Beyond Compaction: Structured Context Eviction for Long-Horizon Agents

• URL: http://arxiv.org/abs/2606.11213v1
• Date: Fri, 01 May 2026 18:39:02 GMT
• Title: Beyond Compaction: Structured Context Eviction for Long-Horizon Agents
• Abstract summary: We present Context Window Lifecycle (CWL), a context-management scheme that gives long-horizon LLM agents an effectively unbounded working horizon.<n>As a session accumulates history, CWL keeps the context within budget through graduated, semantically-aware eviction.<n>We describe the annotation protocol, the episode graph, the eviction policy, and the token-accounting loop, and evaluate CWL on long-horizon agentic benchmarks.
• Score: 0.0
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: We present Context Window Lifecycle (CWL), a context-management scheme that gives long-horizon LLM agents an effectively unbounded working horizon. As a session accumulates history, CWL keeps the context within budget through graduated, semantically-aware eviction: the agent annotates its trajectory as typed, dependency-linked episodes as work proceeds, and a deterministic, LLM-free policy evicts content in priority order within that structure when a token budget is exceeded. CWL preserves user turns and the exploratory context the agent is actively reasoning over, while aggressively shedding action episodes whose effects are already persisted in the environment, keeping active context near a stable ceiling that also avoids the performance degradation associated with very large prompts. Compared to summarization-based compaction, CWL avoids four well-known limitations: unpredictable lossiness, destruction of causal structure, blocking model cost, and compression-induced hallucination. Compared to recency truncation, CWL is semantically aware: it drops the oldest-and-most-recoverable content according to the dependency graph rather than oldest-in-time regardless of relevance. We describe the annotation protocol, the episode graph, the eviction policy, and the token-accounting loop, and evaluate CWL on long-horizon agentic benchmarks: a single agent session completing 89 sequential tasks across 80 million tokens with no measurable degradation in task accuracy relative to per-task isolated sessions

Related papers

• From Profiling to Synthesis: Benchmarking Implicit Behavioral Alignment in Personalized LLM Agents [56.187590188698955]
  Large Language Models have enabled increasingly capable autonomous agents, yet personalization remains critical for making such agents practically useful.<n>Recent benchmarks have begun evaluating personalization in agents, but they largely rely on static preference snapshots, fixed interaction logs, or question answering over predefined user profiles.<n>We introduce IBA-Bench, a benchmark for implicit behavioral alignment constructed from longitudinal interaction histories.<n>We propose IBA-Agent, an agent framework that reconciles conflicting priorities through broad retrieval and trajectory-level alignment.
  arXiv  Detail & Related papers  (2026-08-03T12:52:04Z)
• ACM: Agentic Context Management for Long Horizon Tasks [27.337726990282928]
  Agentic Context Management (ACM) is a framework that equips agents with purpose-built context editing tools.<n>Inspired by the interaction between short-term and long-term human memory, the agent autonomously decides when to compress its context.
  arXiv  Detail & Related papers  (2026-07-26T19:19:24Z)
• ACE: Pluggable Adaptive Context Elasticizer across Agents [65.45147906718154]
  Adaptive Context Elasticizer (ACE) orchestrates historical step information into the agent's context at each decision step.<n>ACE consistently outperforms truncation and summarization baselines.
  arXiv  Detail & Related papers  (2026-06-30T12:20:45Z)
• Governance Decay: How Context Compaction Silently Erases Safety Constraints in Long-Horizon LLM Agents [0.4855342112648282]
  We show that in-context governance constraints reliably obey while visible can be silently removed by compaction.<n>Across 1,323 episodes, violation rises from 0% with the policy in full context to 30% after compaction.<n>We propose Constraint Pinning, a training-free mitigation that quarantines governance constraints from lossy compaction.
  arXiv  Detail & Related papers  (2026-06-21T14:30:20Z)
• IS-CoT: Breaking the Long-form Generation Collapse via Interleaved Structural Thinking [50.381430711141995]
  We introduce the Interleaved Structural Chain-of-Thought (IS-CoT) framework.<n>IS-CoT embeds a dynamic Plan-Write-Reflect cycle into the generation process.<n>We construct a high-quality dataset of interleaved reasoning traces via a multi-teacher pipeline and train IS-Writer-8B.
  arXiv  Detail & Related papers  (2026-06-08T16:31:00Z)
• Agent-BRACE: Decoupling Beliefs from Actions in Long-Horizon Tasks via Verbalized State Uncertainty [70.43119366710778]
  We introduce Agent-BRACE: Agent Belief state Representation via Abstraction and Confidence Estimation.<n>It is a method that decouples an LLM agent into a belief state model and a policy model, jointly optimized via reinforcement learning.<n>It achieves an average absolute improvement of +14.5% across long-horizon, partially observable language environments.
  arXiv  Detail & Related papers  (2026-05-12T02:37:04Z)
• Slipstream: Trajectory-Grounded Compaction Validation for Long-Horizon Agents [8.880437352929315]
  Slipstream is a trajectory-grounded compaction system that uses a judge to validate the candidate summary against the agent's continued reasoning.<n>It improves task accuracy by up to 8.8 percentage points while reducing end-to-end latency by up to 39.7%.
  arXiv  Detail & Related papers  (2026-05-09T00:47:43Z)
• ClawVM: Harness-Managed Virtual Memory for Stateful Tool-Using LLM Agents [0.7377939119373589]
  textscClawVM is a virtual memory layer that manages state as typed pages with minimum-fidelity invariants.<n>textscClawVM eliminates all policy-controllable faults whenever the minimum-fidelity set fits within the token budget.
  arXiv  Detail & Related papers  (2026-04-11T21:38:15Z)
• UT-ACA: Uncertainty-Triggered Adaptive Context Allocation for Long-Context Inference [41.15520696680647]
  Long-context inference is challenging for large language models due to attention dilution and out-of-distribution degradation.<n>We propose Uncertainty-Triggered Adaptive Context Allocation (UT-ACA), an inference-time framework that dynamically adjusts the context window based on token-wise uncertainty.<n>Experiments show that UT-ACA substantially reduces average context usage while preserving generation quality in long-context settings.
  arXiv  Detail & Related papers  (2026-03-19T03:17:50Z)
• V-CAGE: Context-Aware Generation and Verification for Scalable Long-Horizon Embodied Tasks [6.820118518027692]
  V-CAGE is a closed-loop framework for generating semantically aligned manipulation datasets at scale.<n>We propose a context-aware instantiation mechanism that enforces geometric consistency during scene synthesis.<n>We also employ a hierarchical instruction decomposition module to bridge the gap between abstract intent and low-level control.
  arXiv  Detail & Related papers  (2026-01-21T16:41:51Z)
• Deriving Character Logic from Storyline as Codified Decision Trees [67.01182739162142]
  Role-playing (RP) agents rely on behavioral profiles to act consistently across diverse narrative contexts.<n>We propose Codified Decision Trees (CDT), a data-driven framework that induces an executable and interpretable decision structure from large-scale narrative data.
  arXiv  Detail & Related papers  (2026-01-15T05:12:43Z)
• InfiAgent: An Infinite-Horizon Framework for General-Purpose Autonomous Agents [36.740230738304525]
  InfiAgent keeps the agent's reasoning context strictly bounded regardless of task duration.<n>InfiAgent with a 20B open-source model is competitive with larger proprietary systems.
  arXiv  Detail & Related papers  (2026-01-06T17:35:57Z)
• AgentFold: Long-Horizon Web Agents with Proactive Context Management [98.54523771369018]
  LLM-based web agents show immense promise for information seeking, yet their effectiveness is hindered by a fundamental trade-off in context management.<n>We introduce AgentFold, a novel agent paradigm centered on proactive context management.<n>With simple supervised fine-tuning, our AgentFold-30B-A3B agent achieves 36.2% on BrowseComp and 47.3% on BrowseComp-ZH.
  arXiv  Detail & Related papers  (2025-10-28T17:51:50Z)
• CLAUSE: Agentic Neuro-Symbolic Knowledge Graph Reasoning via Dynamic Learnable Context Engineering [47.77119771502664]
  We introduce CLAUSE, an agentic three-agent neuro-symbolic framework that treats context construction as a sequential decision process over knowledge graphs.<n> CLAUSE employs the proposed Lagrangian-Constrained Multi-Agent Proximal Policy Optimization (LC-MAPPO) algorithm to coordinate three agents: Subgraph Architect, Path Navigator, and Context Curator.<n>Across HotpotQA, MetaQA, and FactKG, CLAUSE yields higher EM@1 while reducing subgraph growth and end-to-end latency at equal or lower token budgets.
  arXiv  Detail & Related papers  (2025-09-25T11:43:08Z)
• END: Early Noise Dropping for Efficient and Effective Context Denoising [60.24648712022382]
  Large Language Models (LLMs) have demonstrated remarkable performance across a wide range of natural language processing tasks.<n>They are often distracted by irrelevant or noisy context in input sequences that degrades output quality.<n>We introduce Early Noise Dropping (textscEND), a novel approach to mitigate this issue without requiring fine-tuning the LLMs.
  arXiv  Detail & Related papers  (2025-02-26T08:07:17Z)
• Enhancing Contrastive Learning with Noise-Guided Attack: Towards Continual Relation Extraction in the Wild [57.468184469589744]
  We develop a noise-resistant contrastive framework named as textbfNoise-guided textbfattack in textbfContrative textbfLearning(NaCL) Compared to direct noise discarding or inaccessible noise relabeling, we present modifying the feature space to match the given noisy labels via attacking.
  arXiv  Detail & Related papers  (2023-05-11T18:48:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.