LLM Agent Evaluation

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

115 papers in the last four weeks, up 140% on the four weeks before. 1.1% of all new papers.

Jul 13Week of Sep 28

Latest papers 793

Jul 15, 2026cs.CL

Analogical Deep Research: Retrieving and Integrating Historical Analogies for Foresight Analysis

Systematic comparisons between current situations and structurally similar past events in the historical, i.e., historical analogies, is among the most powerful tools for foresight analysis. In this work, we present a new task called Analogical Deep Research (ADR) to Large Language Model (LLM) agents and construct the first ADR benchmark ADR-bench to study whether LLM agents are able to find and leverage historical analogies when doing foresight analysis. Our investigation reveals a key obstacle: LLM agents are poor at finding analogies because they match on surface features rather than underlying mechanisms. We argue that ADR is inherently a causal question as it requires understanding why the event occurred. Based on our theoretical analysis, we propose two principles required for ADR, including the mechanism alignment and cross-analogy confirmation. Built upon our theoretical results, we propose a new agentic framework called Causal Analogical Researcher (CANA) that guides LLMs to find and integrate historical analogies. CANA incorporates a simple yet effective structural decomposition representation, and integrates structural feedback for reflective improvements of historical analogy identification and integration. We show that CANA brings up to 10% improvements in historical analogy generation, and surpasses the state-of-the-art deep research agents in the ADR-bench. Case studies with the ongoing events confirm the effectiveness of CANA in leveraging historical analogies.

Yongqiang Chen, Guangyi Chen, Yuewen Sun +1

1Mohamed bin Zayed University of Artificial Intelligence · 2Carnegie Mellon University

Analogical Reasoning LLM Agent Evaluation Causal Reasoning in Language Models Autonomous Research Agents

Jul 15, 2026cs.SE

Structured Feedback Improves Repair in an LLM Agent Loop

LLM agents often retry after external validation rejects a candidate, but the interface between validation and the next model call remains underspecified. We introduce VeriHarness, a code-controlled agent loop in which models generate candidates while external validators control acceptance, budgets, and traces. We use it to compare raw diagnostics with feedback that identifies the failure location, observed value, and admissible alternatives. Across 50 paired TextWorld games under a four-call cap, feedback containing all three fields raises terminal success from 14/50 to 36/50 for Qwen2.5-Coder-14B (+44 percentage points) and from 8/50 to 29/50 for Llama-3.1-8B (+42 points). Ablations locate most of the gain in the admissible alternatives: feedback containing only the location and observed value remains near the raw diagnostic baseline. Presenting the complete repair information in prose instead of a keyed JSON record yields nearly the same success, providing no evidence that JSON syntax itself improves repair. The ordering persists across the tested call budgets and one sampled-decoding setting.

Jaideep Ray, Ankit Goyal

Independent Researcher

LLM Self-Correction LLM Agent Evaluation LLM Agent Workflow Optimization LLM Agent Verification LLM Agent Harnesses LLM Agent Reliability

Jul 15, 2026cs.CL

DevicesWorld: Benchmarking Cross-Device Agents in Heterogeneous Environments

LLM-based agents have rapidly improved at operating individual digital environments such as mobile applications, desktop systems, and smart homes. However, real-world user goals often span multiple devices: information may come from a phone, be processed on a desktop, and the result may need to appear on another device. Most existing benchmarks center on a single dominant execution environment, making it difficult to evaluate whether agents can acquire and integrate information across heterogeneous devices and complete end-to-end tasks with cross-device dependencies. We introduce DevicesWorld, a large-scale executable benchmark for cross-device collaborative operation. DevicesWorld contains 6,140 tasks and integrates three classes of device environments -- mobile, desktop, and IoT -- into a unified cross-device interaction and evaluation framework. Each task defines a natural-language user goal, participating devices and initial states, executable actions, rule-based verifiers, and a cleanup procedure. A multi-stage construction and quality-control pipeline keeps tasks close to realistic user needs while allowing final outcomes to be automatically verified from device states and generated files. We evaluate five frontier LLM-agent systems on a fixed evaluation set. All methods achieve low success rates, with the best reaching only 12.5%. Among failed runs, about 28.7% satisfy at least one scoring condition yet still fail the full task. Trajectories show that agents become stuck acquiring information or manipulating interfaces, confuse source and output devices, or terminate before all conditions are jointly satisfied. DevicesWorld turns cross-device collaborative operation into an executable, reproducible, and diagnostically useful evaluation problem for research on reliable cross-device agents.

Huatao Li, Xinwei Geng, Yuheng Wang +9

School of Artificial Intelligence, Shanghai Jiao Tong University · 2Honor Device Co., Ltd

Computer-Use Agent Benchmarks LLM Agent Evaluation AI Agent Benchmarks LLM Agent Reliability

Jul 15, 2026cs.AI

Set-shifting Behavioral Test for Harnessed Agents

What happens to an LLM agent's tool choice when the reliable tool silently changes within an ongoing session? We borrow set-shifting from cognitive psychology to study how well agents adapt to hidden reliability shifts. Our benchmark mounts tool-skill libraries with redundancies, where many tools solve the same task but differ in hidden reliability. In our evaluation framework, a branched schedule shifts the reliable tool group at hidden boundaries and pairs every shift with a no-shift control. We find that agents, by default, settle on a small recurring routine within a few turns of each boundary, with call shares concentrating on a few discrete values after each reliability shift. We score the set-shifting accuracy for each agent trajectory: the joint probability of routing to the target tool group in every post-shift window. We test open-weight LLMs in an open-source agentic harness and find qualitatively distinct failure modes across the same set of routines. We also find that set framing, how the toolset presents the alternatives as competing or complementary, shifts the routing dynamics.

Ziwei Ye

LLM Agent Evaluation AI Agent Benchmarks Tool-Using Agents LLM Agents LLM Agent Routing

Jul 15, 2026cs.CY

Learning Engagement Assistant (LEA): Cross-Course Scalability and Classroom Evaluation of an Agentic AI Tutoring System

This paper is an extension of a paper presented at the ICAART 2026 conference, which introduced LEA (Learning Engagement Assistant), an adaptive AI tutoring agent combining course-specific Retrieval-Augmented Generation (RAG) with structured Knowledge Component (KC) models across integrated Chat, Tutor, and Quiz modes. That prior work validated LEA on a single STEM course (CMP511) exclusively through simulation, using synthetic learner agents. This paper extends that work by reporting the first classroom deployment of LEA with real students (n = 8, CMP511) and the first empirical test of its cross-course scalability, deploying the system across three courses spanning two academic levels and two disciplinary domains. The study reveals a divergence from simulation predictions across modes, showing that synthetic evaluation alone cannot anticipate all aspects of real deployment. A RAGAS-based cross-course scalability evaluation (660 questions) finds Answer Relevancy and Context Precision broadly stable across courses (0.88-0.94 and 0.88-0.90 respectively), while Faithfulness declines with curriculum distance from the system's original course (0.69 to 0.50), a preliminary finding that may reflect generation logic tuned to the system's original subject rather than a scalability limitation. These findings suggest that while the orchestration layer requires no modification, full course-agnosticism of all downstream components requires further investigation.

Teri Rumble, Javad Zarrin, P. George Lovell +1

Faculty of Design, Informatics and Business, Abertay University, Dundee, United Kingdom.

Retrieval-Augmented Generation Intelligent Tutoring Systems AI in Education LLM Agent Evaluation Cross-Domain Generalization

Jul 14, 2026cs.AI

Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution

Large language model (LLM) agents increasingly automate multi-step engineering and informatics workflows, yet they rarely ask how much effort a task actually requires. They often follow a maximum-context-first strategy--re-reading files and dependencies they have already seen--turning a one-line edit into a small code-base audit. We argue the missing capability is task-aware execution-scope estimation: judging a task's difficulty, the information it truly needs, and the shortest reliable path before committing budget. We formalize minimum-sufficient execution and the Agent Cognitive Redundancy Ratio (ACRR), and propose E3 (Estimate, Execute, Expand): the agent estimates an initial operating point, executes a minimum viable path, and expands scope only when verification fails. On MSE-Bench--a deterministic benchmark of 121 edits in a capability-controlled simulator--E3 matches the strongest baseline's 100% success while cutting cost by 85%, tokens by 91%, and inspected files by 92%, and further beats a strong adaptive retrieval baseline by 16%; the gains survive held-out instruction wording and essentially every cost weighting. A companion real-model harness (LLM-Case) corroborates the effect on a live gpt-4o agent editing a real open-source library, with every candidate patch graded by actually running the project's real pytest suite against a measured oracle: the over-reading is milder but real, and E3 is the leanest and fastest policy at comparable task success--its one shortfall a provider rate-limit, not a wrong edit. We frame this as a controlled probe of execution redundancy, not a measurement of any deployed agent, and position task-aware execution as a step toward engineering-grounded AI (EGAI)--agents whose effort is anchored in the engineering reality of the task. We release the framework and benchmark.

Junjie Yin, Xinyu Feng

University of Tennessee, Knoxville, TN 37996, USA · Microsoft, Redmond, WA 98052, USA

Software Engineering Agents LLM Agent Evaluation LLM Agent Workflow Optimization LLM Agents

Jul 14, 2026cs.AI

Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan Evaluation

Plan evaluators can reward a strategic plan for becoming less explicit. This paper studies that failure in a staged expected-value scorer for LLM-generated venture routes. Proposition 1 gives the score change from deleting an interior transition while retargeting its predecessor and retaining downstream value: Delta_k = (prod_{i<k} p_i)[c_k + (1 - p_k)R_{k+1}]. On a frozen 26-route cohort, all 57 admissible deletions matched the analytic identity and threshold sign, and every route had at least one score-improving deletion. A score-seeking optimizer, allowed to restructure routes but not told the exploit mechanism, found baseline-beating uncovered structures in 21/26 routes. GATE refused score release for 26/26 silenced routes with 0/26 honest suspensions; after refusal, 47/54 next revisions repaired to a covered structure, and strict covered improvement rose from 1/26 to 13/26. An adaptive compiler-aware co-author exposed the registry-provenance boundary: obligation-channel evasions remained 6/6 across all four v1/v1.5 conditions, while delta-indexed cost floors reduced beat-honest routes from 6/6 to 3/6 and fundability-by-silence from 5/6 to 0/6 without establishing semantic completeness. If a plan scores better only because it omits necessary work, the plan did not improve; the evaluation created an omission incentive. PCSC detects and neutralizes post-hoc omission splices over model-mediated typed-state records. In the cooperative setting tested, GATE acts as a deterministic search-shaping constraint, not merely a post-hoc filter. It does not verify the semantic completeness or real-world quality of arbitrary LLM-generated strategies.

Aleh Manchuliantsau

Independent Researcher

Reward Hacking LLM Agent Evaluation

Jul 14, 2026cs.AI

Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents

Self-evolving agent systems improve by creating, revising, and retiring their own skills, but every such loop rests on a hidden assumption: a reliable evaluation metric already exists. In many real applications it does not. We make three claims. First, metrics can be \emph{evolved}: our metric loop searches compositions of small drawback detectors under a full evolutionary lifecycle, trained to agree with a ten-item anchored reference set, regularized by consensus over unlabeled outputs, and audited against a held-out anchor it never reads, yielding a transparent, inspectable metric rather than an opaque judge. Second, since no metric exists to beat, the yardstick is recovering what an accurate metric would have enabled, and \emph{Double Ratchet}, our co-evolution of the metric with a lifecycle-managed skill loop, does so: across code generation (MBPP+), enterprise text-to-SQL (Spider~2.0-Snow), and reference-free report generation, it retains 88--110% of the held-out lift achieved by the same skill loop driven by ground truth or the best available rubric. Third, safety comes from anchor discipline plus outer audits: removing anchor guards collapses the metric into a vacuous detector while removing the lifecycle does not; and when evolved skills gamed the report rubric, an independent judge caught it, one detector repaired it, and a task-aware judge then preferred the evolved outputs over the pre-evolution baseline in 77% of decided pairs. We argue this failure-expecting architecture is the right default wherever no reliable automatic verifier exists.

Xing Zhang, Guanghui Wang, Yanwei Cui +4

1AWS Generative AI Innovation Center · 2HSBC Holdings Plc., HSBC Technology Center, China

LLM Agent Evaluation LLM Agent Self-Improvement LLM Agent Skill Learning Metric Learning Rubric-Based Evaluation

Jul 14, 2026cs.AI

Critic Experience Bank: Self-Evolving Step-Level Confidence Estimation for LLM Agents

LLM agents act in external environments where each action changes the state that later decisions condition on, and where a single wrong step can waste interaction budget or trigger irreversible side effects long before the final failure is observed. Reliable deployment therefore requires \emph{step-level confidence estimation}: a calibrated probability that each proposed action is productive, available \emph{before} the action is executed. Existing LLM confidence estimators are designed to score a response from the given prompt, but agent confidence also depends on execution consequences: whether similar actions in similar situations actually advanced the task after the environment responded. We introduce the \method (\methodshort), a self-evolving critic framework in which an LLM critic accumulates evidence from its own past judgments and their observed consequences. After each trajectory, a hindsight LLM that sees the full execution feedback votes on whether each step was productive. The resulting pseudo-labels populate a memory bank from which related productive and unproductive experiences are retrieved into the critic's prompt whenever a similar step recurs. \methodshort requires no training and uses no ground truth step labels. Across three agent benchmarks and three critic backbones, \methodshort attains the best calibration (ECE and Brier) and ranking (AUC) in every dataset--critic combination, reducing ECE by up to

54\%

relative to the strongest training-free baseline.

Yaopei Zeng, Congchao Wang, JianHang Chen +3

Pennsylvania State University · Virginia Tech · Purdue University +1

AI Agent Reliability LLM Agent Evaluation LLM Uncertainty Estimation LLM Agent Reliability Memory-Augmented Language Model Agents

Jul 14, 2026cs.AI

PM-Bench: Evaluating Prospective Memory in LLM Agents

A significant challenge in agentic AI is prospective memory: the ability to execute an intention at a specific future cue or state while other activities are ongoing. We introduce PM-Bench, a text-based benchmark for measuring prospective memory capabilities in modern LLM agents. Inspired by the Virtual Week paradigm from cognitive science, PM-Bench evaluates how well LLM agents maintain user intentions, execute delayed intentions, and monitor latent environment changes. Over the course of a simulated seven-day week, agents must continue an ongoing activity while deciding whether any deferred task is due. We compare eight state-of-the-art LLMs on PM-Bench under eight different agent configurations. PM-Bench proves challenging across all settings: the best method, a GPT-5.4 agent, reaches only 65.1% F1 score under our evaluation. Furthermore, no single strategy for improving prospective memory dominates across models. We release PM-Bench as a controlled testbed for diagnosing these failures and developing training or inference-time interventions that support reliable prospective behavior.

Genglin Liu, Saadia Gabriel

University of California, Los Angeles

LLM Agent Memory LLM Agent Evaluation Memory Agent Benchmarks

Jul 14, 2026cs.AI

How Many Tasks Are Enough for Agent Benchmark Decisions? A Replay Analysis of Public LLM Agent Benchmarks

Agent benchmarks often compare two agents after all tasks have run, but costly evaluations make partial runs tempting. A task fraction alone does not show whether a partial run supports the same pairwise conclusion as the completed benchmark. We study this question by replaying completed public task-level records from SWE-bench, AppWorld, and tau-bench. A partial budget counts as enough only when it supports the completed benchmark's decision, covers required task groups, and leaves no more than a target fraction of comparisons unresolved. The required task fraction varies sharply. At the strict 0 percentage point threshold on a 5 percentage point budget grid, AppWorld first meets all targets at 15 percent, tau-bench at 25 percent, and SWE-bench Verified at 90 percent; SWE-bench Lite does not meet all targets by 95 percent under the primary coverage rule. Partial-evaluation reports should state how much one agent must outperform another, how tasks are selected, what coverage rule is required, what decision rule is used, and how many comparisons may remain unresolved.

Wei-Jung Huang

Independent Researcher · United States

Pairwise Comparison Benchmark Design Software Engineering Benchmarks LLM Agent Evaluation

Jul 14, 2026cs.CL

Fin-Analyst at FinMMEval 2026 Task 3: A Live Hybrid Trading Agent with LLM Specialists and Rule-Based Signals

Large language model (LLM) trading agents show promising performance in equity markets, yet remain narrowly focused on US equities with little evidence from live deployment. We present Fin-Analyst, a hybrid agent for FinMMEval 2026 Task 3: an eight-specialist LLM pipeline over news, SEC filings, fundamentals, analyst forecasts, technical indicators, and social sentiment, aggregated by a Meta-Agent for Tesla (TSLA), and a lightweight rule based three-signal vote for Bitcoin (BTC). On the final official leaderboard (accessed 2026-07-05), Fin-Analyst ranks first of all agents on TSLA with a +13.51% return, +28.33 points over Buy-and-Hold (Sharpe 4.10, 88% win rate), while the BTC vote ends flat yet well above a sharply falling baseline. Relative to the interim performance, the asset ranking reversed, indicating that short live windows yield volatility-sensitive rankings. Ablation identifies event-driven 8-K disclosures as the most influential TSLA signal. Error analysis shows that the memoryless agents repeat wrong calls for days at a time, and that the fixed-threshold BTC rules lost money by trading on noise in a sideways market while the LLM pipeline gained under similar conditions, motivating a memory-aware, LLM-based successor for both assets.

Mohotarema Rashid, Lingzi Hong, Junhua Ding +1

Department of Information Science, University of North Texas, Denton, TX, United States · Department of Data Science, University of North Texas, Denton, TX, United States

Multi-Agent LLM Systems LLM Agent Evaluation AI Agent Benchmarks LLM Agents Finance Algorithmic Trading

Jul 14, 2026cs.AI

Rethinking the Evaluation of Harness Evolution for Agents

Harness evolution is an iterative search procedure that repeatedly evaluates and revises candidate harnesses used for LLM agents using task feedback. We revisit the evaluation of such automatic harness evolution procedures and identify two fundamental issues in the protocol. First, prior work does not compare these approaches with simple task-level search baselines under matched feedback and inference budgets. Second, prior work searches for harness configurations using verification signals (e.g., unit test cases) drawn from the same benchmarks on which it reports the final performance of the evolved harnesses, violating the standard separation between training and test data. To address this, we compare automatic harness evolution with simple test-time scaling and discovery baselines under comparable feedback and inference budgets, and evaluate evolved harnesses on held-out tasks to assess generalization. Following prior work, we experiment on Terminal-Bench 2.1 and find that automatic harness evolution fails to outperform simple test-time scaling methods both with and without test cases, and exhibits limited generalization. However, we find that long-horizon games are a promising setting for automatic harness evolution, as they are difficult enough to leave headroom, rely heavily on adaptation to out-of-distribution dynamics, and provide granular feedback by design. In these settings, task-specific harness evolution improves over the search baseline by 80% on ARC-AGI-3 and by 11% on EdgeBench games under matched budgets. Together, these findings highlight the need for matched-budget baselines and held-out evaluation to distinguish genuine harness improvements from benchmark-specific search and overfitting, and point to a more careful characterization of when automatic harness evolution is actually useful. Our code is available at https://github.com/rethinking-harness-evolution.

Yike Wang, Huaisheng Zhu, Zhengyu Hu +8

Allen Institute for AI · University of Washington · Independent

LLM Agent Evaluation Agent Harness Optimization LLM Agent Harnesses

Jul 13, 2026cs.AI

The Hidden Footprint: Making Storage a First-Class Metric for LLM Agent Evaluation

LLM agent benchmarks measure task completion, reliability, and inference cost, but not the persistent data an agent run leaves on disk, including logs, context snapshots, checkpoints, and debug traces. We introduce AgentFootprint, a cross-framework benchmark of post-run agent storage footprint. Its serialization-aware metric suite measures total retention, channel composition, duplication, growth, compressibility, and conversation-history reconstructability. It addresses a measurement trap: naive byte-level measurement understates duplication by an order of magnitude because database paging and JSON escaping obscure repeated content. A fixed-trace control separates agent-generated logical volume from persistence-layer amplification: replaying the same trajectory through seven persisting frameworks yields a 6.7x spread. Under identical models, tools, and tasks, configurations with 100% accuracy differ by 15.7x in retained bytes, although their defaults support different recovery and audit capabilities. Three full-history configurations grow superlinearly on a repeated-observation stress task. Exported trajectories from 108 instance-normalized SWE-bench Verified submissions span three orders of magnitude per instance, with no detectable correlation with resolve rate. A content-addressed store reduces retention by 4.8x-32.7x while preserving every reconstructability score. These results establish persistent storage as a resource metric to report jointly with accuracy and reconstructability.

Chenglin Yu, Hongquan Gui, Ying Yu +3

Department of Industrial and Systems Engineering, The Hong Kong Polytechnic University · College of Economics and Management, Zhejiang Normal University · Department of Computing, The Hong Kong Polytechnic University +2

LLM Agent Evaluation AI Agent Benchmarks

Jul 13, 2026cs.AI

NextFund: A Unified Performance Tracking Platform for Agentic Portfolio Management

Large language models (LLMs) based agents are beginning to participate in portfolio construction and market analysis, where decisions must be justified under evolving information and risk constraints. Current assessment practice, however, remains poorly aligned with this setting: many studies rely on static examinations or report only terminal portfolio returns, while the intermediate evidence, analyst judgments, and execution steps that produced those returns stay largely invisible. We introduce NextFund, an evaluation platform that makes financial-agent behavior observable under live market conditions. The platform couples time-consistent market access, coordinated multi-agent analysis, and persistent logging of the full decision path from observation to trade. Through an interactive Trading Arena, users can compare models across markets, inspect equity curves, and drill from leaderboard outcomes down to individual justifications. We present NextFund on Hong Kong, U.S., and China A-share equities, illustrating how inspectable decision histories enable fairer benchmarking and more actionable diagnosis. Our demo is available at https://paradoox.cn/nextfund/.

Changlun Li, Peixian Ma, Qiqi Duan +2

Paradoox AI Research

Quantitative Finance LLM Agent Evaluation AI Agent Benchmarks AI Agent Monitoring Finance Algorithmic Trading

Jul 13, 2026cs.SE

AgentCheck: A Reproduce-Intervene-Mitigate Workbench for LLM Agents over MCP

Tool-using LLM agents are mostly evaluated assuming all tools work. When a tool times out, returns a week-stale value, or has its description poisoned in deployment, the developer needs a controlled way to reproduce the failure, test a fix, and confirm the fix worked before deployment. We present AgentCheck, an open-source web workbench that turns an MCP server into an intervention surface. AgentCheck runs an agent against its real tools and records every tool response, then re-runs the agent with the response perturbed by a fault (12 types) injector. Matching tool calls are replayed from cache, and later tool calls go live after the agent diverges. This yields a reproduce-intervene-confirm loop: the developer toggles a mitigation, re-runs against the identical fault, and sees if the failure goes away. Scoring has two parts: deterministic pass/fail rules, plus an LLM judge for interpretive labels, validated against human annotations. Across five agents, the best passes 105/120 scenarios and the weakest only 77. The failures are usually silent, confident use of incorrect tool outputs rather than crashes. On the weakest agent, a retry mitigation raises success on timeout error faults from as few as 30% of cases to 100%, whereas stale-data faults remain near 3-4 of 10 regardless of the mitigation. AgentCheck makes these failure modes reproducible, comparable, and verifiable before deployment.

Aritra Mazumder, Nusrat jahan Lia

University of Utah · University of Dhaka

AI Agent Reliability LLM Agent Evaluation LLM Agent Verification Tool-Using Agents LLM Agent Reliability

Jul 12, 2026cs.SE

How Do Practitioners Build SE Agents? Insights from a Mixed-Methods Study

The rise of Software Engineering (SE) agents, i.e., LLM-based agents that can understand large codebases and carry out engineering tasks with limited human intervention, has been marked by rapid advances and adoption, but little is known about how developers build these systems in practice: existing studies mine repositories or examine deployment, but few investigate how SE agents are constructed. Through semi-structured interviews with 20 practitioners from 12 organizations and an online survey of 80 practitioners, this paper is the first to study how SE processes are changing in the development of SE agents and what challenges developers face. We find that as implementation becomes cheaper, bottlenecks shift rather than disappear: long-standing non-coding work such as requirements, coordination, review, and deployment becomes more visible, while reviewing and evaluating agent output becomes new and central. We characterize a seven-stage workflow and a shift toward evaluation-driven development, in which evaluation steers iteration and specifications become versioned artifacts read by both humans and agents. We further identify six challenges that teams face, together with the practices they adopt to address them, including unreliable evaluation signals, comprehension debt as code outpaces understanding, and behavioral changes introduced by provider-side model updates.

Yunbo Lyu, David Williams, Jieke Shi +5

Singapore Management University, Singapore · University College London, UK · Tencent, China +2

Software Engineering Agents LLM Agent Evaluation LLM Agents

Jul 12, 2026cs.MA

Auditing Belief-Conditioned LLM Agents in Hidden-Information Social Deduction Games

Evaluating LLM agents in hidden-information multi-agent settings is hard: final outcomes are high-variance and rarely reveal why an agent decided as it did. We study this in a 9-player Werewolf environment where agents act under strict, code-level information isolation, and we build an auditable framework that maintains an external belief state over hidden roles, logs belief updates and belief-action deviations as structured evidence, and supports a defensive offline improvement loop that reviews bad cases before any strategy change. Across 1,080 frozen games spanning belief-disabled, active-belief, kernel-ablation, camp-restricted, consumption-policy, and high-load arms, and including a seed-paired A0/A1 comparison, the active-belief condition is associated with substantially better good-side outcomes: in the 200-seed A0/A1 comparison the good-side win rate rises from 0.205 to 0.390 (paired McNemar

χ^2 = 16.4

,

p < 0.001

), with fewer irreversible witch-poison errors. We do not, however, attribute this shift to belief content. Direct action-belief consistency is low (

\approx 0.21

), and giving belief only to the werewolves helps the good side more than giving it only to the good side, which argues against a simple holder-benefit account; we therefore report the effect as an association and treat its mechanism as unresolved. The contribution is the audit framework itself: it makes the effect measurable, exposes low direct action-belief consistency, rejects an unreliable forced-consumption intervention with evidence, and separates strategy effects from load confounds. We accordingly position external belief in high-noise hidden-information games primarily as an auditable cognitive baseline that also carries decision-relevant signal, turning opaque agent behavior into replayable evidence for safer, controlled iteration.

Yuan Gao, Jiangyi Yang, Yao Zhao +1

University of Melbourne

Imperfect-Information Games Social Deduction Games LLM Auditing LLM Agent Evaluation LLM Agents

Jul 12, 2026cs.AI

Opti-Agent-Bench: Benchmarking End-to-End Optimization R&D Agents on Real-World Business Problems

LLM-based agents are increasingly deployed to solve optimization problems, yet existing benchmarks evaluate them on pre-structured mathematical formulations that bypass the most critical challenge: translating complex business requirements into correct models and solve efficiently. We introduce Opti-Agent-Bench, an end-to-end benchmark that evaluates Large Language Models (LLMs) across the complete optimization R&D pipeline, from understanding business-language descriptions through mathematical modeling, algorithm selection, and code implementation, to solution report generation. Our design rests on three pillars: (1) businesssemantic authenticity with anti-template traps that defeat pattern matching; (2) modular evaluation with cross-module consistency checking across Problem Understanding, Formal Modeling, Implementation, and Reporting; and (3) the ORAC bi-level validity framework that simultaneously ensures task quality and scoring integrity. Across several industrialscale tasks spanning integer programming, robust optimization, stochastic programming, and non-convex optimization, we expose critical failure modes of current models, including constraint omission, model-code inconsistency, and report-implementation divergence, that remain invisible under conventional single-metric evaluation.

Yongchang Fu, Xinjie Huang, Chengjun Dai +3

Ding Talk, Alibaba Group, Hangzhou, China · Zhejiang University, Hangzhou, China

Large Language Model-Guided Optimization LLM Agent Evaluation AI Agent Benchmarks Natural Language to Optimization Modeling

Jul 12, 2026cs.CL

MafiaScope: Non-Invasive, Time-Resolved Belief Probing for LLM Agents in Social Deduction Games

An LLM agent's public behaviour reveals little about its social reasoning: an agent that votes correctly may be guessing, and an agent that lies well leaves no trace of what it actually believes. We present MafiaScope, an open testbed that turns the social deduction game Mafia into a measurement instrument for machine Theory of Mind. After every public utterance, every agent privately answers a configurable set of structured probe questions; the answers never re-enter the game and are scored automatically against the ground truth the engine knows. An interactive visualizer renders the belief trajectories: impersonate mode shows the game as one agent sees it, panels chart timeline-aligned accuracy and calibration, and counterfactual replay forks any recorded step. In a 32-game DeepSeek case study with 13{,}815 parsed probe answers, stated confidence is poorly calibrated, with expected calibration error 0.17, agents over-predict being suspected 1.5 times, and a 30-fork replay experiment walks the counterfactual replay workflow end to end. Engine, viewer and a corpus of 200+ cross-model games are released under an open licence; live demo: https://karpovilia.github.io/mafiascope/; screencast: https://vimeo.com/1208920221.

Ilia Karpov

HSE University

Social Deduction Games Theory of Mind LLM Agent Evaluation AI Agent Benchmarks Language Model Probing

Jul 12, 2026cs.AI

Agents Don't Just Agree, They Remember: Benchmarking Persistent Sycophancy in Self-Improving Personal Agents

Self-improving personal agents now write profiles, memories, and reusable skills that carry over from one chat to the next. Prior work asks whether user pressure bends a model's next answer. Yet these agents can also write the user's claim down, so a later chat may read it back as trusted context. We call this persistent sycophancy. We introduce the Personal Agent Sycophancy Benchmark, PASB, with 1,600 tasks run on two real agents, Hermes-Agent and OpenClaw, across twelve models. Each task isolates a first chat containing the claim from a neutral follow-up chat, so any carryover must pass through a note the agent chose to write. Our analysis shows that downstream failure, meaning how often later answers side with the claim, reason from it, treat it as fact, or stretch it, reaches 71.9% when the follow-up chat can read a saved claim, against 45.0% when the claim stays in the first chat. Writing also edits the claim, as agents save it as a stable preference, a background fact, or a reusable procedure in 51.4% of runs. A saved claim still shapes answers in a different domain. Among the mitigations we test, explicit memory editing helps most, cutting downstream failure to 32.7% for Hermes-Agent and 54.5% for OpenClaw on same-domain follow-ups. PASB shows that self-improving agents must govern what they write down before it governs what they say. Our benchmark is available at https://github.com/henrymao2004/agent-sycophancy.

Xutao Mao, Liangjie Zhao, Leyao Wang +6

City University of Hong Kong · Adelaide University · Yale University +3

LLM Sycophancy LLM Agent Memory LLM Agent Evaluation Agent Memory Management

Jul 11, 2026cs.AI

Can Agentic Trading Systems Pay for Their Own Intelligence?

Large language model (LLM) agents are increasingly used in trading systems, where model reasoning, tool use, and continual decisions incur costs that are expected to produce trading value. Existing evaluations typically report performance metrics, but rarely examine agentic viability: whether dynamic LLM-mediated decisions convert their induced costs into measurable incremental profit. To apply this criterion, we introduce TradeLens, a trace-grounded diagnostic toolkit for evaluating agentic trading systems from their trading records, runtime traces, and deployment configurations. It reconstructs trading trajectories, attributes profit and cost to interpretable evidence, and diagnoses whether and why an agent pays for its own intelligence. We conduct extensive analysis across backbone models, capital scales, trading frequencies, and system architectures, together with deployment discussion. Our results show that viability hinges on intelligence-to-profit conversion: models exhibit different failure patterns, such as poor asset selection in DeepSeek-V3.2 and negative timing in GLM-4.7, while capital scale, trading frequency, and architecture matter primarily in these runs, especially through their effects on decision-attributed timing value. These findings reframe the evaluation of LLMbased trading agents from capability-centric performance ranking to trace-grounded diagnosis of intelligence-to-profit conversion. Our code is available at https://github.com/ParadooxAI/TradeLens.

Qiqi Duan, Changlun Li, Chen Wang +10

HKUST(GZ) · Paradoox AI Research · MBZUAI +1

LLM Agent Evaluation Agent Evaluation LLM Agents Algorithmic Trading

Jul 11, 2026cs.AI

Dynamic Agent Skills: A Lifecycle Survey and Taxonomy of Evolving Skill Libraries

Large language model agents increasingly store reusable procedures outside the model. These reusable procedures are often called \emph{skills}: they may be code functions, natural-language instructions, SKILL.md packages, workflow graphs, or learned adapters that a future agent can retrieve and invoke. This taxonomy-driven survey asks how such skill libraries change over time. Across a

124

-paper

2023

--

2026

audit set, we synthesize dynamic skill systems as \emph{lifecycle-managed, verified, evolving artifact stores}: agents collect evidence from interaction, propose skill updates, verify and admit candidates, organize them for retrieval and composition, repair or prune stale entries, and govern sharing through provenance and rollback. We organize the literature around three survey tools. First, a

\text{six}

-sense taxonomy distinguishes the structurally different artifacts called ``skills'' in current papers. Second, an

\text{eight}

-stage lifecycle architecture identifies the recurring design decisions behind evidence acquisition, proposal, verification/admission, storage, retrieval/composition, maintenance, distillation/portability, and governance. Third, a lightweight skill-record schema and

\text{ten}

-operator vocabulary provide common terms for comparing library updates without elevating them into a separate method contribution. Using this structure, we synthesize evidence-graded patterns with explicit caveats: admission and repair are repeatedly important, verifier quality materially affects skill-aware RL, flat retrieval can degrade as libraries grow, and current benchmarks still under-report library trajectories, usage--utility gaps, and safety surfaces. We close with concrete reporting standards and open problems for evaluating dynamic skills as changing libraries rather than static prompt or tool collections.

Yubo Li

Carnegie Mellon University

LLM Agent Evaluation LLM Agent Skill Learning LLM Agent Skill Retrieval

Jul 11, 2026cs.AI

AgentAbstain: Do LLM Agents Know When Not to Act?

Agent systems based on large language models (LLMs) are increasingly deployed for autonomous tasks, yet existing evaluations mostly focus on task success rather than whether agents know when to abstain. This gap poses real risks: under ambiguity, conflicting constraints, or tool failures, agents may execute unintended and irreversible actions. To close this gap, we present the first systematic evaluation framework for agentic abstention: the calibrated ability of tool-using LLM agents to recognize when not to act. At its core, AgentAbstain is a paired-task benchmark built on an agent-native taxonomy of 8 abstention scenarios across pre-execution reasoning and runtime discovery. It contains 263 paired tasks across 42 executable sandbox environments, where each pair consists of a should-act task and a should-abstain variant produced through a controlled perturbation to the instruction, tool, or environment state. To scale this paired design and resist data contamination, we propose AbstainGen, a fully automated pipeline that synthesizes sandbox environments and generates paired tasks end-to-end, validated by deterministic replay and semantic LLM judges; fresh task instances can be regenerated on demand, and three independent annotators rate 94-98% of sampled tasks as well-designed. Across 17 frontier LLMs in 4 agent harnesses, the best agent (Gemini 3.1 Pro) achieves only 59.5% paired accuracy (correct on both the act and abstain sides of each paired task). More importantly, abstention capability is largely independent of general task-solving capability, indicating that scaling task-solving alone will not close this gap. We further identify failure modes such as post-hoc abstention, in which agents execute irreversible actions before recognizing abstention triggers. Our code and dataset are open-sourced at agentabstain.github.io.

Xun Liu, Yi Evie Zhang, Vira Kasprova +5

University of Illinois Urbana-Champaign

AI Agent Reliability LLM Agent Evaluation Tool-Using Agents AI Agent Safety Benchmarks LLM Agent Reliability

Jul 10, 2026cs.AI

Who&When Pro: Can LLMs Really Attribute Failures in AI Agents?

Automated failure attribution uses LLMs to identify where and why agentic systems fail. As agents become more capable, their failures become subtler, making automated attribution increasingly important. We introduce Who&When Pro, a large-scale benchmark for automated failure attribution in agentic systems. Using a strictly controlled pipeline that injects a failure only after exactly replaying a successful prefix, we construct 12,326 failed trajectories with golden labels across 3 modalities and 26 benchmarks covering various scenarios. Beyond benchmarking, we conduct extensive experiments and analyses, revealing systematic patterns in how models attribute failures across modalities, protocols, and model families, and providing empirical guidance for future automated failure attribution systems.

Jiale Liu, Huajun Xi, Shaokun Zhang +6

Penn State University · AG2ai, Inc. · Mohamed bin Zayed University of Artificial Intelligence +2

Agent Failure Analysis LLM Agent Evaluation AI Agent Benchmarks

Jul 10, 2026cs.AI

DynamicMCPBench: A Trace-Grounded, Effect-Scored Benchmark for LLM Agents over Live MCP Servers

Large language model (LLM) agents are increasingly deployed over Model Context Protocol (MCP) servers, yet the benchmarks used to evaluate them score the final answer or a fixed "ground-truth" list of tools, both of which are fragile once the underlying data is live and stateful. We present DynamicMCPBench, a reusable framework rather than a fixed dataset. A practitioner can run it on their own MCP servers to test models on their own tasks, or let it collect servers automatically to measure a model's general ability to solve agentic tasks. Given the servers and any set of models, it generates realistic goals, pursues each one live to record a successful trajectory, distills that trajectory into path-agnostic effect checkpoints, and scores an agent on whether it reproduces those effects, never on the final answer. To show what the framework reveals, we run it at scale: 24 models over 121 servers and 750 tasks spread evenly over 15 task categories (50 each), where each category targets a distinct tool-use challenge of the generated questions. Each task is scored by pass^3: it counts as solved only if all three independent attempts succeed. Even the strongest agents solve only about half of the tasks, 31% of tasks are solved by no model at all, and accuracy collapses as the required tool chain grows longer (from 39% on the shortest chains to 13% on the longest). A human validation study confirms the automatic scoring is reliable (chance-corrected agreement of 0.76). DynamicMCPBench thus turns benchmark construction into something practitioners can rerun on their own servers and models, while exposing a consistent inability of current agents to handle long, multi-step agentic tasks.

Jerzy Kamiński, Ilya Galyukshev, Artem Kuznetsov +4

ITMO University · Central University

Computer-Use Agent Benchmarks LLM Agent Evaluation Long-Horizon Agent Evaluation AI Agent Benchmarks Model Context Protocol Tool-Using Language Agents

Jul 8, 2026cs.AI

Do LLM-Generated Skills Make Better AI Data Scientists? A Component Ablation Across Data-Science Workflows

Product data scientists often ask LLM-based agents to help with recurring execution tasks such as cleaning data, writing SQL, choosing statistical tests, and formatting results. Reusable skill files are meant to avoid prompting from scratch by packaging guidance for a task family. Expert-written skills can encode high-quality guidance, but writing and maintaining them across many data-science task families creates a manual bottleneck. We ask whether LLM-generated skills offer a useful low-curation alternative: do they improve performance over the task prompt alone? We test this question across four lifecycle stages: data preparation, data extraction, statistical analysis, and reporting, using one generated skill per stage. We find no reliable improvement from full generated skills over No-Skill prompting. We then ask whether any part of the skill is useful by ablating different skill components. The main ablation covers 56 tasks, nine model configurations, and three providers, yielding 7,560 runs. Compared with prompting using the task alone, neither the full generated skill nor any ablated skill variant significantly improves performance; all p-values are at least 0.396, and the total spread across variants is only 1.2 pp. A supplemental token-matched control adds 1,512 runs and finds that Full skills perform similarly to task-irrelevant skill-formatted content. The results caution against using one LLM-generated skill per data-science workflow as a default single-shot prompting strategy.

Wei-Jung Huang

Independent Researcher United States

LLM Prompting LLM Agent Evaluation LLM Agent Skill Retrieval

Jul 7, 2026cs.AI

Evaluating SageMath-Augmented LLM Agents for Computational and Experimental Mathematics

Recent advances in AI for Mathematics have focused largely on autoformalization and theorem proving, leaving the role of Computer Algebra Systems (CAS) in agentic LLM workflows underexplored. We propose a ReAct-style agentic setup that combines LLM reasoning with verifiable feedback from SageMath, together with Context7 for the up-to-date documentation. We evaluate this agentic setup across frontier models for solving research-level mathematical problems from the RealMath benchmark in a setting that emulates a computational-mathematics research loop. We also propose a refinement to the RealMath benchmark by introducing a multi-step post-processing procedure and a multi-stage validation pipeline, both of which improve the quality and reliability of the extracted problem set. Our experiments reveal substantial performance gains from SageMath access across all evaluated models on +9.7~~pp on average, the gains range from 1.5~~pp to 27.8~~pp and narrow the gap between open-weight and closed models. Qwen~~3.7-Max benefits from SageMath the most, while GPT-5.5 achieves the highest solve rate of

75.2\%

and the lowest token usage among tool-enabled configurations. Our findings suggest that CAS-augmented agents represent a promising direction for assisting mathematicians in computational exploration, and we believe that this work is a step towards automated conjecture discovery. The project repository is available online.

Pavel Snopov, German Magai

School of Mathematical and Statistical Sciences, The University of Texas Rio Grande Valley, USA · Noeon Research, Tokyo, Japan.

Mathematical Reasoning Benchmarks Knowledge Augmentation for Language Models LLM Agent Evaluation LLM Mathematical Reasoning

Jul 7, 2026cs.SE

Reliable and Developer-Aligned Evaluation of Agents for Software Engineering

Large language models are rapidly moving towards closing the development cycle, transitioning from simple assistive companions to autonomous contributors deeply embedded into collaborative development environments. Despite their accelerated adoption, existing evaluation techniques are limited due to their fragmented nature and distorted projection of true model capabilities, often obtained from hypothetical syntactic scenarios. This research aims to bridge this gap by providing a comprehensive evaluation methodology for LLM-powered agents that is grounded in real-world software development practice. Our evaluation approach focuses on contamination-awareness, in-the-wild agentic behavior assessment, and trajectory-aware benchmarks and metrics capturing realistic coding contexts, human-aligned behavior, and model failure modes.

Razvan Mihai Popescu

Delft University of Technology Delft, The Netherlands

Software Engineering Agents LLM Agent Evaluation Agent Evaluation AI Agent Benchmarks LLM Agent Reliability

Jul 7, 2026cs.SE

RuBench: A Repository-Level Agentic Coding Benchmark with Natively Authored Russian Task Specifications

Developers increasingly delegate real maintenance work to product-grade coding agents, and many state tasks in their native language, in the style of a customer request rather than a curated English issue. Existing repository-level agentic benchmarks do not measure this setting: their task statements are English by design. We introduce RuBench 1.0, a benchmark of 25 tasks mined from recent fix commits in five live open-source repositories (aiohttp, aiogram, Laravel, NestJS, Fastify; Python, PHP, TypeScript, JavaScript), where each task is specified natively in Russian -- written from scratch in the style of an actual customer request, not translated -- and judged by the upstream maintainer's regression tests, which we withhold from release. All 25 fix commits postdate the training-data cutoffs of every evaluated model, giving a contamination argument that holds task-by-task. We evaluate deployed product configurations (CLI agent + model + reasoning effort) -- Claude Code with Opus 4.8, Sonnet 5, and Haiku 4.5, and Codex CLI with GPT-5.5 -- with three independent runs each, reporting pass@1 with task-level confidence intervals, paired comparisons, dollar cost, and token usage. The best configuration resolves 78.7% of tasks; at N=25 only the gaps to the weakest model are statistically resolvable, which we state explicitly. Auditing full trajectories of a fifth, hors-concours configuration (Claude Code + Fable 5, July 2, 2026 release), we caught the product silently substituting the model: on 5 of 25 tasks (20%) an official safeguard fallback re-routed routine HTTP-protocol fixes to Opus 4.8 -- direct, reproducible evidence that the deployed product, not the model, is the unit actually measured. We release task statements, metadata, full agent trajectories, and diffs; grading oracles are withheld, with a SHA-256 manifest committed at publication time.

Evgeny Shilov

Independent Researcher

Computer-Use Agent Benchmarks Tool-Augmented Language Model Agents AI Coding Agents Software Engineering Benchmarks LLM Agent Evaluation