Contract

• Input: task description, target LLM, evaluation criteria.
• Output: prompt library + evaluation report + recommendation.
• Side effects: may call LLM APIs (cost, data privacy).
• Dependencies: LLM API access (OpenAI, Anthropic, local model server).
• Stop condition: best variant selected with evaluation evidence.
• Risk: medium — LLM outputs can be unreliable; requires validation; data privacy concerns.
• Boundary: evaluates and selects prompts; does not deploy to production without additional validation.

Prompt Engineering

Design LLM prompts that produce correct, consistent, and useful outputs — with evaluation, iteration, and failure-mode analysis.

Process