The advice I find myself repeating every time someone asks how to get started with Claude Code
1. Setup Add tight, layered rules Claude Code loads CLAUDE.md files into context on every single message. Every line you add is tokens the model has to process and remember. Treat it like a config file, not a wiki. Use two layers: Global (~/.claude/CLAUDE.md) - things that apply to every project: language preferences, security rules, things to never do Project (<repo>/CLAUDE.md or <repo>/.claude/CLAUDE.md) - repo-specific context: build commands, architecture decisions, testing patterns ~/.claude/CLAUDE.md # always loaded, every project <project>/CLAUDE.md # loaded for that project only <project>/.claude/CLAUDE.md # alternative location (same effect) Think about the signal-to-noise ratio. If a rule isn’t relevant to the current task, it’s just noise misleading the prediction engine. Be clear, concise and specific. Don’t blindly import someone else’s entire rule set (including mine). ...