Chapter 18
18Prompting and System Prompts
18.1The system prompt as configuration
Planned: How a provider sets persona, rules, and defaults; the layering of system/developer/user roles.
18.2In-context learning
Planned: Zero-, few-shot, and why examples work without weight updates.
18.3Chain-of-thought and its descendants
Planned: Eliciting reasoning; when it helps and when it is theater.
18.4Prompt injection and the trust boundary
Planned: Why everything from tools and the web is untrusted data, not instructions.