Patterns, not magic words

This guide distils the prompting patterns visible across the directory's 255 descriptions. It does not quote prompts the sources never published, and it does not promise magic phrases. What the successful stories share is structure, and structure can be copied. The entries below were chosen because their descriptions make the working method unusually visible.

Pattern one is the outcome-first brief: start with the finished state you want, then the constraints, then the evidence. Pattern two is the context dump: upload the policy, paste the thread, forward the receipts, because agents act on what they can see. Pattern three is the standing job: turn a repeated chore into an ongoing instruction with a reporting rhythm, like the price monitors and back-office runs in this directory.

Checkpoints and finish lines

Pattern four is the approval checkpoint. The careful users in these stories let Muse research and prepare freely, but require a yes before money moves or messages send. Pattern five is the verifiable finish line: a confirmation number, a booked reference, a filed claim. If a task has no finish line, the stories suggest shrinking it until it does.

A final habit is naming. Several people give their Muse a name and a role, and then brief it like a staff member with standing context. Whether or not the name matters to the model, it seems to matter to the human: the briefs get clearer. Take the patterns, test them on one small errand from the list below, and submit your result so the next version of this guide has your story in it.

The examples in this guide

Want more like these in your inbox? The weekly email below rounds up the five best new use cases. Or submit your own and help the directory grow.