When OpenAI launched Dots (always-on agents in ChatGPT that run in the background, connect to your apps, and ask before risky actions), I wanted to set one up and went looking for examples of what other people were telling theirs. I couldn't find any.

Each Dot comes down to three things: its instructions, the apps it can reach, and Custom Rules (Allow / Require approval / Block). The rules matter most, because they decide what the agent does without asking you. They're also the part nobody was sharing.

So I built DotBotDirectory.

Each entry has the Dot's job, its instructions, its connected apps, and its three rule lists. One click copies the instructions and rules together so you can paste them into your own Dot.

Where it stands:

- 26 starter entries, written by me from ideas people published in early Dot write-ups. Each one credits its source.

- None of them have been tested on a real Dot yet, and they're labelled "Starter - untested". I'll mark entries as tested once someone confirms one works.

- Copy counts start at zero, so nothing looks more popular than it is.

- Free, and not affiliated with OpenAI.

Two questions for anyone who has run a Dot:

1. What did you put in Require approval vs Block? That's where I think most of the value is.

2. Did your Dot ever do something you didn't expect? I'd like to turn those cases into rules other people can copy.

Happy to answer anything.