Agent Lifecycle — References

Prior art for instance states, supervision, and reclamation, mostly from runtimes that solved this long before agents existed.

State models and reclamation

• Kubernetes pod lifecycle — an explicitly enumerated state model with phases, conditions, termination grace periods, and garbage collection of finished objects.
• Erlang/OTP design principles: supervision — restart strategies, child specifications, and the let-it-crash argument for recovering an instance rather than defending it. Cited by name; the documentation path has moved between releases.
• Orleans virtual actors — the model in which an instance is always logically alive and is activated and deactivated transparently, which removes most explicit lifecycle code and replaces it with placement concerns. Named without a link.

Durability across restarts

• Temporal documentation — durable execution with timers, signals, and in-flight versioning, including the constraints that durability imposes on how the code may be written.
• AWS Step Functions documentation — managed state machines with explicit execution states, history, and callback tokens for waiting on the outside world.

Operating long-lived processes

• Release It! (Michael T. Nygard) — stability patterns and antipatterns, including the resource-leak and blocked-thread failures that this capability exists to prevent. Cited by name; no canonical online edition is recorded.