15 tricks separating casual Claude users from true masters. None require being technical: just knowing they exist.
1. Start with "Before answering..."
Force explicit reasoning: "Before answering, tell me what you assume about the problem".
2. Use XML to separate blocks
Wrap context in <context> and tasks in <task> so Claude processes them separately.
3. Ask for "output only, no explanation"
When you want just result (JSON, table, code), specify "respond ONLY with output, no explanatory text".
4. Switch models mid-task
Start with Opus to design plan, switch to Sonnet to execute. Save money without sacrificing critical quality.
5. Save prompts as Skills
If you repeat a prompt >3 times, encapsulate in Skill.
6. Use /clear strategically
Every time you change big task, clear context.
7. Ask for alternatives
"Give me 3 different versions of X. At end say which is best and why".
8. Prompt caching in API
If your prompt has long repeated context, activate cache. Save up to 90%.
9. Use prefill to force format
In API, start response with "{" to force JSON. With "Here's:" force direct tone.
10. Metaprompting
Ask Claude to help improve your prompts.
11. Test with edge cases
When writing prompt for automation, test with weird inputs.
12. Ask for critique of your own text
"Act as demanding critic. Find 5 weakest points of this text".
13. Combine Skills + subagents
Skill orchestrates; subagents execute parts.
14. Use Projects for persistent memory
Each Project can have instructions and files Claude remembers between sessions.
15. Learn from history
Periodically review your best prompts and patterns. Document them. In 6 months you'll have a personal library worth gold.
Conclusion
These tips won't change your life alone, but consistently applied they mark the difference between someone who "uses Claude" and someone who masters it.