THE CORE AGENT LOOP

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS These prompts are written for you to give to an AI assistant while you work on the lab or on a real job. Replace the words in square brackets with your own material before sending. [START OF PROMPT] Review this agent run trace for ReAct-loop health: [PASTE LAB RUN TRACE / THOUGHT-ACTION-OBSERVATION LOG HERE]. Identify any step where reasoning did not logically follow from the prior observation, any redundant action repeated without new information, and whether the step/turn limit was adequate or wastefully high/low given how many steps were actually needed. Output format: a per-step annotation list plus a final recommendation on step-limit tuning. [END OF PROMPT] [START OF PROMPT] Recommend plan-and-execute versus step-by-step reactive looping for this task, and specify the exact termination condition and step/turn limit to configure: [TASK DESCRIPTION AND CONSTRAINTS]. Justify the choice against the task's structure (is it decomposable upfront, or does it need to adapt based on intermediate results), and state the specific risk of getting the step-limit number wrong in either direction. Output format: Recommendation (one paragraph), Config Values (step limit, turn limit, termination condition), Risk Notes (bullets). [END OF PROMPT]