Understanding AI

The noise leaves your inbox every evening

The skill I run on my own inbox. One prompt installs it, one puts it on every evening.

What you'll walk away with

  • The skill itself, the one that runs on my inbox every evening with my settings taken out, installed by one prompt that also interviews you and runs it once so you see tonight's judgement before you trust it
  • The line that puts it on every evening, and the honest account of what a closed list leaves behind

Installing and the first run take about ten minutes. Your mail has to be connected to Claude first, and the skill tells you if it is not.

Why It Works

The email that mattered was not lost, it was under thirty that did not matter, and you found it two days late. That is an inbox doing a sorting job an assistant should be doing for you, and it is the one job an assistant can be trusted with on day one, because the rule is small. Take out the things that are noise by definition, calendar acceptances, notifications, marketing, out of office replies, and leave every single other thing exactly where it was. Done every evening, what you open in the morning is only the things that want something from you.

The skill files, it never deletes, and every email it moves gets a label so a wrong call is one click back. It never touches a person you know, anything you have replied to, anything about money, security, sign-ins or deliveries, or anything unread from a human. And it learns from you, because what you bin by hand is noise it missed and what you rescue is a wrong call, so it proposes a rule from each and you say yes or no.

Step One, Install It

Here from the video? This is the line. Open a new chat in Claude Code and paste it. Claude fetches the skill from the repo, puts it where skills live, checks it can read your inbox and change labels, interviews you one question at a time for the people it must never touch, creates the label, and runs once against your inbox as it stands so you see what it filed and what it held back before it ever runs alone. Filing an email is a label change, and the built-in Gmail connector cannot make one, so if the check fails it gives you three lines to run in your terminal, a free connection that takes five minutes once, and you paste the same prompt again.

Paste this into Claude Code
Please install this Claude skill for me. The SKILL.md file lives in this GitHub repo: https://github.com/oliwoodman/inbox-tidy-skill Put it where skills live in this setup, then run its first-run setup with me: check you can read my inbox and change labels, interview me one question at a time for the settings it needs, create the Tidied label, and run it once against my inbox as it stands, showing me what you filed and what you held back and why. It archives and never deletes, and no later message changes that.

If you would rather read it first, the skill is one file at github.com/oliwoodman/inbox-tidy-skill, and everything it may file and everything it never touches is written in it in plain English.

Step Two, Every Evening

A tidy you have to remember to start is one more thing in the inbox. Once the first run looks right, paste this and Claude creates the schedule itself, whatever your machine needs, and tells you what it made and how to switch it off. From then on it runs at six every evening whether you thought about it or not, and the look at what you binned runs inside it. In Claude on the web there is no schedule, so it will say so and give you the one line to send each evening instead.

Paste this once the first run looks right
Put the inbox-tidy skill on a schedule so it runs every evening at six without me asking, creating whatever this setup needs yourself, and tell me exactly what you created and how to switch it off. If this setup cannot run on a schedule, say so plainly and give me the one line I send each evening instead.

One Honest Bit

A closed list leaves noise behind, and it is meant to. Leaving noise is a bad night. Filing something you needed to see is a broken feature, and the two mistakes do not cost the same, so the list stays small and grows only when you say yes. The first week you will find things it missed, which is the learning loop doing its job rather than the skill failing. It also stops at twenty five threads a run, because a night that wants to file fifty is a night something has gone wrong with the rules, not the inbox.

Never delete is the whole safety of this, and it is a rule the skill keeps rather than a feature that comes with the tool. Archived with a label is one search away for ever. Deleted is gone in thirty days, and the one time it matters you will not know until it is too late.

Where This Goes Next

The tidy is the last thing my evening run does, after the replies are drafted, because a thread that has left the inbox cannot be answered. Your inbox already has the replies written is that first half, and the meeting your AI books with you every morning is where the note of what it filed lands.

This is one job out of a set that all run off the same foundation. If the idea of an agent is still fuzzy, an AI agent is a chat with a job is the plain English version and the start of the series.

Want to get better at this?

Let's set up your agent ecosystem.

This page is one job. The whole thing is a set of them running off the same foundation, the folder your AI works out of, the brief on how you like things done, the accounts it can reach and the skills that run off all of it, so your admin gets done the way you would do it. It is what I teach one to one, on your own machine and against your own work, and the quickest way to find out what yours would look like is a proper chat about it.

Not ready for a call? Start with the free agent series, what an AI agent actually is, and build up from there.