10.1Concepts

Operations

Every action the platform can perform against a directory is an operation. Operations are discovered automatically from the platform itself; you never add or remove them by hand. Each one has:

• a display name, such as "Delete User";
• a command name, the technical identifier shown in small type next to it (for example DeleteActiveDirectoryUserCommand), which you need when you refer to the operation in configuration;
• a category, such as the area of the directory it belongs to;
• a risk classification that groups it on the screen.

Operations are either commands, which change something, or queries, which only read. Only a command can make an agent wait for approval; a query changes nothing, so there is nothing to approve.

The three policy settings

Each operation carries three settings.

When an operation is disabled, any request for it is refused with HTTP status 400 and the error code Operation.Disabled. The refusal itself is recorded in the operations log with the status Blocked.

Note: "Always ask agents" never affects people working in the console. It is a pin that applies to API keys and connected AI agents only.

Risk tiers on the screen

The Directory Policies screen groups the operations in each category into risk tiers, so that you can see at a glance what matters most.

Hovering over an operation's name shows a one-line explanation of its risk, for example "Removes an object, or revokes memberships or grants — consider requiring approval."