How RepoBird Prompts Work

Understanding RepoBird's AI Prompting

RepoBird uses carefully structured prompts powered by Claude Sonnet 4.5 (the world's strongest coding model with 77.2% on SWE-bench Verified) to ensure our AI agents deliver consistent, high-quality code solutions. Here's how our prompting system works behind the scenes.

The RepoBirdBot Identity

Every RepoBird agent introduces itself as RepoBirdBot, a professional AI coding agent that:

• Thoroughly researches codebases before making changes
• Follows existing project patterns and conventions
• Tests solutions comprehensively
• Works in a secure Debian environment with modern development tools

Prompt Structure

RepoBird uses XML-structured prompts for clarity and consistency. Each prompt includes:

1. System Context

Establishes the agent's identity, working environment, and professional approach.

2. Task Information

Depending on how you trigger RepoBird, the prompt structure adapts:

From GitHub Issues:

• Issue title and description
• Additional context (if provided)

From Pull Request Comments:

• Your specific request
• PR title and description for context

From Dashboard UI:

• Task title and description
• Any additional context you provide

Prompt Examples

Example: PR Comment Request

<system>
RepoBirdBot introduction and approach...
</system>

<user_request>
Please add unit tests for the new authentication module
</user_request>

<pr_title>
feat: Add OAuth2 authentication support
</pr_title>

<pr_description>
This PR adds OAuth2 authentication...
</pr_description>

Example: GitHub Issue

<system>
RepoBirdBot introduction and approach...
</system>

<issue_title>
Fix email validation in login form
</issue_title>

<issue_description>
The login form is not validating email addresses correctly...
</issue_description>

<context>
Reported by multiple users. High priority bug.
</context>

How This Benefits You

1. Consistency: Every run follows the same structured approach
2. Context Awareness: The agent understands where requests come from
3. Professional Results: Clear structure leads to better code quality
4. Transparency: You know exactly what information the agent receives

Tips for Better Results

• Be Specific: Clear, detailed descriptions lead to better solutions
• Provide Context: Additional context helps the agent understand priorities