Prompt Engineering¶
Beginner → Intermediate · 8 topics
The LLMs notes introduced prompts: the parts of a good one, few-shot examples and templates. This series goes deeper, into the skills that separate a prompt that worked once from one you can ship: instructions a model follows on the first try, examples chosen on purpose, reasoning and chaining for hard tasks, reliable structured output, defence against prompt injection, and testing prompts like code.
# Nothing to install. Every example uses only the Python standard library and a fake model, so it runs offline.
# Code that needs an API key or a project layout is marked "no-run".
| # | Topic | Sub-topics |
|---|---|---|
| 1 | Writing clear instructions | The new-colleague test · reasons behind rules · success criteria · numbers over adjectives · ordering long prompts · prefill · linting prompts |
| 2 | Few-shot examples in depth | Zero-, few- and many-shot · auditing an example set · examples as messages · order effects · retrieving similar examples · when examples hurt |
| 3 | Reasoning techniques | Chain of thought with tags · self-consistency voting · decomposition · step-back prompting · self-critique · prompting reasoning models |
| 4 | Prompt chaining & routing | One prompt vs a chain · a working chain · validation gates · routers · map-reduce for long inputs · chains vs agents |
| 5 | Structured output prompting | Field descriptions · nulls, not guesses · validation · repair loops · evidence and confidence · classification · native schema mode |
| 6 | Prompt injection & defensive prompting | Direct vs indirect · delimiters and spotlighting · instruction hierarchy · input and output checks · least privilege · attack sets |
| 7 | Testing & versioning prompts | Eval sets · code checks vs LLM judges · comparing versions · regression tests in CI · a prompt registry · cost optimisation |
| 8 | Prompt recipes for common tasks | Summaries · classification · extraction · grounded RAG answers · rewriting · code review · agent system prompts |
flowchart LR
W[Clear instructions] --> E[Examples] --> R[Reasoning] --> C[Chains & routing]
C --> S[Structured output] --> D[Injection defence] --> T[Tests & versions]Before you start
Read Prompt engineering in the LLMs notes first; it covers the basics these pages build on. For ready-made prompts by industry, see the Prompt library.
Back to: Notes overview