Appendix - Prompt & Context Engineering

What We Learned

• Specify the task, constraints, evidence and output format using the model's message structure.
• Use examples, decomposition, tool use or verification when they improve measured performance.
• Constrained output formats enforce structure; semantic correctness still needs validation.
• Manage context with retrieval, compression, offloading and stable cacheable prefixes.
• Version prompts, evaluate changes and optimize on training/dev cases while keeping a held-out test.

Key Acronyms, Concepts & Jargon

Term	Short meaning
Zero-shot / few-shot	Instructions without examples / instructions with example demonstrations.
ICL	In-Context Learning: adapting behavior from examples in the prompt without weight updates.
CoT / ReAct	Chain of Thought / Reasoning and Acting: intermediate reasoning / reasoning interleaved with tools.
ToT / self-consistency	Tree of Thoughts / choosing among multiple sampled reasoning paths.
Prompt chaining	Passes one model step's output into the next step.
Instruction hierarchy	Rules for resolving conflicting instructions by source priority.
JSON Schema	Specifies the expected structure and types of JSON data.
Constrained decoding	Restricts allowed next tokens to enforce an output grammar or schema.
Context engineering	Selects and organizes instructions, evidence, history and tool results.
Prompt caching	Reuses computation for a matching prompt prefix.
Prompt injection	Attempts to turn untrusted content into controlling instructions.
DSPy	Framework for declarative model programs and metric-driven optimization.
Prompt optimization / overfitting	Search for better prompt behavior / tailoring it too closely to development cases.

Back to section overview