AgenticSTS proposes a bounded-memory contract for long-horizon LLM agents: each decision receives a fresh user message assembled through typed retrieval, with no raw cross-decision transcript appended. This keeps prompts bounded and allows individual memory or skill layers to be ablated. The authors instantiate the testbed in Slay the Spire 2, a stochastic deck-building game requiring hundreds of decisions. In a fixed-A0 comparison, adding triggered strategic skills increased wins from 3/10 to 6/10. The released package contains 298 completed trajectories, condition tags, frozen memory and skill snapshots, prompt records, and analysis scripts. The result is directional, not statistically conclusive.
No heat snapshots are available in the last 24 hours.