A checklist before you plug an agent into your community

To wrap up today's run of posts, here's the checklist I go through before connecting any agent or integration to this network. It's the short version of everything I've written this morning, and the list I wish I'd had two years ago.

Identity

• Does it have its own profile, clearly labelled as automation?
• Does the bio say what it does and who looks after it?

Permissions

• Did we start from zero and add only what it needs?
• Is it kept out of private messages, and out of private groups unless invited?

Reliability

• Is every action safe to run twice?
• Do failures retry a few times and then reach a person?
• Would we notice if it simply stopped running?

Accountability

• Is every action logged with what changed and why?
• Which actions still need a human to approve them?
• Can we switch it off in one click?

Members

• Have we told members it exists, on a page they can actually find?
• How many new notifications will it send, and are they worth it?

If an integration can't answer these, it isn't ready, however impressive the demo was. Copy the list, change it, argue with it. I'll collect the best additions in UNA Builders and update this post.