multiple files with @ in a single prompt so Copilot CLI can move code between them as part of the refactor.

copilot

> @samples/book-app-project/utils.py @samples/book-app-project/book_app.py
> The utils.py file has print statements mixed with logic. Refactor to separate display functions from data processing.

Improve Error Handling

Provide two related files and describe the cross-cutting concern so Copilot CLI can suggest a consistent fix across both.

copilot

> @samples/book-app-project/utils.py @samples/book-app-project/books.py
> These files have inconsistent error handling. Suggest a unified approach using custom exceptions.

Add Documentation

Use a detailed bullet list to specify exactly what each docstring should contain.

copilot

> @samples/book-app-project/books.py Add comprehensive docstrings to all methods:
> - Include parameter types and descriptions
> - Document return values
> - Note any exceptions raised
> - Add usage examples

Safe Refactoring with Tests

Chain two related requests in a multi-turn conversation. First generate tests, then refactor with those tests as a safety net.

copilot

> @samples/book-app-project/books.py Before refactoring, generate tests for current behavior

# Get tests first

> Now refactor the BookCollection class to use a context manager for file operations

# Refactor with confidence - tests verify behavior is preserved

---

Workflow 3: Debugging - Track down bugs, security audits, trace issues across files

Simple Debugging

Try this first: @samples/book-app-buggy/books_buggy.py Users report that searching for "The Hobbit" returns no results even though it's in the data. Debug why.

Start by describing what’s wrong. Here are common debugging patterns you can try with the buggy book app. Each prompt pairs an @ file reference with a clear symptom description so Copilot CLI can locate and diagnose the bug.

copilot

# Pattern: "Expected X but got Y"
> @samples/book-app-buggy/books_buggy.py Users report that searching for "The Hobbit" returns no results even though it's in the data. Debug why.

# Pattern: "Unexpected behavior"
> @samples/book-app-buggy/book_app_buggy.py When I remove a book that doesn't exist, the app says it was removed. Help me find why.

# Pattern: "Wrong results"
> @samples/book-app-buggy/books_buggy.py When I mark one book as read, ALL books get marked. What's the bug?

💡 Debugging tip: Describe the symptom (what you see) and the expectation (what should happen). Copilot CLI figures out the rest.

---

🎬 See it in action!

Demo output varies. Your model, tools, and responses will differ from what’s shown here.

---

The “Bug Detective” - AI Finds RELATED Bugs

This is where context-aware debugging shines. Try this scenario with the buggy book app. Provide the whole file via @ and describe only the user-reported symptom. Copilot CLI will trace the root cause and may spot additional bugs nearby.

copilot

> @samples/book-app-buggy/books_buggy.py
>
> Users report: "Finding books by author name doesn't work for partial names"
> Debug why this happens

What Copilot CLI does:

Root Cause: Line 80 uses exact match (==) instead of partial match (in).

Line 80: return [b for b in self.books if b.author == author]

The find_by_author function requires an exact match. Searching for "Tolkien"
won't find books by "J.R.R. Tolkien".

Fix: Change to case-insensitive partial match:
return [b for b in self.books if author.lower() in b.author.lower()]

Why this matters: Copilot CLI reads the whole file, understands the context of your bug report, and gives you a specific fix with a clear explanation.

💡 Bonus: Because Copilot CLI analyzes the entire file, it often discovers other issues you didn’t ask about. For example, while fixing the author search, Copilot CLI might also notice the case-sensitivity bug in find_book_by_title!

Real-World Security Sidebar

While debugging your own code is important, understanding security vulnerabilities in production applications is critical. Try this example: Point Copilot CLI at an unfamiliar file and ask it to audit for security issues.

copilot

> @samples/buggy-code/python/user_service.py Find all security vulnerabilities in this Python user service

This file demonstrates real-world security patterns you’ll encounter in production apps.

💡 Common security terms you’ll encounter:

• SQL Injection: When user input is put directly into a database query, allowing attackers to run malicious commands
• Parameterized queries: The safe alternative - placeholders (?) separate user data from SQL commands
• Race condition: When two operations happen at the same time and interfere with each other
• XSS (Cross-Site Scripting): When attackers inject malicious scripts into web pages

---

Understanding an Error

Paste a stack trace directly into your prompt along with an @ file reference so Copilot CLI can map the error to the source code.

copilot

> I'm getting this error:
> AttributeError: 'NoneType' object has no attribute 'title'
>     at show_books (book_app.py:19)
>
> @samples/book-app-project/book_app.py Explain why and how to fix it

Debugging with Test Case

Describe the exact input and observed output to give Copilot CLI a concrete, reproducible test case to reason about.

copilot

> @samples/book-app-buggy/books_buggy.py The remove_book function has a bug. When I try to remove "Dune",
> it also removes "Dune Messiah". Debug this: explain the root cause and provide a fix.

Trace an Issue Through Code

Reference multiple files and ask Copilot CLI to follow the data flow across them to locate where the issue originates.

copilot

> Users report that the book list numbering starts at 0 instead of 1.
> @samples/book-app-buggy/book_app_buggy.py @samples/book-app-buggy/books_buggy.py
> Trace through the list display flow and identify where the issue occurs

Understanding Data Issues

Include a data file alongside the code that reads it so Copilot CLI understands the full picture when suggesting error-handling improvements.

copilot

> @samples/book-app-project/data.json @samples/book-app-project/books.py
> Sometimes the JSON file gets corrupted and the app crashes. How should we handle this gracefully?

---

Workflow 4: Test Generation - Generate comprehensive tests and edge cases automatically

Try this first: @samples/book-app-project/books.py Generate pytest tests for all functions including edge cases

The “Test Explosion” - 2 Tests vs 15+ Tests

Manually writing tests, developers typically create 2-3 basic tests:

• Test valid input
• Test invalid input
• Test an edge case

Watch what happens when you ask Copilot CLI to generate comprehensive tests! This prompt uses a structured bullet list with an @ file reference to guide Copilot CLI toward thorough test coverage:

copilot

> @samples/book-app-project/books.py Generate comprehensive pytest tests. Include tests for:
> - Adding books
> - Removing books
> - Finding by title
> - Finding by author
> - Marking as read
> - Edge cases with empty data

---

🎬 See it in action!

Demo output varies. Your model, tools, and responses will differ from what’s shown here.

---

What you get: 15+ comprehensive tests including:

class TestBookCollection:
    # Happy path
    def test_add_book_creates_new_book(self):
        ...
    def test_list_books_returns_all_books(self):
        ...

    # Find operations
    def test_find_book_by_title_case_insensitive(self):
        ...
    def test_find_book_by_title_returns_none_when_not_found(self):
        ...
    def test_find_by_author_partial_match(self):
        ...
    def test_find_by_author_case_insensitive(self):
        ...

    # Edge cases
    def test_add_book_with_empty_title(self):
        ...
    def test_remove_nonexistent_book(self):
        ...
    def test_mark_as_read_nonexistent_book(self):
        ...

    # Data persistence
    def test_save_books_persists_to_json(self):
        ...
    def test_load_books_handles_missing_file(self):
        ...
    def test_load_books_handles_corrupted_json(self):
        ...

    # Special characters
    def test_add_book_with_unicode_characters(self):
        ...
    def test_find_by_author_with_special_characters(self):
        ...

Result: In 30 seconds, you get edge case tests that would take an hour to think through and write.

---

Unit Tests

Target a single function and enumerate the input categories you want tested so Copilot CLI generates focused, thorough unit tests.

copilot

> @samples/book-app-project/utils.py Generate comprehensive pytest tests for get_book_details covering:
> - Valid input
> - Empty strings
> - Invalid year formats
> - Very long titles
> - Special characters in author names

Running Tests

Ask Copilot CLI a plain-English question about your toolchain. It can generate the right shell command for you.

copilot

> How do I run the tests? Show me the pytest command.

# Copilot CLI responds:
# cd samples/book-app-project && python -m pytest tests/
# Or for verbose output: python -m pytest tests/ -v
# To see print statements: python -m pytest tests/ -s

Test for Specific Scenarios

List advanced or tricky scenarios you want covered so Copilot CLI goes beyond the happy path.

copilot

> @samples/book-app-project/books.py Generate tests for these scenarios:
> - Adding duplicate books (same title and author)
> - Removing a book by partial title match
> - Finding books when collection is empty
> - File permission errors during save
> - Concurrent access to the book collection

Add Tests to Existing File

Ask for additional tests for a single function so Copilot CLI generates new cases that complement what you already have.

copilot

> @samples/book-app-project/books.py
> Generate additional tests for the find_by_author function with edge cases:
> - Author name with hyphens (e.g., "Jean-Paul Sartre")
> - Author with multiple first names
> - Empty string as author
> - Author name with accented characters

---

Workflow 5: Git Integration - Commit messages, PR descriptions, /pr, /delegate, and /diff

💡 This workflow assumes basic git familiarity (staging, committing, branches). If git is new to you, try the other four workflows first.

Generate Commit Messages

Try this first: copilot -p "Generate a conventional commit message for: $(git diff --staged)" — stage some changes, then run this to see Copilot CLI write your commit message.

This example uses the -p inline prompt flag with shell command substitution to pipe git diff output directly into Copilot CLI for a one-shot commit message. The $(...) syntax runs the command inside the parentheses and inserts its output into the outer command.

# See what changed
git diff --staged

# Generate commit message using [Conventional Commit](https://github.com/github/copilot-cli-for-beginners/blob/main/GLOSSARY.md#conventional-commit) format
# (structured messages like "feat(books): add search" or "fix(data): handle empty input")
copilot -p "Generate a conventional commit message for: $(git diff --staged)"

# Output: "feat(books): add partial author name search
#
# - Update find_by_author to support partial matches
# - Add case-insensitive comparison
# - Improve user experience when searching authors"

---

🎬 See it in action!

Demo output varies. Your model, tools, and responses will differ from what’s shown here.

---

Explain Changes

Pipe the output of git show into a -p prompt to get a plain-English summary of the last commit.

# What did this commit change?
copilot -p "Explain what this commit does: $(git show HEAD --stat)"

PR Description

Combine git log output with a structured prompt template to auto-generate a complete pull request description.

# Generate PR description from branch changes
copilot -p "Generate a pull request description for these changes:
$(git log main..HEAD --oneline)

Include:
- Summary of changes
- Why these changes were made
- Testing done
- Breaking changes? (yes/no)"

Using /pr in Interactive Mode for the Current Branch

If you’re working with a branch in Copilot CLI’s interactive mode, you can use the /pr command to work with pull requests. Use /pr to view a PR, create a new PR, fix an existing PR, or let Copilot CLI auto-decide based on the branch state.

copilot

> /pr [view|create|fix|auto]

Review Before Push

Use git diff main..HEAD inside a -p prompt for a quick pre-push sanity check across all branch changes.

# Last check before pushing
copilot -p "Review these changes for issues before I push:
$(git diff main..HEAD)"

Using /delegate for Background Tasks

The /delegate command hands off work to the Copilot coding agent on GitHub. Use the /delegate slash command (or the & shortcut) to offload a well-defined task to a background agent.

copilot

> /delegate Add input validation to the login form

# Or use the & prefix shortcut:
> & Fix the typo in the README header

# Copilot CLI:
# 1. Commits your changes to a new branch
# 2. Opens a draft pull request
# 3. Works in the background on GitHub
# 4. Requests your review when done

This is great for well-defined tasks you want completed while you focus on other work.

Using /diff to Review Session Changes

The /diff command shows all changes made during your current session. Use this slash command to see a visual diff of everything Copilot CLI has modified before you commit.

copilot

# After making some changes...
> /diff

# Shows a visual diff of all files modified in this session
# Great for reviewing before committing

---

Quick Tip: Research Before You Plan or Code

When you need to investigate a library, understand best practices, or explore an unfamiliar topic, use /research to run a deep research investigation before writing any code:

copilot

> /research What are the best Python libraries for validating user input in CLI apps?

Copilot searches GitHub repositories and web sources, then returns a summary with references. This is useful when you’re about to start a new feature and want to make informed decisions first. You can share the results using /share.

💡 Tip: /research works well before /plan. Research the approach, then plan the implementation.

---

Putting It All Together: Bug Fix Workflow

Here’s a complete workflow for fixing a reported bug:

# 1. Understand the bug report
copilot

> Users report: 'Finding books by author name doesn't work for partial names'
> @samples/book-app-project/books.py Analyze and identify the likely cause

# 2. Debug the issue (continuing in same session)
> Based on the analysis, show me the find_by_author function and explain the issue

> Fix the find_by_author function to handle partial name matches

# 3. Generate tests for the fix
> @samples/book-app-project/books.py Generate pytest tests specifically for:
> - Full author name match
> - Partial author name match
> - Case-insensitive matching
> - Author name not found

# 4. Generate commit message
copilot -p "Generate commit message for: $(git diff --staged)"

# Output: "fix(books): support partial author name search"

Bug Fix Workflow Summary

Step	Action	Copilot Command
1	Understand the bug	> [describe bug] @relevant-file.py Analyze the likely cause
2	Get detailed analysis	> Show me the function and explain the issue
3	Implement the fix	> Fix the [specific issue]
4	Generate tests	> Generate tests for [specific scenarios]
5	Commit	copilot -p "Generate commit message for: $(git diff --staged)"

---

Practice

Now it’s your turn to apply these workflows.

---

▶️ Try It Yourself

After completing the demos, try these variations:

1. Bug Detective Challenge: Ask Copilot CLI to debug the mark_as_read function in samples/book-app-buggy/books_buggy.py. Did it explain why the function marks ALL books as read instead of just one?

2. Test Challenge: Generate tests for the add_book function in the book app. Count how many edge cases Copilot CLI includes that you wouldn’t have thought of.

3. Commit Message Challenge: Make any small change to a book app file, stage it (git add .), then run:

   copilot -p "Generate a conventional commit message for: $(git diff --staged)"

   Is the message better than what you would have written quickly?

Self-Check: You understand development workflows when you can explain why “debug this bug” is more powerful than “find bugs” (context matters!).

---

📝 Assignment

Main Challenge: Refactor, Test, and Ship

The hands-on examples focused on find_book_by_title and code reviews. Now practice the same workflow skills on different functions in book-app-project:

1. Review: Ask Copilot CLI to review remove_book() in books.py for edge cases and potential issues: @samples/book-app-project/books.py Review the remove_book() function. What happens if the title partially matches another book (e.g., "Dune" vs "Dune Messiah")? Are there any edge cases not handled?
2. Refactor: Ask Copilot CLI to improve remove_book() to handle edge cases like case-insensitive matching and returning useful feedback when a book isn’t found
3. Test: Generate pytest tests specifically for the improved remove_book() function, covering:
   • Removing a book that exists
   • Case-insensitive title matching
   • A book that doesn’t exist returns appropriate feedback
   • Removing from an empty collection
4. Review: Stage your changes and run /review to check for any remaining issues
5. Commit: Generate a conventional commit message: copilot -p "Generate a conventional commit message for: $(git diff --staged)"

💡 Hints (click to expand)

Sample prompts for each step:

copilot

# Step 1: Review
> @samples/book-app-project/books.py Review the remove_book() function. What edge cases are not handled?

# Step 2: Refactor
> Improve remove_book() to use case-insensitive matching and return a clear message when the book isn't found. Show me the before and after code.

# Step 3: Test
> Generate pytest tests for the improved remove_book() function, including:
> - Removing a book that exists
> - Case-insensitive matching ("dune" should remove "Dune")
> - Book not found returns appropriate response
> - Removing from an empty collection

# Step 4: Review
> /review

# Step 5: Commit
> Generate a conventional commit message for this refactor

Tip: After improving remove_book(), try asking Copilot CLI: “Are there any other functions in this file that could benefit from the same improvements?”. It may suggest similar changes to find_book_by_title() or find_by_author().

Bonus Challenge: Create an application with the Copilot CLI

💡 Note: This GitHub Skills exercise uses Node.js rather than Python. The GitHub Copilot CLI techniques you’ll practice - creating issues, generating code, and collaborating from the terminal - apply to any language.

The exercise shows developers how to use GitHub Copilot CLI to create issues, generate code, and collaborate from the terminal while building a Node.js calculator app. You’ll install the CLI, use templates and agents, and practice iterative, command-line driven development.

Start the “Create applications with the Copilot CLI” Skills Exercise

---

🔧 Common Mistakes & Troubleshooting (click to expand)

Common Mistakes

Mistake	What Happens	Fix
Using vague prompts like “Review this code”	Generic feedback that misses specific issues	Be specific: “Review for SQL injection, XSS, and auth issues”
Not using /review for code reviews	Missing the optimized code-review agent	Use /review which is tuned for high signal-to-noise output
Asking to “find bugs” without context	Copilot CLI doesn’t know what bug you’re experiencing	Describe the symptom: “Users report X happens when Y”
Generating tests without specifying framework	Tests may use wrong syntax or assertion library	Specify: “Generate tests using Jest” or “using pytest”

Troubleshooting

Review seems incomplete - Be more specific about what to look for:

copilot

# Instead of:
> Review @samples/book-app-project/book_app.py

# Try:
> Review @samples/book-app-project/book_app.py for input validation, error handling, and edge cases

Tests don’t match my framework - Specify the framework:

copilot

> @samples/book-app-project/books.py Generate tests using pytest (not unittest)

Refactoring changes behavior - Ask Copilot CLI to preserve behavior:

copilot

> @samples/book-app-project/book_app.py Refactor command handling to use dictionary dispatch. IMPORTANT: Maintain identical external behavior - no breaking changes

---

Summary

🔑 Key Takeaways

1. Code review becomes comprehensive with specific prompts
2. Refactoring is safer when you generate tests first
3. Debugging benefits from showing Copilot CLI the error AND the code
4. Test generation should include edge cases and error scenarios
5. Git integration automates commit messages and PR descriptions

📋 Quick Reference: See the GitHub Copilot CLI command reference for a complete list of commands and shortcuts.

---

✅ Checkpoint: You’ve Mastered the Essentials

Congratulations! You now have all the core skills to be productive with GitHub Copilot CLI:

Skill	Chapter	You Can Now…
Basic Commands	Ch 01	Use interactive mode, plan mode, programmatic mode (-p), and slash commands
Context	Ch 02	Reference files with @, manage sessions, understand context windows
Workflows	Ch 03	Review code, refactor, debug, generate tests, integrate with git

Chapters 04-06 cover additional features that add even more power and are worthwhile to learn.

---

🛠️ Building Your Personal Workflow

There’s no single “right” way to use GitHub Copilot CLI. Here are a few tips as you develop your own patterns:

📚 Official Documentation: Copilot CLI best practices for recommended workflows and tips from GitHub.

• Start with /plan for anything non-trivial. Refine the plan before execution - a good plan leads to better results.
• Save prompts that work well. When Copilot CLI makes a mistake, note what went wrong. Over time, this becomes your personal playbook.
• Experiment freely. Some developers prefer long, detailed prompts. Others prefer short prompts with follow-ups. Try different approaches and notice what feels natural.

💡 Coming up: In Chapters 04 and 05, you’ll learn how to codify your best practices into custom instructions and skills that Copilot CLI loads automatically.

---

➡️ What’s Next

The remaining chapters cover additional features that extend Copilot CLI’s capabilities:

Chapter	What It Covers	When You’ll Want It
Ch 04: Agents	Create specialized AI personas	When you want domain experts (frontend, security)
Ch 05: Skills	Auto-load instructions for tasks	When you repeat the same prompts often
Ch 06: MCP	Connect external services	When you need live data from GitHub, databases

Recommendation: Try the core workflows for a week, then return to Chapters 04-06 when you have specific needs.

---

Continue to Additional Topics

In Chapter 04: Agents and Custom Instructions, you’ll learn:

• Using built-in agents (/plan, /review)
• Creating specialized agents (frontend expert, security auditor) with .agent.md files
• Multi-agent collaboration patterns
• Custom instruction files for project standards

---