A comprehensive guide to LLM prompting fundamentals covering: role assignment, structured input/output with XML tags, response prefilling, n-shot prompting (recommending 12+ examples rather than just a few), Chain-of-Thought with sketchpads, multi-step prompt decomposition, context placement, instruction formatting, hallucination reduction techniques, stop sequences, and temperature tuning. Practical code examples use the Claude Messages API throughout. Key insights include why 'think step by step' alone is insufficient for CoT, how to use sketchpads to reduce hallucinations, and why starting temperature at 0.8 and lowering is preferable to starting low.
