Study · Claude Course

Prompt Engineering

Iteratively apply prompt techniques and re-measure with [[Prompt-Evaluation]]. Biggest levers, in order of impact seen in the course:

type conceptstatus activeprompting · technique

Key points

  • Be clear & direct: the first line matters most — action verb + clear task + output spec.
  • Be specific: Type A (output attributes: length/format/structure — use almost always) + Type B (reasoning steps — for complex problems); often combined.
  • Structure with XML tags: wrap sections in descriptive tags (<sales_records>) so Claude distinguishes content types; specific names beat generic ones.
  • Provide examples (one-/multi-shot): wrap example input/ideal-output pairs in XML; great for corner cases (sarcasm), formatting, and quality bars; place after instructions.
  • Measured gains compound (course example: 2.32 → 3.92 → 7.86).
Prompt-Evaluation · Anthropic-API-Basics

Sources

  • 2026-06-28-claude-course
Compiled from wiki/study/claude-course/Prompt-Engineering.md · git is the source of truth