A city-bike repair-stand team wants every GenAI session to stay concise and avoid invented part numbers, and they also need one ask that lists boundary values for the torque field. How should those instructions be assigned?
Select an answer to reveal the explanation.
Short Explanation
Sticky house style is the shop sign you hang once; the torque boundary list is today’s work ticket. Put lasting “be concise, don’t invent part numbers” in the system prompt, then send the boundary-value ask as a user prompt. That is the usual set-once, ask-many pattern.
Full Explanation
Typical usage sets the system prompt once with durable role and constraint text, then sends successive user prompts for individual tasks. Conciseness and bans on invented part numbers are standing rules; listing torque-field boundary values is a one-off analysis ask. Swapping those layers, rewriting the system prompt for every task, or treating the LLM as a CT-AI-style SUT misses the intended separation.