The onboarding process for new developers at MacCoss Lab, which has maintained a vast 700,000-line codebase for over 17 years, has proven effective when applied to Claude Code. Brendan MacLean, a Developer Ambassador at Claude, details this approach.
Skyline, an open-source protein analysis software developed at the University of Washington's MacCoss Lab, has been continuously improved since 2008. This tool is essential for researchers working on protein detection and quantification, aiding in biomarker discovery and drug development. The lab's extensive codebase, primarily in C#, is supported by a robust testing framework that includes over 200,000 automated nightly tests.
Lessons from Onboarding
Brendan MacLean has spent nearly three decades onboarding various lab members, from undergraduates to postdocs. As developers transitioned in and out, the codebase evolved, leading to certain sections becoming increasingly difficult to manage.
Brendan's extensive experience in training new developers provided him with insights on how to effectively integrate Claude into the lab's workflow. Initially skeptical about AI tools' capabilities with C#, he began experimenting with Claude.ai and discovered that the same onboarding principles could be applied.
Building Effective Context
To facilitate Claude's understanding of the codebase, Brendan created a dedicated repository for AI context, separate from the main codebase. This repository includes a CLAUDE.md file that outlines the environment setup and directs Claude to essential documentation, serving as a foundational guide.
Brendan developed a library of skills that Claude can utilize, focusing on debugging and project-specific conventions. This approach allows Claude to engage in root cause analysis rather than merely guessing solutions, significantly improving its effectiveness.
Accelerating Development
With the established context, the process of teaching Claude about the codebase became more efficient. Brendan successfully completed a long-stalled project, the Files View panel, in just two weeks with Claude's assistance. Previously, such projects often remained incomplete due to developer turnover.
In another instance, Brendan revived a Java module for Skyline that had been neglected after the original developer left. With Claude's help, he was able to implement desired features and enhancements quickly.
Automation and Efficiency
Claude has transformed the development process at MacCoss Lab. Developers are now primarily instructing Claude to generate automation scripts and implement features, allowing them to focus on more creative tasks. For example, one developer who was initially doubtful about AI tools successfully created a new plotting extension with Claude's support.
Key Takeaways for Developers
- Prioritize Context: Maintaining context is crucial for success. Developers should invest in creating and updating a context layer that persists across sessions.
- Build a Skill Library: Skills should encapsulate domain knowledge, ensuring Claude can access the necessary information without redundancy.
- Utilize Data Access Integrations: Establish integrations that allow Claude to access real data, enhancing its functionality and relevance.
Brendan emphasizes that building context is essential, especially in open-source projects where institutional memory is limited. By treating context as a project artifact, it can be preserved and utilized by future contributors.
In summary, the onboarding strategies developed over 17 years at MacCoss Lab have not only streamlined the integration of new developers but have also enhanced the capabilities of Claude Code, demonstrating the value of effective context management in software development.
