Templated Prompt Task

Implement a workflow task as a prompt template that fills inputs, elicits a completion, and post-processes the text into the task's output schema.

A templated prompt task hydrates a task-specific template with upstream inputs, runs the model, and parses the completion into the fields task I/O schema design requires — the "link" style many orchestration libraries encourage. Before reaching for this pattern, confirm the task actually needs an LLM at all — see non-LLM task implementation.

Gather, rank, and trim context under context engineering, then assemble a document whose natural completion is the task output. A useful framing is prefix/suffix completion: open with role and context sections, start the desired artifact (for example Dear {owner_name},), and close with a fixed sign-off so the completion body is exactly the extractable field — nothing more. Iterate by clarifying tone and labeling interpolated blocks so the model knows what it is reading.

Every templated task must post-process the completion for downstream consumers. When the needed output is a rigid multi-field structure rather than prose, prefer forced tool-choice extraction.

---

Sources

• Prompt Engineering for LLMs (Berryman), ch. 9