Agent Systems: Prompt Chaining Pattern

Intent

Replace one vague prompt with a sequence of explicit, testable transformations.

Outline

1. Recurring problem: one prompt mixes extraction, reasoning, transformation, and presentation.
2. Structure: input → extract → transform → reason → compose.
3. Stable interface: each stage consumes a defined input and returns a defined output.
4. Variable implementation: the prompt, model, or tool used inside a stage.
5. Strength: inspectable intermediate results and local error handling.
6. Risks: latency, cost, and error propagation.
7. Example: paper → metadata → outline → theorem notes → proof notes.

Suggested figure

02-prompt-chaining.svg in the Agent design pattern illustration folder.