How to turn Claude Code into a domain specific coding agent

Coding agents can be improved for domain-specific tasks by providing high-quality, condensed information rather than just raw documentation. Experiments were conducted using Claude Code to generate code for the LangGraph library, testing configurations with and without access to documentation tools and a curated guide (`Claude.md`). An evaluation framework assessed code quality using smoke tests, task-specific tests, and an LLM-as-a-Judge for subjective aspects. The results showed that a concise guide combined with tools for accessing more details produced the best outcomes, as the guide helped the agent use the tools more effectively and avoid common errors.