Title: ReasonAlloc: Hierarchical Decoding-Time KV Cache Budget Allocation for Reasoning Models

Source: http://arxiv.org/abs/2606.11164v1

Summary:

ReasonAlloc introduces a hierarchical KV cache allocation strategy that significantly optimizes memory usage during the long chain-of-thought trajectories characteristic of modern reasoning models. By identifying "Reasoning Wave" demand patterns, this training-free framework provides a foundational primitive for scaling inference efficiency in complex reasoning tasks.

Title: Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory

Source: http://arxiv.org/abs/2606.09365v1

Summary:

This paper proposes SkeMex, a foundational architecture for self-evolving agent memory that enables the distillation and governance of procedural skills from interaction trajectories. It introduces a novel "Read-Write-Assess-Govern" reasoning loop that allows agents to continuously improve their capabilities post-deployment without the need for model retraining.

Title: End-to-End Context Compression at Scale

Source: http://arxiv.org/abs/2606.09659v1

Summary:

This paper introduces Latent Context Language Models (LCLMs), a novel architectural primitive that utilizes encoder-decoder compression to efficiently handle long-context sequences at scale. It establishes a new Pareto frontier for accuracy and efficiency, providing a foundational backbone for next-generation agents that require massive context windows.

Title: Capability-Aligned Hierarchical Learning for Tool-Augmented LLMs

Source: http://arxiv.org/abs/2606.09371v1

Summary:

This paper proposes Capability-Aligned Hierarchical Learning (CAHL), a novel framework that jointly optimizes high-level planning and low-level execution policies using reinforcement learning. It addresses the fundamental bottleneck of planner-executor misalignment, creating a more robust and foundational reasoning loop for tool-augmented agentic systems.

Title: Bittensor Agent Arenas as a Trajectory Primitive: Distilling a Shopping Agent from ShoppingBench Subnet Traces

Source: http://arxiv.org/abs/2606.10064v1

Summary:

This paper introduces the concept of Agent Arenas as a "trajectory primitive," establishing a novel framework for generating diverse, incentive-aligned training data for agentic post-training. This approach represents a significant breakthrough in scaling agent capabilities by moving beyond the limitations of synthetic data and unjudged production logs.

Title: When RL Fails after SFT: Rejuvenating Model Plasticity for Robust SFT-to-RL Handoff

Source: http://arxiv.org/abs/2606.09932v1

Summary:

This paper identifies and solves the critical 'loss of plasticity' bottleneck in the standard LLM post-training pipeline where excessive SFT inhibits subsequent RL optimization. It introduces 'Rejuvenation', a foundational training primitive that uses model fusion and neuron resets to enable robust reasoning gains during RL while preserving SFT-acquired knowledge.

Title: Trajectory-Refined Distillation

Source: http://arxiv.org/abs/2606.08432v1

Summary:

This paper identifies and mitigates 'prefix failure' in on-policy distillation, a structural issue that hampers the efficiency of reasoning-scale post-training. By introducing trajectory-level corrections, it provides a foundational efficiency breakthrough that improves exploration and reasoning accuracy for large language models.

Title: ConMem: Structured Memory-Guided Adaptation in Training-Free Multi-Agent Systems

Source: http://arxiv.org/abs/2606.08702v1

Summary:

ConMem establishes a novel framework for multi-agent adaptation using relation-aware memory graphs to distill and coordinate reusable strategies from historical trajectories. It represents a foundational advancement in agentic reasoning loops by enabling robust, training-free adaptation with significantly reduced planning overhead.

Title: VATS: Exploiting Implicit Authority in Error-Path Injection via Systematic Mutation

Source: http://arxiv.org/abs/2606.07992v1

Summary:

This study exposes a foundational vulnerability in agentic reasoning by identifying 'implicit authority' within error-handling loops as a primary vector for bypassing safety heuristics. It provides a critical analysis of the Model Context Protocol (MCP) and demonstrates how systematic mutations in tool feedback can compromise the integrity of autonomous agent workflows.

Title: Semantic Quorum Assurance: Collective Certification for Non-Deterministic AI Infrastructure

Source: http://arxiv.org/abs/2606.08021v1

Summary:

This paper establishes Semantic Quorum Assurance (SQA) as a new architectural primitive for the reliable governance of non-deterministic agentic infrastructure. It introduces a multi-agent consensus framework that shifts focus from deterministic state replication to the semantic validation of agentic intent, addressing a core challenge in autonomous system safety.