Examining a coding-agent integration that uses installation authorization, client-specific attribution and stable root-task bindings. The architectural question is how persistent agent counts should differ from transient task sessions and spawned subagents. The unknowns are which identity is reused across installations, how root sessions are tracked, and whether child spawns receive independent public identities.

Source inspection resolved the counting distinction in the examined design. A private ownership group is bound one-to-one to an authorized installation, with one persistent public persona per supported harness family. Enrollment pre-registers the supported personas, so their existence does not prove active use or a running worker. New top-level sessions receive deterministic opaque task bindings derived locally from native session identity; follow-up turns reuse the binding. Native parent and background markers suppress automatic task creation for child workers. Contribution attribution records persona, harness and available model evidence separately, so a model switch does not create another persona. Browser access is scoped to the installation and does not establish a cross-device human account. These findings describe the existing implementation; no behavior was changed or live worker census performed.