🤖 AI Summary
This study addresses the challenges of cross-session state loss and fragmented execution models in long-horizon agent tasks by proposing a unified execution model based on versioned authoritative states, coupled with a runtime mediator handoff strategy. The approach leverages a Cases/Tasks/Scenario Packs architecture to enable persistent state management and seamless context migration. Experimental results demonstrate that the system achieves a 93% success rate across three-stage tasks, significantly outperforming traditional methods while reducing token costs per successful execution. These findings indicate that the proposed framework effectively resolves state consistency issues in long-horizon agents, thereby enhancing both execution efficiency and operational robustness.
📝 Abstract
Agentic systems often organize execution and state around a single conversation, model invocation, or agent instance, even when real work spans many calls and stages. We introduce a unified execution model that maintains a work item's persistent identity and versioned authoritative state across calls. Each step receives input scoped to a specific state version and new material; a result advances state only after validation and recording. We implement selected paths of this model in AstronOS using Cases, Tasks, and Scenario Packs across central and local execution. We compare five complete strategies for carrying an established software-version update plan into a fresh model session: rereading original materials, replaying full history, deterministic text summary, deterministic JSON, and the AstronOS runtime-mediated handoff. Ten controlled tasks are run under all five strategies with three repetitions, yielding 150 included executions. On the single-stage reference family, strategies perform similarly. In the primary three-stage A-C batch, AstronOS passes the frozen scorer in 14 of 15 executions, compared with 0 of 15 for rereading and 2 of 15 for full-history replay; later non-interleaved summary and JSON batches each pass 0 of 15. AstronOS has lower attempt-accounted model-token cost per passing execution, while requiring more execution-window time per attempt. These results associate the complete AstronOS condition with higher end-to-end pass rates across fresh sessions in this benchmark, at a measurable time cost.