Prompt & Context Engineer Course

Contents

1. How to use this book
2. Chapter 1. Instruction design

   The model is a capable colleague on their first day · Identify the task type first · Four elements of a good instruction · A role sets a perspective; it does not supply knowledge · Turn vague requirements into verifiable instructions · Say what to do rather than only what not to do · A reason helps the model generalize · Number procedures when order matters · Message roles and priority · A template separating rules from data · The author resolves conflicting instructions · A policy for missing information · Acting versus suggesting, and the trap of emphasis · Keep the right altitude and start with a minimal prompt · Edit instructions down and keep secrets in the right place

3. Chapter 2. Examples and structure

   Examples are the sample garments you hand a tailor · What makes a good example · A table for designing diversity · Boundary cases and contrastive pairs · A process for reducing example bias · Keep examples separate from evaluation data · Draw boundaries with XML tags · Choose a format for the recipient · Data dictionaries and missing values · Structured outputs: enforcing format · Format compliance is not content accuracy · How prefill works and what replaces it · Request short explanations in evidence fields · Show methods to models with thinking enabled · The display environment determines formatting

4. Chapter 3. Context engineering

   Context is the space on your desk · Budgeting the context window · Include only the information that is needed · Attach source, version and scope to documents · Long documents first, the question at the end · Get the quotes first · Citation features and search-result blocks · How retrieval augmentation works · Diagnose retrieval failures separately from answer failures · Conversation summarization and compaction · Context strategies for long-running agents · Prompt caching and arrangement · Where untrusted content belongs · Put everything in a long window, or retrieve it? · Tool design is context design · Entrusting conversation state to a server

5. Chapter 4. Reasoning and multi-step tasks

   Reasoning is scratch paper · Reasoning models and general models · Reasoning effort as a dial · Match reasoning effort to cost and latency · Output limits and reasoning tokens · Prompts for reasoning models · Do not request internal thinking in the output · Task decomposition and workflow patterns · Input/output contracts for each stage · Make self-review a concrete check · Parallel, sequential and subagent work · The cost of over-splitting a chain · Long work across multiple context windows

6. Chapter 5. Evaluation and improvement

   Evaluation is like retaking the same exam · Turn success criteria into evaluation items · How to write a rubric · Building a test set · Separate improvement and confirmation sets · Three ways to grade · Limits and calibration of LLM evaluators · Variability and repeated runs · Compare before and after on the same set · Managing regressions · Classify failures to choose the next fix · Agent evaluation: outcomes and records · When scores fail to represent reality · Consistency in human grading

7. Appendix A. Glossary
8. Appendix B. Self-check answers and explanations
9. Appendix C. Exam day guide
10. Appendix D. Reference documents