Skip to main content
To get the best results from Codegen, treat it like a skilled teammate: provide clear, specific instructions and sufficient context. Vague requests lead to ambiguous outcomes.
Codegen is based on Anthropic’s Claude 3.7. You can prompt it similarly to ChatGPT or other LLM-based assistants, but with access to powerful development tools.

​
The Core Principle: Specificity

Instead of “Fix the user service,” try:
In the my-web-app repo (PR #42), refactor the UserService class in src/services/user.ts to use the UserRepository pattern shown in ProductService/ProductRepository.
If there are specific implementation details you want included, make sure to specify. For example:
Ensure all tests in tests/services/user.test.ts pass and add new tests for the repository with 90%+ coverage. Update the diagram in docs/architecture/user-service.md.

​
Elements of a Strong Prompt

  1. Scope: What repository, branch, or files are involved? (e.g., my-web-app repo, PR #42, src/services/user.ts)
  2. Goal: What is the high-level objective? (e.g., Refactor UserService, improve testability)
  3. Tasks: What specific actions should the agent take? Use a numbered or bulleted list for clarity. (e.g., Extract logic to UserRepository, use dependency injection, update tests, update diagram)
  4. Context/Patterns: Are there existing patterns, examples, or documentation to reference? (e.g., ProductService, ProductRepository)
  5. Success Criteria: How will you know the task is done correctly? (e.g., Tests pass, 90%+ coverage, diagram updated)

​
Platform-Specific Guidance

​
Slack Integration

When prompting Codegen in Slack:
  • Use threads for complex, multi-step requests to keep context organized
  • Tag specific repos early in the conversation: “In the frontend repo…”
  • Reference PR numbers directly: “Review PR #123 and suggest improvements”
  • Ask for status updates on long-running tasks: “What’s the progress on the refactoring?”
Example Slack prompt:

​
Linear Integration

When working with Linear issues:
  • Reference issue numbers for context: “This relates to CG-1234”
  • Break down large tasks into sub-issues when appropriate
  • Specify acceptance criteria clearly in the issue description
  • Use @codegen to assign implementation tasks directly
Example Linear prompt:

​
GitHub Integration

When requesting PR reviews or code changes:
  • Link to specific files and line numbers when possible
  • Reference related issues or previous PRs for context
  • Specify review criteria: security, performance, maintainability
  • Request specific types of feedback: “Focus on error handling”
Example GitHub prompt:

​
Common Task Types

​
Code Review

​
Bug Fixing

​
Feature Development

​
Refactoring

​
Advanced Techniques

​
Providing Context Effectively

When working with large codebases:

​
Iterative Prompting

Build on previous work:

​
Handling Complex Dependencies

For interconnected changes:

​
Troubleshooting Common Issues

​
”I don’t have enough context”

Problem: Codegen asks for more information Solution: Provide file paths, existing patterns, and specific requirements upfront

​
”The changes broke existing functionality”

Problem: Modifications caused test failures Solution: Always specify which tests should continue passing and request regression testing

​
”The implementation doesn’t match our patterns”

Problem: Code doesn’t follow team conventions Solution: Reference specific examples of preferred patterns in your codebase

​
”The task is too large”

Problem: Request is overwhelming or unclear Solution: Break down into smaller, specific tasks with clear dependencies

​
Best Practices

Be Specific

Include file paths, function names, and exact requirements rather than general descriptions.

Provide Examples

Reference existing code patterns, similar implementations, or desired outcomes.

Set Success Criteria

Define what “done” looks like: passing tests, performance metrics, or specific functionality.

Consider Dependencies

Mention related systems, APIs, or components that might be affected by changes.
Clear, detailed prompts empower Codegen agents to deliver accurate results faster, significantly streamlining your workflow. Remember that Codegen has access to your entire codebase and can understand complex relationships between files.

​
Getting Help

If you’re unsure how to structure a prompt for your specific use case:
  • Start with a simple request and iterate based on the results
  • Ask Codegen to suggest how to break down complex tasks
  • Reference the capabilities documentation to understand available tools
  • Check out real examples in our use cases section