creating-agents-md
About
This Claude Skill generates plain markdown agents.md files without YAML frontmatter for AI coding assistants. It provides project context guidelines in a free-form structure, typically placing a single agents.md file in the project root. Use it specifically when creating documentation to guide AI assistants on project-specific setup and conventions.
Quick Install
Claude Code
Recommended/plugin add https://github.com/pr-pm/prpmgit clone https://github.com/pr-pm/prpm.git ~/.claude/skills/creating-agents-mdCopy and paste this command in Claude Code to install this skill
Documentation
Creating agents.md Files
Overview
The agents.md format provides project-specific context for AI coding assistants. It's the simplest format: plain markdown only with NO YAML frontmatter, NO special syntax.
CRITICAL:
- No frontmatter - Pure markdown only (no YAML)
- Free-form content - No required structure
- Single file - Typically
agents.mdin project root
Quick Reference
| Aspect | Requirement |
|---|---|
| Format | Plain markdown |
| Frontmatter | None (forbidden) |
| Structure | Free-form |
| File location | agents.md in project root |
Creating agents.md Files
Plain markdown with no frontmatter:
# TaskMaster Development Guide
## Project Overview
TaskMaster is a task management application for remote teams, built with real-time collaboration features and offline-first architecture.
## Architecture
### Frontend
- React 18 with TypeScript
- Vite for build tooling
- Zustand for state management
- React Query for server state
- Tailwind CSS for styling
### Backend
- Node.js with Express
- PostgreSQL with Prisma ORM
- WebSocket for real-time features
- Redis for caching and pub/sub
- JWT for authentication
## Coding Conventions
- Use TypeScript strict mode
- Functional components with hooks (no class components)
- Server components by default in Next.js
- Colocate tests with source files (*.test.tsx)
- Use Zod for runtime validation
## File Structure
\`\`\`
src/
components/ # Reusable UI components
features/ # Feature-based modules
hooks/ # Custom React hooks
lib/ # Utility functions
pages/ # Route pages
types/ # TypeScript types
\`\`\`
## Development Workflow
1. Create feature branch from `main`
2. Write tests first (TDD)
3. Implement feature
4. Run `pnpm test` and `pnpm lint`
5. Create PR with description
6. Merge after approval
## Testing
- Use Vitest for unit tests
- Use Playwright for E2E tests
- Aim for 80% coverage on new code
- Mock external dependencies
What to Include
Focus on project-specific information AI doesn't already know:
High Priority:
- Project overview and purpose
- Architecture decisions and patterns
- Tech stack and dependencies
- File structure and organization
- Coding conventions
- Development workflow
- Testing approach
- Domain knowledge and business logic
Skip:
- General programming best practices
- Language syntax explanations
- Framework basics
- Obvious code quality rules
Example: Backend API Project
# Payment Gateway API
## Overview
RESTful API for payment processing with support for multiple payment providers.
## Tech Stack
- Node.js 20.x
- Express
- PostgreSQL 15
- Redis for rate limiting
- Stripe and PayPal integrations
## API Design
### Endpoints
All endpoints follow REST conventions:
- `GET /api/payments` - List payments
- `GET /api/payments/:id` - Get payment details
- `POST /api/payments` - Create payment
- `PUT /api/payments/:id` - Update payment
- `DELETE /api/payments/:id` - Cancel payment
### Error Handling
Return consistent error format:
\`\`\`json
{
"error": {
"code": "PAYMENT_FAILED",
"message": "Payment could not be processed",
"details": {...}
}
}
\`\`\`
## Security
- All endpoints require JWT authentication
- Rate limiting: 100 requests/minute per IP
- Input validation with Zod schemas
- SQL injection prevention via Prisma
- PCI DSS compliance for payment data
## Database
### Conventions
- Use snake_case for table/column names
- Add timestamps (created_at, updated_at) to all tables
- Use UUIDs for primary keys
- Foreign keys follow `{table}_id` pattern
Example: Frontend Component Library
# Design System Components
A React component library following Atomic Design principles.
## Component Structure
All components follow this structure:
\`\`\`
ComponentName/
ComponentName.tsx # Main component
ComponentName.test.tsx # Tests
ComponentName.stories.tsx # Storybook stories
index.ts # Exports
\`\`\`
## Styling
- Use Tailwind CSS utility classes
- Create custom classes in `styles/components/` for complex components
- Follow BEM naming for custom classes
- Responsive by default (mobile-first)
## TypeScript
\`\`\`typescript
// Good: Explicit prop types
interface ButtonProps {
variant: 'primary' | 'secondary' | 'ghost';
size?: 'sm' | 'md' | 'lg';
disabled?: boolean;
onClick?: () => void;
children: React.ReactNode;
}
export function Button({ variant, size = 'md', ...props }: ButtonProps) {
return <button className={cn(variants[variant], sizes[size])} {...props} />;
}
\`\`\`
## Accessibility
- All interactive elements must be keyboard accessible
- Use semantic HTML (button, nav, main, etc.)
- Include ARIA labels for icon-only buttons
- Test with screen readers
- Maintain minimum 4.5:1 contrast ratio
Content Format
Free-form markdown including:
- Project overview: Purpose and goals
- Architecture notes: Technical decisions and patterns
- Conventions: Coding standards and practices
- Context: Domain knowledge and business logic
- Workflows: Development processes
- File structure: Directory organization
- Dependencies: Key libraries and tools
Common Mistakes
| Mistake | Fix |
|---|---|
| Adding YAML frontmatter | No frontmatter allowed - plain markdown only |
| Generic best practices | Focus on project-specific patterns |
| Verbose explanations | Be concise, AI already knows general concepts |
| Language tutorials | Skip basics, focus on project conventions |
| Missing context | Include domain knowledge and business logic |
Writing Style
Concise (Good):
## Testing
- Vitest for unit tests
- Playwright for E2E
- 80% coverage target
- Mock external dependencies
Verbose (Bad):
## Testing
When you are writing tests, it's important to understand that we use Vitest
for our unit tests because it's fast and modern. For end-to-end testing,
we have chosen to use Playwright because it provides excellent cross-browser
support and has a great developer experience...
File Placement
Typically in project root:
project-root/
agents.md # Main file
src/
tests/
package.json
Can also be in subdirectories for monorepos:
monorepo/
packages/
frontend/
agents.md # Frontend-specific context
backend/
agents.md # Backend-specific context
Validation
Documentation: /Users/khaliqgant/Projects/prpm/app/packages/converters/docs/agents-md.md
Schema location: /Users/khaliqgant/Projects/prpm/app/packages/converters/schemas/agents-md.schema.json
Best Practices
- Be concise: Focus on project-specific info (AI knows general practices)
- Keep updated: Review and update as project evolves
- Real examples: Show actual code patterns from your project
- Plain markdown: No YAML frontmatter or special syntax
- Human-readable: Write for both AI and human developers
- Project-specific: Avoid generic advice that AI already knows
- Natural structure: Organize however makes sense for your project
Migration from Other Formats
When converting to agents.md:
- Strip all frontmatter - Remove YAML headers completely
- Focus on content - Keep only markdown content
- Combine files - Merge multiple rule files into one cohesive document
- Simplify - Remove format-specific features (globs, regex, etc.)
- Plain markdown only - Use standard markdown syntax
Official Specification
For the authoritative specification, see: https://github.com/openai/agents.md
Remember: agents.md uses plain markdown with NO frontmatter. Free-form structure. Focus on project-specific context AI doesn't already know.
GitHub Repository
Related Skills
sglang
MetaSGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.
evaluating-llms-harness
TestingThis Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.
llamaguard
OtherLlamaGuard is Meta's 7-8B parameter model for moderating LLM inputs and outputs across six safety categories like violence and hate speech. It offers 94-95% accuracy and can be deployed using vLLM, Hugging Face, or Amazon SageMaker. Use this skill to easily integrate content filtering and safety guardrails into your AI applications.
langchain
MetaLangChain is a framework for building LLM applications using agents, chains, and RAG pipelines. It supports multiple LLM providers, offers 500+ integrations, and includes features like tool calling and memory management. Use it for rapid prototyping and deploying production systems like chatbots, autonomous agents, and question-answering services.
