Agents and modes
Agent Control > Agents lists every credential that has called your site through the Abilities API, from its first call. Each row shows the label, type, user, mode, call and write counts, and when it was last seen.
Agent types#
| Type | What it is | Key |
|---|---|---|
| Application Password | A REST call signed with an Application Password. Its label is the password’s name. | app_password:<uuid> |
| Agent user | A user created by the WordPress AI plugin as an agent. | agent_user:<user id> |
| MCP STDIO | The MCP Adapter’s STDIO server run over WP-CLI, used by local desktop agents. | mcp_stdio:<user id> |
| AI Client | An AI feature on your site calling abilities in-process through the WordPress AI Client. Named after the plugin that started the prompt. | ai_client:<plugin> |
You can rename an agent; the label is used in emails and the audit trail.
Modes#
| Mode | What happens | When to use it |
|---|---|---|
| Paused | Every call is denied. | A new or unknown agent, or one that misbehaves. |
| Restricted (default) | Rules apply first, then the default decision for the ability’s risk level. | Most agents. |
| Trusted | Writes run without approval, still snapshotted and logged. Critical abilities still need approval unless a rule says otherwise. | A well-tested agent doing routine work. |
The order of decision is: paused, then a matching rule, then trusted, then the risk default.
Give each agent its own Application Password with a clear name. Then you can pause, trust or limit one agent without touching the others.
Was this article helpful?
Tell us what was missing
