The author shares their extensive experience using Claude Code, detailing both personal and professional applications. They emphasize the importance of the CLAUDE.md file, which acts as the primary source of truth for project-specific information. For personal projects, they allow Claude to generate content freely, while in a professional setting, they maintain strict documentation standards. This includes limiting content to tools used by a significant portion of the team, ensuring clarity and conciseness. The author suggests that the document should serve as a high-level guide rather than a comprehensive manual, pushing for simplicity and effectiveness. They outline key practices for writing a useful CLAUDE.md, such as starting with guardrails and avoiding bloated references to external documents. Negative constraints should be minimized to prevent confusion for the agent. The author also discusses their strategies for managing context during coding sessions, advocating against relying on auto-compaction features in favor of manual resets that help preserve valuable context. The use of custom slash commands is framed as a tool for efficiency rather than complexity. The author warns against building extensive command lists that could hinder usability, suggesting that commands should remain straightforward. Additionally, while custom subagents are touted as a powerful feature, the author points out their limitations, such as disrupting the agent's holistic understanding and imposing rigid workflows. Instead, they prefer using the built-in Task feature for better context management, allowing the main agent to manage its own delegation.