What is plan mode in Claude Code?

Plan mode restricts Claude Code to read-only tools so it compiles a plan of action before touching your code. Anthropic engineers note the newer models increasingly plan well on their own, so some now skip it for auto mode — but it remains the safe default for bigger changes.

What should go in a CLAUDE.md file?

As little as possible. Claude Code's creator recommends doing the minimal thing that keeps the model on track, and if the file bloats, deleting it and adding back a little at a time. The team's own habit: every time Claude makes a mistake, write the correction into CLAUDE.md or a skill instead of just re-prompting.

How do I get Claude Code to verify its own work?

Don't stop at unit tests, lint, and type checks — those were already automated. Anthropic engineers frame agent verification as 'can the agent run the thing?': give Claude a way to actually execute and observe its output, and it can loop until results are correct.

Claude Code Best Practices, From Anthropic

Keep exploring

MCP explained by the people building it
Agent skills & progressive disclosure
The future of coding, from the people building it
Podcast MCP server — bring transcripts into Claude