Docs Zorem AI Agent Connector WordPress Tools & Risk Ratings

WordPress Tools & Risk Ratings

A tool is one thing an agent can do on your site, such as Get posts or Reply to comment. Your Admin Assistant starts with the WordPress tools this plugin adds. You see them all, and set what the agent may do with each, in the agent’s Permissions section.

AI Agent Connector → Agent → Permissions

Setting What the Agent May Do

Every tool has three choices: None, Read or Write. A read-only tool offers None and Read only, because Write would mean nothing for it. Each section heading carries Grant all, All read and Revoke so you can set a whole section at once, and the All abilities row above the table does the same for every tool on the page, with Revoke all to clear them. Nothing is saved until you click Save permissions.

Risk Ratings

Every tool is rated automatically, so you can see at a glance what it can do:

OptionDescription
Destructive tools never delete permanently: they move posts, pages and items to the trash, where a person can restore them. They stay off until you turn on Allow destructive abilities in the agent’s Connection section.

The 43 WordPress Tools

OptionDescription

The agent also has Who am I, which tells the AI app which agent it is connected as, and the WordPress core tools for site, user and environment information.

Tools From Your Other Plugins

Any plugin that registers abilities with the WordPress Abilities API appears in its own section on the Permissions page as soon as it is active, rated like the rest. Those tools start at None: the agent gets none of them until you grant them yourself. When a plugin does not say how risky a tool is, it is rated from its name, and anything that cannot be rated counts as Write, never Read.

Safe Defaults

  • Drafts first. New posts and pages are saved as drafts unless the agent asks to publish them.
  • Trash, never delete. Deleting only moves to the trash.
  • Users are read only. The agent can list users but never create, change or remove them.
  • Protected content stays protected. Password-protected posts and pages return no content.
  • Each item is checked. A tool handed a specific post, page, comment or term checks that the agent may read or edit that item, not just that it holds the capability in general.