**Tip of the Day: Context Engineering Over Prompt Engineering**

Stop optimizing how you ask and start optimizing what you provide. Maintain lightweight identifiers (file paths, stored queries) in your CLAUDE.md and dynamically load full context via tools at runtime. A lean CLAUDE.md with just-in-time retrieval outperforms a massive one that front-loads everything. Move infrequently-used instructions into Skills (loaded on-demand at ~100 tokens scanning cost). This alone can save 30%+ on token costs while improving output quality.