Agentic Workflow Step

Overview

This lab teaches you how to set up reusable prompt templates and repository rules that make AI agents more effective and consistent in your projects.

Prompt Template System

Directory Structure

your-project/
├── .github/
│   └── prompts/
│       ├── feature.md
│       ├── bugfix.md
│       ├── refactor.md
│       └── review.md
├── AGENTS.md
└── ...

Feature Request Template

Create .github/prompts/feature.md:

# Feature Request: {{feature_name}}

## Requirements
{{requirements}}

## Technical Constraints
- Follow existing patterns in src/
- Add tests for new functionality
- Update documentation if needed

## Deliverables
1. Implementation in appropriate module
2. Unit tests with >80% coverage
3. Update CHANGELOG.md

## Quality Checklist
- [ ] TypeScript types are complete
- [ ] No eslint warnings
- [ ] Tests pass
- [ ] Reviewed for security concerns

Bug Fix Template

Create .github/prompts/bugfix.md:

# Bug Fix Request

## Issue
{{issue}}

## Expected Behavior
{{expected}}

## Actual Behavior
{{actual}}

## Investigation Steps
1. Identify root cause
2. Propose minimal fix
3. Explain why fix works

## Constraints
- Minimal change to fix the issue
- Don't refactor unrelated code
- Add regression test

Repository Rules (AGENTS.md)

Create a comprehensive AGENTS.md:

markdown
Copy
# AGENTS.md - AI Assistant Guidelines

## Project Context
This is a [project type] built with [tech stack].

## Code Standards

### Naming Conventions
- Files: kebab-case (user-profile.ts)
- Components: PascalCase (UserProfile.tsx)
- Functions: camelCase (getUserProfile)
- Constants: SCREAMING_SNAKE_CASE

### File Organization
- Components in src/components/
- Utilities in src/lib/
- Types in src/types/
- Tests alongside source files

## AI Behavior Rules

### DO
- Follow existing code patterns
- Add types to all new code
- Write tests for new features
- Explain changes in commit messages

### DON'T
- Modify package-lock.json directly
- Add new dependencies without approval
- Use console.log in production code
- Skip error handling

## Testing Requirements
- Unit tests for utilities
- Integration tests for API routes
- Component tests for complex UI

## Security
- Never commit secrets
- Validate all user input
- Use parameterized queries
- Sanitize output

## Review Checklist
Before suggesting changes, verify:
- [ ] Code compiles without errors
- [ ] Tests pass
- [ ] Lint passes
- [ ] No security vulnerabilities

Workflow Automation

Pre-Commit Hooks

Create .husky/pre-commit:

bash
Copy
#!/bin/sh
npm run lint
npm run typecheck
npm test -- --passWithNoTests

AI-Friendly Git Messages

Configure commit message templates:

bash
Copy
git config commit.template .gitmessage

Create .gitmessage:

# <type>(<scope>): <subject>
#
# Types: feat, fix, docs, style, refactor, test, chore
#
# Body: explain what and why
#
# Footer: reference issues

Prompt Chaining

For complex tasks, break into steps:

Step 1: Analysis

Before making changes, analyze:

1. What files are involved?
2. What are the dependencies?
3. What tests cover this code?
4. What could break?

Provide your analysis before suggesting code.

Step 2: Plan

Based on your analysis, create a plan:

1. List files to modify
2. Order of changes
3. Risks and mitigations
4. Testing approach

Get approval before implementing.

Step 3: Implement

Implement the approved plan:

- One file at a time
- Show diff for each change
- Run tests after each file
- Stop if tests fail

Verification

Your workflow is set up correctly if:

• AGENTS.md exists with clear guidelines
• Prompt templates are in .github/prompts/
• Pre-commit hooks run on commit
• AI responses follow your conventions

Common Mistakes