Playbooks
When a prompt can't do the job
A prompt can score well and still be impossible to run, because it asks a model for something that model cannot do on its own: writing to a specific folder, running on a schedule, remembering last week. Rewording does not reach that. Each playbook below names the gap and the smallest fix that is not a better prompt.
Recurring report that must land in the right folder
You have one long master prompt you paste in every week. Some weeks the document comes out right, some weeks the sections are reordered or renamed, and some weeks the assistant says it created the file and nothing appears.
Chat that repeats what it already did
You open a new chat, paste roughly the same prompt you used last time, and the assistant repeats something it already handled. It re-drafts a note you already sent, or asks a question you answered a few sessions ago.
Assistant that says it sent an email and didn't
You asked the assistant to email a client, it replied as if the email went out, and nothing arrived. Or it did send, straight past you, and you only found out because the client wrote back.
Output format that changes every run
You run the same prompt on a new record and the output comes back shaped differently each time: a field renamed or a section missing, and the numbers themselves sitting in a table one run and buried in a sentence the next.
Missed a clause in a long document
You pasted a 60-page policy into a single prompt and asked for a summary of the exclusions, and the answer missed a clause that was sitting in plain sight on page 40.
Quoted rate that was already stale
You asked for a rate, and the model answered with a specific number as if reading it off a current table. The number turned out to be months out of date, because nothing about the question told the model otherwise.
Batch of fifty that quietly finishes as twelve
You handed the assistant a list of fifty clients and one instruction to apply to each of them. It stopped at around a dozen, with no error and no explanation, just fewer results than records you gave it.
Coverage term the document never actually said
You asked for a summary of a policy, and it stated a coverage term as a plain fact. You went back to check, and that term wasn't in the document anywhere. The model filled in a gap with something that sounded exactly right.
Draft that went out before anyone looked at it
You told the assistant to check with you before anything went out, and it sent anyway. There's no state anywhere tracking whether you'd actually looked at this particular one yet, so nothing stopped it.
Prompt that was supposed to run every Monday
You asked the assistant to do this every Monday, and it ran the task once, right there in that conversation. The following Monday, nothing happened, because nothing anywhere was watching for Monday to arrive.