CONTEXT AS THE SCARCE RESOURCE

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS The following prompts are written so that you can paste them into an AI assistant and replace the bracketed parts with your own details. Treat them as a skilled colleague you consult on the job. Always check the assistant's numbers yourself, because assistants can make arithmetic mistakes. Prompt 1 is for planning what goes into a window that is too small for everything. [START OF PROMPT] Given this context window size and these candidate items competing for space, produce a prioritized include, summarize, or exclude plan that fits the budget while preserving task-critical information: context window = [TOKEN LIMIT], items = [LIST ITEMS WITH APPROXIMATE SIZES, e.g. lab resume files, job posting text, prior tool outputs]. Output format: a table with the columns Item, Size, Decision, and Justification, plus the resulting total token count. Please remember to reserve room for the model's own answer and a safety margin, and show your arithmetic so I can check it. [END OF PROMPT] Prompt 2 is for diagnosing a real usage log. [START OF PROMPT] Our agent's context usage is tracked in this log: [PASTE LAB OR PRODUCTION TOKEN-USAGE LOG HERE]. Diagnose whether we are at risk of running out of context mid-task given current trends, and recommend a specific numeric budget-allocation split, for example the percentage reserved for system instructions, recent history, and retrieved documents. Output format: a diagnosis paragraph plus a budget-allocation table. Explain which numbers in the log led to each conclusion. [END OF PROMPT]