💡 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.
You are planning an agentic task. Before calling any tool, output a numbered plan of the steps you intend to take, the tools each step needs, and an estimate of how many tool calls the whole run should take. Stop and wait for my approval. Do not begin executing until I reply with GO.
Here is the task below. Tell me which thinking level you would run it at - low, medium, high or xhigh - and why. Give me the trade off in one sentence: what the higher level would buy and what it would cost. Then run it at the level you recommended.
I am going to paste a large document. Pricing is flat across the full context, so do not summarise to save tokens. Read all of it. Then answer only the question I ask, citing the exact section each claim comes from. If the document does not answer the question, say so instead of inferring.
After completing the task, output a table with one row per tool call you made: the tool name, why you called it, what came back, and whether it changed your plan. Mark any call that turned out to be unnecessary. I want to see the waste, not a clean summary.
For this task, if you reach a point where you need information you do not have, stop and ask me rather than making an assumption. Do not fill gaps with plausible values. If you are more than slightly unsure about a fact, label it clearly as unverified and tell me what would confirm it.
Below is a process I currently do by hand, written as prose. Rewrite it as a sequence of discrete steps an agent could execute, naming the tool or data source each step needs. Flag any step that requires a judgement call a human should keep. Then tell me which steps are worth automating first and why.