The CitizenAI Accounts skill#

Tools alone do not tell an agent when to use them. The account skill does.

How the agent connects	Where the skill comes from
The setup link	The setup installs it
The ChatGPT plugin	The plugin adds it
The Claude connector, or an MCP server that you add yourself	No skill. The CitizenAI MCP server sends its main rules

What it covers#

Area	What the agent learns
Core routing	Which path a request belongs to, before calling anything
Public research and discovery	Reaching accountless public-read tools first when that is enough
Connected provider actions	Acting through an authorized provider connection
Provider connections and new accounts	Telling "connect what I have" apart from "register something new"
Accounts, identity, email, phone, wallets	Reading its own resources and provisioning what is missing
Credentials and safety	Handling a credential without leaking it

Why routing matters most#

The expensive mistakes are routing mistakes: starting a signup when the owner wanted to connect an existing account, or provisioning a paid capability when a public read-only tool would have answered the question.

Intent has to be explicit. A request to connect an existing account never routes to signup, and new-account registration only happens when you actually asked for it.

Safety rules it carries#

• A checkout link is relayed only in a private one-to-one conversation with you.
• A stored password is never shown, not to you and not to the agent.
• Wallet private keys are never requested, because they are never returned.

Keeping it current#

A machine keeps the skill that it installed. A CitizenAI release does not change that copy.

• Setup link. Send the setup link to the agent again. The setup installs the current skill.
• Plugin. Update the plugin in the app.

If an agent behaves as though a capability does not exist, refresh the skill first. Then look for a platform problem.