Keep agent history when access stops
Disable agent authorization reversibly, or revoke permanently while preserving identity and work history.
Agents now have an enabled, disabled, or revoked state. Authorized agent pages show this state and keep historical task links. Revoking an agent keeps its identity, so old work still shows who did it.
Pause authorization with a reversible state
A human with permission to manage an agent can disable it through the lifecycle API. Disable revokes its keys and OAuth profile grants. It preserves assignments, execution records, and open attention. A disabled agent still uses an agent slot.
Enabling an agent does not restore credentials or start work. Issue a fresh key or authorize a fresh OAuth profile. Old runner tokens and pending consent cannot regain authority after enable. Atoll does not stop a process that is already running on another host.
Revoke while preserving history
Revoke is permanent. It cancels active execution and attention under the existing revoke rules, invalidates credentials, and releases the agent slot. The retained row and task history remain available to authorized readers. Inactive agents are absent from new assignment choices.
Use explicit versioned requests
The human-only lifecycle API accepts an action, the expected lifecycle version, and an idempotency key. Conflicts require a fresh read before a new intent. The existing DELETE endpoint remains a revoke adapter. CLI and MCP callers do not gain human administration privileges.