💡 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.
Before you act on this task, write out:
- What you believe I am asking for
- The approach you intend to take, and why that one
- What you are assuming that I have not told you
- What would make this approach wrong
Then wait for me to confirm. Treat this as part of the deliverable, not preamble.
You have finished. Produce a record of what happened:
- Every action taken, in order
- What each one changed
- Anything that did not go as expected
- Anything you did that was not in the plan, and why
Be complete rather than brief. Assume someone else will read this without context.
I am building a workflow where an agent will: [DESCRIBE]
Tell me what to log so that in six months I could establish what happened on any given run. Specifically:
- What to capture at each step
- What would be missing if I only logged inputs and outputs
- What I would want if something went wrong and I had to explain it
- What is cheap to log and what is expensive
Keep it to what I would realistically maintain.
I need to catch silent errors in this workflow: [DESCRIBE]
Design a set of test cases where I already know the correct answer, chosen so that a wrong result would be obvious rather than plausible.
Include at least two cases where a subtly wrong answer would look right to a casual reader.
On a previous run you did this: [PASTE THE ACTION LOG]
Reconstruct what the reasoning must have been. Then tell me:
- Which steps follow clearly from the inputs
- Which required a judgement I cannot verify from the log
- What additional logging would have made this reconstructable
Do not invent a rationale. Where you cannot tell, say so.
Here is what I currently log for agent runs: [DESCRIBE]
If someone asked me to account for a specific run, what could I establish and what could I not?
Be specific about the gaps rather than reassuring. I would rather find them now.