A Step-by-Step Guide

I was spending 30+ minutes a day just going through my email (and I’m sure it’s not just me). Most of the time, it’s more about sorting, rather than reading → deciding what needs my input, what's worth reading, and what's noise.

So I built an AI agent that sorts my inbox, labels emails, kills the noise, and sends me a 30-second digest.

https://www.loom.com/share/e1ac46d5482e4dbca5de138344142b6c

https://youtu.be/cfUWI3kvdIM

🛠 What I used:

Claude Code (the AI that does the magic)
Prompt Engineering (SKILL.md → the brain + rules.md → the learning layer)
Gemini 3 (to write the Python script → the execution engine)
Google Gmail API (direct OAuth → for labeling, archiving, sending digest)

🚀 How I Built It

A 3-file system:

File	Purpose
`gmail_cli.py`	Connects to Gmail
`SKILL.md`	The brain → 3 labels, 3 tiers, digest format, guardrails
`rules.md`	Sender lookup table, with tiers + exceptions

1. Start with the problem, not the tool

I started by defining the simplest possible system. For this use case: 3 labels, 3 newsletter tiers, 2 steps. That's it.

3 Labels

Label	Meaning
🔥 Action	Reply, decide, or do something
📚 Read	Worth reading later
🗑️ Delete (from Email-Ops)	Noise — out of sight

A Step-by-Step Guide

🛠 What I used:

🚀 How I Built It

1. Start with the problem, not the tool

3 Labels

3 Newsletter Tiers