Worked on the marimo-team/marimo repository to resolve a critical duplication issue in the PydanticAI completions API. Addressed the problem by centralizing system prompt handling at agent construction, ensuring the prompt is passed only once and removing it from agent.run. This approach reduced duplication in code completions, particularly in tab completion scenarios, and improved API consistency. Implemented end-to-end tests using Python to verify the fix and maintain reliability across the completions pipeline. Prepared release notes and aligned documentation to support the update. Demonstrated skills in AI development, backend development, and unit testing while enhancing the user experience and codebase reliability.