Muse prompts for your calendar and inbox

By the Chat Judge team · Updated · 5 min read · Links go to prompts in our library; creators are credited on each prompt.

In short: Muse prompts read like short instructions to an assistant that can see your email, calendar and accounts. This guide collects the ones shared publicly for that kind of work, explains what each expects to be able to access, and flags where you should keep approval for yourself.

The daily briefing and the calendar

Morning day briefing and inbox triage, shared on X by Muse Cases, is three lines and does two jobs: a briefing of your calendar, bookings and reminders, then an inbox triage that flags bills, surfaces priority mail and names what needs a decision today. Its usefulness depends entirely on the assistant being able to read those accounts. If it cannot, you will get a confident briefing built from nothing, so test it on a morning when you already know what is in the inbox.

Travel-time buffers for your calendar was posted by Trevin Chow as a prompt he runs with Muse and Grok Bot. It protects travel time so you can physically reach meetings that are not where you usually work, looking 14 days ahead and re-scanning on a schedule rather than pretending to watch the calendar continuously. It asks once for your usual workplaces and how you travel, then remembers them. That "ask once, then remember" pattern is worth copying into your own standing prompts.

If you want scheduling handled rather than just briefed, Executive assistant for scheduling meetings, shared by Michael for Grok Bot, sets an assistant up with its own email address and rules for when it may book. The setup steps are worth reading even if you use Muse, because they make the boundaries explicit.

Turning your inbox into a record

Build a personal CRM from your email and calendar, shared by jkpelaez, goes through a year of email and calendar, ranks people by how recently and how often you have been in touch, and builds one page per person. It is a large task, so expect to review the merges before you trust them. People with similar names and group threads are where this kind of pass goes wrong.

Briefing about you for another AI assistant, shared by Trevin Chow, produces a briefing so a second assistant can help without you re-explaining your context. He frames it as knowledge sharing rather than a migration, and it tells the assistant to use discretion with other people's private information and to leave out secrets. Read the result before you paste it anywhere. You are the only person who knows which details should not move.

Errands: refunds, warranties, tax paperwork

Muse Cases has shared several prompts that are really errands with a paper trail. Find duplicate hotel charges looks for duplicate charges and drafts a dispute email for each, and it ends with the line that matters: do not contact anyone without approval. Cancel a home internet plan asks the assistant to confirm the cancellation date and the refund in writing, and not to accept a new offer without asking first. File an extended-warranty claim checks whether the card you paid with extends the warranty and, if so, files and tracks the claim.

Gather tax and income documents, shared by Bill D'Alessandro, compares this year's folder with last year's and with your accountant's email to find what is still missing. It is a good model for any recurring paperwork: point at last year's example and ask what is absent, rather than asking the assistant to decide what you need from scratch.

One more, for completeness: Build a simple company website, from Muse Cases, is a three-line brief for a site with a menu and an order button that saves every version. Short briefs like this work when you are happy to iterate out loud afterwards.

Write the rules of the road once

Before any of these prompts, write five lines your assistant must always follow: which accounts it may read, what it may draft but not send, what needs your approval every time (anything that pays, cancels, signs or contacts a person), and how it should behave when it lacks access. It should say so rather than guess. Several prompts here already carry part of this. The hotel-charge prompt forbids sending without approval, and the cancellation prompt forbids accepting a new contract. Copy those sentences into prompts that lack them.

Then test with a case you know the answer to. Ask for the morning briefing on a day you have already read your inbox, and compare. Ask it to find a duplicate charge you planted in a statement. An assistant that passes both is worth trusting with the next step; one that invents either is not worth connecting to more accounts. Audit every permission your bot has is written for Grok Bot, but the questions it asks, what can you do, who granted it, is it still needed, are the right questions to ask of Muse too.

How to get good results from any of these prompts

  1. Fill in every placeholder. Anything in [BRACKETS] is a fact only you know. Leave one empty and the model will guess.
  2. Tell it what not to invent. Add "use only the facts above, and flag anything missing" if the prompt does not already say it.
  3. Ask for one version first. Fix the direction, then ask for variations.
  4. Check every name, number and promise before anything leaves your hands.
  5. Edit it into your own voice. Read it aloud once; if it does not sound like you, change it.

Prompts in this guide

Questions people ask

What does Muse need access to for these prompts?

Each prompt assumes access to the accounts it mentions, such as email, calendar, bookings, cards or a drive folder. If the assistant cannot see them, the output will not be reliable.

Should an assistant send emails or accept offers on my behalf?

Keep approval for anything that contacts people, accepts terms or spends money. Several of these prompts include that rule; add it to the ones that do not.

Can I use these with another assistant?

Yes, if it can reach the same accounts. The instructions are plain text.

Keep exploring

More guides

Written by the Chat Judge team from prompts in our library. We do not run model benchmarks, and nothing here is a claim about results you will get. Spot a problem? Tell us or submit a better prompt.