Why not just let an AI assistant manage my inbox?

Because you cannot audit what it did, and mail is exactly where a silent mistake is expensive. Written rules give you the time saving with a record of the reasoning.

How many rules is too many?

Past about ten you have stopped triaging and started building a system you will not maintain. If your residue is large, the honest answer is usually fewer rules and a bigger ASK ME bucket.

Workflow automation

Inbox triage rules

Do not ask an assistant to sort your inbox. Ask it to propose explicit triage rules you can read, argue with, and reuse every week.

Use this when the same kinds of message keep arriving and you keep making the same sorting decision by hand.

The pattern

Copy this and replace everything in [SQUARE BRACKETS] with your own specifics.

Prompt template

Here are [15-20] real items from my [INBOX / MESSAGES / NOTIFICATIONS], with sender and subject: [PASTE, anonymised.]

My situation: [WHAT I AM RESPONSIBLE FOR, what I must never miss, and what I am happy to ignore for a week.]

Step 1 — Propose triage rules as a numbered list. Each rule: the observable signal (sender, keyword, timing), the bucket it goes to, and the reason. Buckets: DO NOW / SCHEDULE / DELEGATE / ARCHIVE / ASK ME.
Step 2 — For each rule, show which of my sample items it would catch, and name the item it would catch WRONGLY.
Step 3 — List the items no rule covers. Those stay in ASK ME; do not invent a rule to absorb them.

Do not sort anything yet. I want the rules first.

Why each part is there

Real items, anonymised: Rules written from imagined mail sort imagined mail. Your actual senders and subject patterns are the whole input — with names and identifiers stripped first.
What I must never miss: Triage is a risk decision, not an efficiency one. The asymmetry between a missed exam notice and a missed newsletter is what shapes every rule.
The observable signal: A rule you cannot apply mechanically is not a rule. “Important-sounding” is a vibe; “from the school domain, subject contains 'fee'” is a rule.
The item it would catch wrongly: Every rule has a false positive, and asking for it up front is how you find the rule that would have archived something that mattered.
Items no rule covers: The honest residue. A rule set that covers everything has been stretched to fit, and the stretched rules are the ones that fail next month.

Before and after

The lazy version

Go through my inbox and tell me what is important.

What goes wrong: You get an opaque verdict with no reasoning you can inspect, no way to check what it skipped, and nothing reusable — so next week you ask again and get different answers on the same mail.

The pattern, filled in

Worked example

Here are 18 real items from my inbox, sender and subject only, anonymised: [pasted list]

My situation: I am a Class 12 student. I must never miss anything from the school domain about exams, fees or forms, or anything from two specific teachers. I am happy to ignore course marketing, coding-contest promos and newsletters for a week or more.

Step 1 — Propose triage rules as a numbered list. Each rule: the observable signal, the bucket, and the reason. Buckets: DO NOW / SCHEDULE / DELEGATE / ARCHIVE / ASK ME.
Step 2 — For each rule, show which sample items it catches, and name the item it would catch WRONGLY.
Step 3 — List items no rule covers. Those stay in ASK ME; do not invent a rule for them.

Do not sort anything yet.

What that buys you: Seven rules you can read in a minute, one of which you immediately reject because its false positive is a genuine internship email — and three uncovered items honestly left in ASK ME rather than forced into a bucket.

Now do it on your own work

Get the rules, then apply them BY HAND for one week and keep a tally: how many times each rule fired, and how many times you overrode it. Rewrite or delete any rule you overrode more than twice.

Check yourself — every box should be true

Every rule uses a signal you could check without judgement.
You rejected or edited at least one rule because of its false positive.
You ran them by hand for a week and have the tally.
You deleted or rewrote at least one rule based on the tally, not on how it sounded.

Then explain it back

You overrode one rule repeatedly. What signal were you using that the rule could not see, and can it be written down at all?

Write two or three sentences in your own words. If you cannot, you copied a prompt rather than learning a pattern — which is the whole difference this library is trying to make.

Does this depend on which AI tool I use?

Keep rule-writing and rule-running separate. Write them here; run them in your mail client's own filters, which are inspectable, reversible and do not send your mail anywhere. Granting an assistant live mailbox access is a much bigger decision than it appears — the rules are the valuable part, and they are portable.

Why it actually works

A numbered rule list with buckets, false positives and uncovered cases is a decision tree that you wrote by hand — including the overfitting you just found by tallying overrides.

The same idea, learned from data →

Prompting well is a skill you can pick up in a week. Understanding what is happening underneath is the 18-month curriculum — and it is what stays useful when the tools change.

FAQ

Why not just let an AI assistant manage my inbox?: Because you cannot audit what it did, and mail is exactly where a silent mistake is expensive. Written rules give you the time saving with a record of the reasoning.
How many rules is too many?: Past about ten you have stopped triaging and started building a system you will not maintain. If your residue is large, the honest answer is usually fewer rules and a bigger ASK ME bucket.

Next patterns

The Prompt Patterns Library is a separate, optional track — it does not change the 78-week curriculum or your roadmap. See what the curriculum actually covers →