Chapter L10-C02 · Language models and AI application patterns
Context and output contracts
Separate trusted instructions from untrusted source content; Reject a schema-valid but unsupported field when evidence is missing.
Units in this chapter
- L10-U03 Prompt and context engineering Separate trusted instructions from untrusted source content
- L10-U04 Structured output and validation Reject a schema-valid but unsupported field when evidence is missing
Reading that connects these units
Reading
- Prompt engineering guide Clear instructions, examples, and context; actual behavior depends on the model.About this reference (W16)
- OWASP GenAI Security Project Prompt injection, sensitive information, excessive agency, and application threat models.About this reference (W26)
- FastAPI tutorial Validated request/response models and service interfaces; choose relevant tutorial units.About this reference (W14)
- Evaluation best practices Evaluation datasets, criteria, comparisons, and ongoing regression checks.About this reference (W17)
- OpenAI Cookbook Use one relevant example and validate against current API docs before running.About this reference (G10)
Selected, not required. Reviewed 2026-11-17.