Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
175 lines (130 loc) · 5.41 KB

CONTRIBUTING.md

File metadata and controls

175 lines (130 loc) · 5.41 KB

Contribution Guidelines

Welcome to submit PRs! We recommend keeping each agent "small and specialized" and focused on solving problems in specific scenarios.

TRAE Agents Banner

English | 中文

🚀 Create a New Agent

1. Fork the Repository and Create a Branch

git checkout -b feature/your-agent-name

2. Create Agent Directory

  • Create a new folder in the agents/ directory
  • Use kebab-case naming: your-agent-name
  • Copy the template file agents/_template/README.md to your directory

3. Fill in Configuration Information

Fill in the following content according to the template requirements:

✅ Required Sections

  • Basic Information: Name, description, applicable scenarios
  • Prompt: Complete system instructions
  • Usage Examples: At least 1 input-output example

🔧 Optional Sections

  • MCP Service Configuration: If MCP is needed
  • Tool Configuration: Recommended built-in tools
  • Collaborative Agents: Other agents that can work together
  • Configuration Suggestions: Model selection, parameter settings, etc.
  • Related Resources: Documentation and reference links

4. Test Your Agent

Actually create and test in TRAE:

  1. Open TRAE
  2. Create an agent according to your written configuration
  3. Test with different inputs
  4. Optimize prompts based on test results

5. Update README

Add your agent to the agent list in README.md:

| your-agent-name | One-line description | ✅ Stable | [Configuration Details](./agents/your-agent-name/) |

6. Submit PR

  • Ensure all files are committed
  • Write clear commit messages
  • Push to your fork
  • Create Pull Request

📋 Quality Standards

Your agent should meet the following requirements:

✅ Content Requirements

  • Clear Positioning: Clearly explain what this agent does
  • Complete Prompt: Include role definition, objectives, workflow, and constraints
  • Practical Examples: Show real input-output examples
  • Reproducible: Others can create agents with the same effect following your configuration

❌ Issues to Avoid

  • Do not include sensitive information (API keys, tokens, internal URLs, etc.)
  • Do not be too broad (one agent doing everything)
  • Do not lack examples
  • Do not use vague instructions (like "do your best")

📁 File Structure

Each agent directory should include:

your-agent-name/
├── README.md          # Main configuration file (required)
└── resources/        # Related resource files (optional)
    └── reference.md

🏷️ Status Indicators

Use emoji indicators for agent status in README:

  • 🚧 Under Development: Still testing, may be adjusted
  • ✅ Stable: Tested and working well
  • ⚠️ Deprecated: Still available but not recommended for new use

🎯 Naming Conventions

Agent Name

  • Use English with capitalized first letters: Git Commit Generator
  • Or use kebab-case: git-commit-generator
  • Clearly express the purpose, avoid ambiguity

Directory Name

  • Must use kebab-case: git-commit-generator
  • All lowercase
  • Clear and concise

✅ PR Checklist

Before submitting, please confirm:

  • Created agent directory and README.md
  • Includes complete Prompt (can be directly copied to TRAE)
  • Provided at least 1 usage example
  • No sensitive information (API keys, tokens, etc.)
  • Updated the main README agent list
  • Actually tested in TRAE
  • Followed the template structure
  • Used correct naming conventions

💡 Best Practices

Tips for Writing Good Prompts

  1. Clear Role: "You are an XXX expert"
  2. Define Objectives: "Your responsibility is to..."
  3. Step by Step: "First step... Second step..."
  4. Give Examples: "For example..."
  5. State Constraints: "Do not..."

Common Patterns

Analysis + Recommendation Type

You are a code review expert. When users provide code:
1. Analyze code structure and logic
2. Identify potential issues
3. Provide improvement suggestions

Task Execution Type

You are a Git assistant. When users need to write commit messages:
1. Read code changes
2. Determine change type
3. Generate standardized commit messages

Q&A Consulting Type

You are a technical consultant. When answering user technical questions:
1. Understand the core of the problem
2. Provide multiple solutions
3. Analyze pros and cons of each solution

🤔 Frequently Asked Questions

Q: How long should the Prompt be?

A: There's no fixed standard, typically 200-500 words is appropriate. The key is to be clear and complete, not longer is better.

Q: Do I need to configure MCP?

A: Most agents don't need it. Only use when you really need to access external data or services.

Q: How to test the effect?

A: After creating in TRAE, test with 3-5 different real scenarios to see if they all give reasonable responses.

Q: Can I reference other agents?

A: Yes, but ensure the referenced agents already exist and are stable.

📚 Reference Resources

🙏 Acknowledgments

Thank you to every developer contributing wisdom to the community! 🎉

Feel free to discuss any questions through Issues.