Skill: cc-performance-tuning

STOP - Measure First

• Don't optimize based on intuition—profile first
• Correctness before speed - Make it work, then make it fast
• <4% of code causes >50% of runtime - Find the hot spot before touching anything

CRITICAL: Even In Emergencies

Production down? Losing money? User panicking? Especially then:

• Guessing wrong costs MORE time than 60-second profiling
• Multiple shotgun changes make rollback impossible
• Wrong "fix" can mask the real problem for days