MEMORY HIERARCHIES FOR CONTEXT

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS Prompt 1 helps you design the tiers and the movement rules. [START OF PROMPT] Design a three-tier memory hierarchy (hot/active context, warm/compressed summaries, cold/full transcripts) and the exact tier-movement rules for this long-running agent task: [TASK DESCRIPTION, e.g. "processing hundreds of resumes over several days"]. Specify the trigger condition for moving data between each tier and what retrieval looks like when old information is needed again. Output format: a tier-definition table plus a movement-rule list. Include storage format suggestions and one failure case to watch for. [END OF PROMPT] Prompt 2 helps you design and test a scratchpad. [START OF PROMPT] Design a structured scratchpad file format for this agent to track its own progress, including fields for completed steps, open questions, key decisions, and next action: agent = [AGENT DESCRIPTION, e.g. Resume Review Agent]. Then, given this mid-task scratchpad snapshot, determine the correct next action: [PASTE LAB SCRATCHPAD/NOTES FILE CONTENTS HERE]. Output format: the scratchpad schema, then the determined next action with justification. If the snapshot is inconsistent or missing information, say so plainly instead of guessing. [END OF PROMPT]