AgenticSTS: A Bounded-Memory Testbed for Long-Horizon LLM Agents
Xiangchen Cheng, Yunwei Jiang, Jianwen Sun, Zizhen Li, Chuanhao Li, Xiangcheng Cao, Yihao Liu, Fanrui Zhang, Li Jin, Kaipeng Zhang
arXiv:2607.02255 · 55▲ · cs.AI, cs.CL
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
The paper introduces AgenticSTS, an LLM agent for the deck-building game Slay the Spire 2 that replaces the usual growing chat transcript with a bounded memory contract: every decision prompt is rebuilt fresh from five typed slots (protocol, state schema, game rules, episodic summaries, and triggered strategic skills) via typed retrieval, so no raw cross-decision history is appended. It also ships a testbed of 298 tagged trajectories, frozen memory snapshots, and analysis scripts so individual memory layers can be ablated in isolation.
Why it matters
For anyone building long-horizon agents, this is a concrete design point showing you can keep per-decision prompt size flat regardless of run length instead of letting a transcript balloon into hundreds of thousands of tokens. The measured cost gap against two accumulating-context agents is large (roughly 4x wall-clock per floor and one to two orders of magnitude fewer tokens per unit of progress), though the causal attribution is weak because the compared systems differ in many ways beyond memory.
Practical takeaway
If your agent loop currently appends full history every turn, consider distilling prior experience into typed, retrievable records with capped top-k retrieval so context stays bounded; the released archive lets you slice trajectories by condition and rerun the ablations yourself.
Key result
In the fixed-difficulty (A0) matrix, enabling the triggered strategic skill layer moved wins from 3/10 (no scaffold) to 6/10, but at N=10 per cell this is directional only (Fisher exact p is about 0.37, overlapping confidence intervals), so it is not statistically significant. The stronger evidence is operational: their bounded baseline won 3/10 versus 0/5 for two open-source transcript-accumulating StS2 agents on the same backbone, character, and difficulty, while using far fewer tokens, but those competitors are untuned community projects rather than matched baselines.
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.