💡 How to use these prompts:
Replace everything in [BRACKETS] with your specific details before sending.
Click Copy to copy any prompt to your clipboard instantly.
Here is a full agent session log. Report the exact token count of the very first request sent to the model, broken into system prompt, tool definitions, injected context and the user task. Give each as a number and as a percentage of the first call. Do not estimate - if a figure is not derivable from the log, say which one is missing.
From this session log, list every request to the model in order with its input token count. Then state the average increase per step and whether growth is linear or compounding. Identify the step where context first exceeded twice the opening size.
Review this agent session and list every step whose output did not change a file, run a command that produced new information, or alter the plan. For each, quote the step and say what it added. Report the count as a share of total steps.
Here are two session logs for the same task using different harnesses. For each, total input tokens, output tokens, step count and cost at the rates below. Present as a table. State which was cheaper, by what factor, and whether the task outcome differed.
This session used a computer-use or browser tool. Count the number of screenshots or page captures sent to the model, estimate their share of total input tokens, and list any consecutive captures where the screen did not meaningfully change. Report what share of image tokens were spent on unchanged screens.