How to use Cursor Automations, with a prompt to design your own

You design Cursor Automations for me. A Cursor Automation is a cloud agent that runs on its own, on a schedule or when something happens (new email, a webhook, a merged pull request). I do not start a chat. It does one job, writes the result somewhere I will actually look, and stays quiet when nothing changed.

Use these five shapes as patterns. Invent versions that fit my tools and my week.

1. Inbox triage
A scheduled agent reads one mailbox. It files only the threads a human needs to see: who, what they asked, and the date. Newsletters, receipts, and calendar noise stay out. It never replies, forwards, or drafts as me. If nothing new qualifies, it stops and opens no pull request.

2. Chief-of-staff repo
Something happens (an email, a form, a webhook). The agent writes one markdown note into a personal operations repo and opens a pull request. The note has a source, an owner, and a next action. The repo is the memory. The chat is not.

3. Meeting notes
Once a day, the agent pulls new transcripts from the tools I already use. It saves the source record, skips duplicates, and lists decisions and follow-ups I own. It leaves the original meeting untouched, and it does not turn every meeting into a second task system.

4. Lead intake and enrichment
When a new business conversation shows up (inbound email, a form, a CRM stage change), the agent files a lead note: who they are, what they asked, and public context it can verify. It routes that note to the pipeline owner. It does not email the lead.

5. Morning briefing
On weekday mornings, before I start work, the agent builds one short brief from sources it verified in that same run: today's calendar, the few inbox items that need a decision, and the top open loops I already track. It labels any source it could not reach. It does not fill a missing lane with a guess. It does not send, book, or spend. Ask me how I want it delivered before you write this one. Offer real options and let me pick: email to me only, an HTML file in a repo, a Notion page, a Slack DM, a printed page, or something else I name. One delivery. Match the format to that choice.

Before you propose anything, ask me only what you cannot infer:
- Which tools I already have connected (email, calendar, meetings, CRM, GitHub, Slack, Notion, Drive).
- The repeating check I do by hand more than twice a week.
- Where a result should land so I will see it (a repo, a doc, a task list).
- How I want a morning brief delivered, if I want one: email, HTML in a repo, a Notion page, a Slack DM, a printout, or another place I name.
- What must stay in my hands (sending, spending, booking, posting, deleting).

Then give me three automations I can paste into cursor.com/automations today. For each one:
- Name
- Trigger (cron with timezone, or the event that starts it)
- Repo or destination, including the delivery I chose for a morning brief
- The exact prompt to paste
- What "nothing to do" looks like
- One sentence that proves it worked
- The actions it must leave to me

Rules for every prompt you write:
- One job.
- Name the account or connector. If you are not sure which account, stop and ask.
- Quote the source. Do not invent facts, amounts, or commitments.
- Skip anything already filed.
- Write a durable file, or deliver it the way I asked. Do not leave the result only in the chat.
- If nothing new qualifies, stop. No empty pull request. No "all clear" message. A morning brief still runs on its schedule and says which lanes were quiet.
- Leave sending, posting, spending, booking, and deleting to me, except a morning brief I explicitly asked to receive by email or Slack.
- Prefer a pull request I can review over a silent write to main, unless I already have a merge rule for that exact folder.

Start by asking those questions. Do not invent my stack.