LLM Agent Evaluation

As large language model (LLM) agents evolve into personalized companions, memory has emerged as a core capability. However, LLMs face a knowledge utilization problem: they may fail to act on relevant user preferences even when they are fully present in context. When an agent fails to tailor its response in a context where previously shared user preferences should matter, it is unclear whether the model failed to remember that information or remembered it but failed to use it. To isolate this breakdown, we introduce a decoupled evaluation paradigm that administers paired Know and Act tests to the same user preference. We conduct large-scale experiments across 16 systems and five memory architectures, evaluating 1,000 preferences embedded at three levels of expression strength. Our results show a large gap between Know and Act outcomes: agents often pass the recall test for a user preference but fail to reflect that same preference in the paired behavioral scenario. While memory architectures reduce this gap, utilization remains especially weak for health and therapy-related preferences, where failures to act carry the greatest real-world stakes.

With the prosperity of the large language models (LLMs), it has become an interesting topic: how do LLM-based agents work in Minecraft? Unfortunately, most existing benchmarks evaluate them under fixed game mechanics. High performance in these settings does not show whether an agent can continue making progress when familiar recipes, drops, and other rules change. In this paper, we introduce MirrorCraft, a paired benchmark for evaluating agents under hidden rule changes in Minecraft. Each Mirror world is a copy of its paired Vanilla world, with selected server-side rules modified by the corresponding datapack. Terrain, spawn, resource placement, objective, interface, and action budget remain matched within every Vanilla-Mirror pair. MirrorCraft includes five controlled biomes, six rule suites, three progression objectives, two model families, and six agent configurations under a shared Mineflayer interface. We evaluate task progress with deterministic advancement milestones and success rate and use the Rule Intervention Effect (RIE) to measure the performance change between matched Vanilla and Mirror worlds. The experiments show that hidden rule changes have strongly different effects across suites. Among the configurations evaluated without rule descriptions, ReAct achieves the highest pooled Mirror score. Providing the exact rules yields modest gains in average progress and completion across all three objectives. MirrorCraft extends Minecraft evaluation beyond fixed mechanics and provides a controlled setting for studying how agents use gameplay outcomes when the rules of the current world differ from familiar ones.

Evaluating LLM coding agents in algorithmic trading is difficult because static benchmarks risk data contamination and numerical backtest outputs require ground truth from actual code execution. We present Backtrader-Bench, a framework with two complementary pipelines. A deterministic multiple-choice question (MCQ) pipeline generates questions from backtest configurations across five trading strategies, 33 templates, and three difficulty tiers, with an independent checker that re-derives every answer. A generator-solver filtering pipeline autonomously mines harder questions: a generator writes questions verified by executable code, converts them to MCQs, and discards any that a no-tool solver can answer without code execution. We evaluate 11 models without tools (10 runs each) and four with-tools configurations on a 30-question curated set. Tool-augmented agents reach 90.0% accuracy in a single pass (GPT-5.5 and Opus 4.7), outperforming the best no-tools baselines (73.0%, averaged over 10 runs) by 17 percentage points. On 38 separately mined questions, no-tools accuracy drops further, with half the models falling to roughly random-chance level (25%). Beyond evaluation, the scalable MCQ infrastructure is designed to produce a training corpus for reinforcement learning, with the ultimate goal of building a specialized agent for quantitative trading workflows.

Large language model agents are increasingly evaluated as autonomous tool users, yet most benchmarks focus on bounded tasks with immediate success criteria. Real-world deployments often require Long-Term Coherence, the capacity to preserve purposeful behavior across extended horizons while adapting decisions to accumulated evidence. Evaluating this capacity requires a persistent environment in which actions constrain future choices, feedback arrives at heterogeneous delays, and incoherent behavior produces measurable cumulative effects. Seller-side e-commerce provides a suitable setting for this evaluation through recurrent and interdependent decisions over Product Sourcing, Listing and Pricing Control, Cash-Flow Management, and Mixed-Latency Feedback Adaptation. We introduce MerchantBench, a 365-day order-level simulation grounded in 98,843 real e-commerce product records and equipped with 26 tools for agent interaction. MerchantBench couples promptly observable Upstream Supplier Events with delayed Downstream Order Outcomes, requiring agents to follow individual order lifecycles and revisit earlier decisions. We evaluate eight LLMs under two agent frameworks in 48 runs, each spanning 365 simulated days. Our results reveal a substantial gap between even the latest LLMs and human participants, with the best LLM configuration attaining only 27.3% of the mean final net assets achieved by human participants.

Agentic commerce is moving from concept to deployed infrastructure: payment networks, retailers, and AI platforms are setting the stage for agents to transact on behalf of merchants and consumers. Yet whether the LLMs behind these agents can price competently in real markets, where customer preferences are hidden, competitors adapt in real time, and demand can shift without warning, has not been systematically tested. We introduce Bazaar, a dynamic sealed-bid benchmark for multi-attribute auction under these conditions. Despite its dynamics, the benchmark is grounded in closed-form customer utilities, enabling exact evaluation. Across 11 frontier LLMs from four providers, the leading agents on customer acquisition (e.g. Gemini 3.1 Pro) are often not the leading agents on profit (e.g. Opus 4.6). The ranking shifts again under demand shocks: agents that learned fastest pre-shock are typically the slowest to revise their beliefs afterwards, while Gemini 3.1 Pro recovers fastest despite not leading on profit. However, even the strongest agent captures less than a third of hindsight-optimal profit, suggesting current LLMs are progressing in agentic commerce but leave substantial headroom.

When a repair agent runs a test and sees it pass, the result is treated as evidence about the reported defect. We measure how often that treatment is warranted. BSG-VA (buggy-state/candidate-state/gold-fix validation analysis) captures each validation command at its exact working-tree state, extracts a test-only patch, and replays the command on the original buggy code (B), the candidate state (S), and the developer gold fix (G). The captured outcome and the replay results assign every event an evidence role, from gold-aligned bug-discriminating through regression-only to misleading. Across 3,730 events in 643 rollouts on 110 tasks, 46.0% of positive comparable events carry no bug-discriminating information; 23.8% of baseline rollouts, with no feedback injected, close with a patch whose entire positive evidence base is of this kind. A three-arm experiment tests whether returning the B-replay outcome to the agent changes this pattern. Bug-contrast feedback reduces evidence-inadequate closure by 7.8 percentage points relative to an attention-matched reminder (p = 0.0029) and raises bug-discriminating evidence by 7.4 points (p = 0.011), with no detectable cost to repair success. Both estimates fall below the prespecified 10-percentage-point smallest effect size of interest, so practical magnitude remains uncertain. Roughly a third of the improvement traces to the reminder alone; across two exploratory replications, varying the scaffold and the model, the B-replay content adds a detectable increment only with gpt-5.6-sol under the unconstrained tool-use loop. BSG-VA applies post hoc to any replayable repair trajectory that preserves the required code states and execution environment. Keywords: program repair agents, validation evidence, test adequacy, large language models, software quality, controlled experiment.

Existing evaluations often reduce agent failures to system-level outcomes, obscuring where the fault originated and which intervention would improve the agent system. This creates a repair-assignment problem: the same visible failure may call for model post-training, harness engineering, environment redesign, or benchmark repair depending on its source. Because agent behavior emerges from interactions among models, harnesses, users, tools, memory, and environments, outcome-level labels are often insufficient for improvement. Most failure taxonomies do little to resolve this problem because they are benchmark-specific and lack a shared structure. We introduce an interaction-centric taxonomy that localizes failures to the interactions in which they originate and identifies the responsible component. It organizes 41 failure modes by assigning each to an edge between two components and a fault side indicating where the repair belongs. This makes the taxonomy actionable: model-side failures identify targets for post-training, harness-side failures point to scaffolding and tool-integration fixes, and environment or grader failures reveal evaluation conditions requiring redesign. The schema applies across agent architectures, from coding assistants to long-horizon personal assistants and multi-agent systems. We ground the taxonomy in worked examples from public benchmarks, model system cards, published reports, and logged agent trajectories, and evaluate its reproducibility using independent reasoning agents as judges. Across four frontier models, the strongest judge reaches Cohen's $κ=0.76$ against human category labels, suggesting that the categories capture shared structure rather than annotator-specific preferences.

As large language models (LLMs) are deployed as agents in high-stakes settings, such as medical and legal systems, understanding their deceptive capabilities is fundamental to safety. Controlled social deduction games provide a reproducible proxy for isolating and evaluating these complex adversarial behaviors. We present the open-source benchmark framework ParliamentBench based on the game Secret Hitler to evaluate LLMs in scenarios that require deception, persuasion, and reasoning under information asymmetry. We evaluate 16 LLMs across 1,600 simulated matches playing each other, playing against humans, and compare them against a large set of online games. We introduce three novel metrics that isolate social deduction, reasoning, and deceptive consistency. Our experiments reveal that frontier models achieve strong performance across cooperative and deceptive roles, with a strong top-four cluster (GPT-5.4, Kimi K2.5, Grok 4.1 Fast, and DeepSeek 3.1 Terminus), whereas the weakest models fall short of random (33%) and simple algorithmic (45%) baselines. Most LLMs struggle to maintain a consistent deceptive persona throughout an entire game, with deception retention dropping below 50%.

As LLM-based agents are deployed in complex, multi-step workflows, a critical evaluation gap has emerged: most existing benchmarks judge only final outcomes, unable to distinguish reliable reasoning from lucky success or attribute failures to specific process deficiencies, hindering attribution in long-horizon tasks. In this work, we present ClawTrack, a dual-assessment benchmark that simultaneously measures what an agent achieves (Task Score) and how it achieves it (Process Score). ClawTrack comprises 320 tasks across 8 domains with 25+ deterministic mock services. A Process Grader scores each reasoning turn along four dimensions (goal alignment, efficiency, information utilization, and result verification), anchored by 12,541 task-specific rubric items. Evaluating 21 models over 16,000+ trials, we find that: (1) process scores effectively attribute success and failure to specific reasoning dimensions, filtering lucky passes invisible to outcome-only evaluation; (2) the four dimensions are complementary, with result verification as the systematic bottleneck; (3) the framework is robust to evaluator choice across different judge LLMs; and (4) process-based trajectory filtering yields consistent post-training improvements across model scales.

Large language models (LLMs) and LLM-based agents are increasingly being deployed to automate complex workflows, promising to revolutionize data management and processing. However, existing benchmarks predominantly focus on simplified Text-to-SQL translation or data analysis, leaving the critical and complex domain of end-to-end data engineering largely unexplored. To bridge this gap, we introduce DataClawEval, the first comprehensive benchmark designed specifically to evaluate the end-to-end task completion capabilities of autonomous agents in real-world data engineering scenarios. Built upon production-grade code authored by professional enterprise data engineers, it comprises 100 rigorous, end-to-end tasks spanning five execution engines: PySpark, MySQL, HiveSQL, PrestoSQL/Trino, and FlinkSQL. Rather than non-deterministic LLM-as-a-judge scoring, each task is executed within a case-specific, isolated sandbox and graded by deterministic, rule-based scripts. Evaluating 16 frontier agents exposes critical limitations: The strongest model attains only 74.9 overall, and no single model dominates, as each excels on a different engine, revealing strict domain specialization rather than omnipotent proficiency. Thus, autonomous data engineering remains a formidable, unresolved challenge. We release our dataset, containerized environments, and deterministic evaluation scripts at https://github.com/Dicemy/DataClawEval/tree/master

Role-playing agents (RPAs) have become one of the most important consumer applications of large language models. Users engage in multi-turn conversations with RPAs for experiences such as emotional comfort, making reliable evaluation essential for measuring capability, comparing systems, and guiding further improvement. Existing benchmarks, however, typically require an RPA to continue a fixed dialogue history and then evaluate the continuation using a fixed rubric detached from the user. We identify and empirically demonstrate two limitations of this design. First, an RPA's output is shaped by the preceding dialogue history, preventing a scientifically grounded assessment of its role-playing ability in real multi-turn settings. Second, user experience varies substantially across individuals, and conventional fixed rubrics need not align with user satisfaction. We therefore introduce PALATE (Person-Aligned LLM-Simulated-User Assessment with Tailored Evaluation), a scalable RPA benchmark built on user simulators. PALATE is accompanied by a pool of 300 character profiles. Its main evaluation trains five per-user simulators and lets them engage candidate RPAs in free-form, multi-turn conversations over a pre-frozen panel of character profiles. Alongside a general quality rubric, we construct personalized rubrics to measure user satisfaction; on held-out annotated data, the personalized rubrics show higher agreement with human judgments than the general rubric. In the main evaluation of 16 candidates, PALATE separately characterizes generic turn quality, long-horizon session capability, and per-user experience on multi-turn trajectories co-constructed by each candidate. It thereby produces interpretable evaluations of specific user-RPA pairs rather than compressing systems into a single user-independent ranking.

Large language model agents increasingly plan, execute, and interpret biological analyses, yet fluent responses, successful tool calls, and benchmark performance alone do not establish scientific credibility. Existing reviews primarily organize biological agents by application, architecture, and agentic capability, but do not jointly operationalize the accountability of agent-generated workflows. We address this gap by treating the inspectable workflow trajectory, rather than architecture or final output alone, as the primary unit of analysis. We introduce the Function--Evidence--Validation (FEV) framework, which separates demonstrated workflow operations, traceable support for actions and claims, and use-case-specific validation. Using FEV, we map 109 agentic or agent-adjacent systems and 28 benchmark or evaluation resources, representing 128 unique publications across genomics, single-cell and spatial omics, protein science, drug discovery, computational pathology, and general bioinformatics automation. Across domains, planning and tool-mediated execution have advanced more rapidly than replayability, provenance, robust scientific assessment, external validation, and prospective empirical testing. We therefore argue that agentic bioinformatics should be assessed through workflow correctness rather than final-answer correctness alone. FEV provides a practical basis for comparing systems and designing transparent, auditable, and scientifically accountable bioinformatics workflows.

Reusable skills are becoming a standard interface for extending language agents with task procedures. Yet evaluators usually infer skill use from visible reasoning or the agent's own attribution. These signals show what the agent appears to use, not whether the skill changed its decision. We ask whether skill-augmented agents exhibit a \textbf{Reasoning Backroom}, a systematic gap between stated skill use and intervention-measured influence. We introduce BACKTRACE, an evaluation framework that pairs each skill-conditioned answer with a matched no-skill counterfactual, intervenes on skill meaning, wording, identity, content, and assignment, and elicits attribution only after the answer is committed. We instantiate the framework as BACKROOMBench, a verified testbed spanning controlled logic and competition mathematics, multiple skill conditions, single-agent and multi-agent settings, and diverse model families. Our evaluation reveals a pervasive provenance failure. Across models and domains, stated skill use often remains stable while causal reliance and signed utility vary, producing both silent uptake and performative use. Behavioral effects follow procedural content more reliably than displayed skill identity, whereas stated attributions respond strongly to artifact availability. Observational detectors based on direct skill-use claims, text mentions, trace similarity, and an LLM judge do not identify which decisions actually depend on the skill. In multi-agent systems, skill influence can survive communication even after its source is lost, while no-skill teams still name skills and sources that were never supplied. These findings establish the Reasoning Backroom as a general AI provenance problem whose audit requires intervention.

Production voice agents span cascaded, speech-to-speech, and hybrid architectures. Voice-agent benchmarks typically measure component quality and conversational properties such as word error rate, latency, naturalness, and turn-taking. Fewer measure whether the agent handled a phone call correctly on its own. Contact centers refer to this as ``containment'': the share of phone calls the automated system resolves without handing off to a human. On some phone calls the right outcome is refusal or a redirect. To address this gap, we introduce VAmoS Bench, the Voice Agent Simulation Bench. It measures complete voice-agent systems end to end in a stateful customer-support task. The agent is Riley, a credit-card support representative for a fictional bank who can freeze, cancel, replace, or activate a card. Each of 100 scenarios supplies a simulated caller with a private goal and a seeded PostgreSQL backend. The platform uses each scenario to populate and activate an isolated simulation in which the caller reaches Riley over audio; roughly one-third apply adversarial pressure. The agent can use five tools that execute real SQL against the backend. Each scenario also defines binary assertions. A grader evaluates them against the complete trace of what the caller and agent said and what the agent did, including tool invocations, arguments, and returned rows. This catches an agent that claims to have changed a card without updating the database, as well as one that makes the right database change while disclosing protected information. This first benchmark version focuses on financial services. Its evaluation protocol supports an evolving leaderboard: additional voice agents can be evaluated on the same version, while later versions can expand the tasks and scenarios.

Large language model (LLM) agents are increasingly expected to assist users in completing tasks. However, existing benchmarks provide limited support for evaluating whether agents can carry out office-suite workflows at a reasonable cost. We introduce OmegaUse-OfficeVal, a benchmark for evaluating LLM agents on long-horizon office-suite tasks with task-level economic grounding. The benchmark comprises 100 tasks derived from office-suite requests proposed by practitioners and adapted through a privacy-preserving process. On average, these tasks require 2.32 hours of human labor to complete. An important feature of the benchmark is that each task is paired with two economic signals: human labor time and task price proxy. These signals enable direct comparisons between human costs and LLM inference costs, as well as value-weighted evaluation. To support stable evaluation, we develop code-based verifiers from fine-grained rubrics. We evaluate several frontier LLMs together with a human baseline. Although all evaluated LLMs are substantially cheaper and faster than human workers, they have not yet approached human-level deliverable quality. The code and dataset are fully open-sourced, and more information is available on our project website: https://omegause-officeval.github.io.

Travel planning is a demanding stress test for tool-using LLM agents: a usable itinerary is a single artifact that must be right along many axes at once - every flight, hotel, and attraction must exist and be bookable, the days must be physically traversable, the total must clear a budget, and the plan must serve a traveler whose needs are only partly stated. Existing agent benchmarks reward these properties one at a time and grade the final output with soft or LLM-judged rubrics, which cannot certify that a returned plan is executable and are neither reproducible nor auditable. We introduce TREK (Travel Reasoning and Evaluation Kit), a benchmark for feasible itinerary synthesis: producing a single plan that is jointly constraint-correct, hallucination-free, spatio-temporally executable, budget-valid, and responsive to the traveler's unstated persona needs. TREK comprises 800 multi-constraint tasks - 533 feasible and 267 provably infeasible with typed route/entity/budget causes - over a synthetic, internally consistent knowledge base of 212,530 records across 375 cities and 13 personas, served through a production-style tool sandbox of validated RESTful APIs. Every task is scored by a fully deterministic, rule-based evaluator with no LLM judge and ships a human-verified gold reference that scores a perfect 1.0 under that same evaluator, so the ceiling is demonstrably achievable and every remaining gap is an agent limitation rather than scorer strictness. Evaluating 15 LLM agents across nine constraint dimensions, we find that even the strongest (GPT-5.6) produces a fully-feasible plan on only 46.2% of solvable tasks, with a median of 6.6% and a floor of 0.0%; satisfying travelers' unstated needs emerges as the universal bottleneck, unsolved even at the frontier. We release the dataset, tool sandbox, deterministic evaluator, and agent code as a fully reproducible benchmark.

Multi-agent LLM pipeline systems break down the task among multiple roles for better reasoning, but are benchmarked mainly with large-scale commercial models. In this study, we investigate Parishad, a structured multi-agent system involving five roles, by deploying it on Qwen2.5-7B-Instruct, a local model, on two datasets: GSM8K (500 questions) and HumanEval (164 questions), compared with prompting directly and two-call self-refinement. The multi-agent system drops GSM8K accuracy from 75.0% to 45.0% with JSON data format due to the error accumulation problem. With plaintext format, the accuracy is restored to 82.0%. A two-call self-refinement strategy (V1) can achieve 86.2% accuracy on GSM8K, with 7.4$\times$ lower token usage. However, the same V1 implementation on HumanEval---where direct accuracy is already 96.3%---actively destroys performance (66.5%). A task-aware gated redesign (V2) applied to HumanEval preserves accuracy at 95.1%. Our results demonstrate that communication format and implementation details determine outcomes more than architectural complexity, and that simpler approaches match or outperform multi-agent pipelines for local 7B model deployment. All code and data are released.

Large Language Model (LLM) agents are increasingly adopted in real-world security operations with access to host artifacts and command-line interfaces (CLIs), making it critical to thoroughly assess their security capabilities. However, existing cybersecurity benchmarks focus on pre-compromise settings where agents are placed in a clean and idealized environment before an attack occurs. This leaves the post-compromise setting underexplored. To address this gap, we introduce SecRespond, the first benchmark for evaluating LLM agents on the post-compromise incident-response workflow. Given a forensic disk snapshot of a compromised host together with the alerts, vulnerability scans, and baseline checks reported by a host security product, agents are required to produce forensic reports on intrusions, baseline risks, and vulnerability risks, together with a remediation plan. We instantiate this task across 10 cyber ranges, each constructed from a distinct compromised cloud host, spanning 4 entry-point types, 21 ATT&CK techniques, and 5 operating systems. We evaluate 23 frontier LLMs on the OpenCode agent harness. Experimental results show that although current agents can reliably uncover the problems exposed by alerts, they struggle to proactively investigate the disk for silent intrusions and to produce comprehensive, verified remediation plans, with no model achieving complete detection and remediation on any single range. This reveals a fundamental bottleneck in building agents for real-world incident response. The benchmark is publicly available at https://github.com/Alibaba-NLP/qqr/tree/main/data/secrespond.

As LLM agents move from decision support to autonomous procurement, firms need to know whether delegated negotiators create value, divide it predictably, and avoid money-losing contracts. We study this in a canonical supply chain bargaining problem: a buyer with private demand information negotiates a quantity-payment contract with an uninformed seller. We benchmark nine LLMs from OpenAI, Google, and Alibaba against a validated Perfect Bayesian Equilibrium across 9,840 LLM-to-LLM negotiations. First, capability governs value creation. Agents agree in 98.9% of negotiations and capture 95.4% of first-best surplus undiscounted, but average 2.98 rounds against the benchmark's 1.25, and this delay erodes 21-34% of surplus. Capability also governs reliability: baseline models accept individually irrational contracts in 19.2% of cases, versus 0.0-0.6% at mid-tier and flagship, making automated profit verification the binding guardrail below that threshold. Second, surplus capture is relational. Provider identity predicts who captures surplus better than capability rank: self-play buyer shares average 40% for OpenAI, 50% for Google, and 70% for Alibaba's Qwen, an ordering that survives restricted communication and no discounting. Reversing which provider sells moves the division by 7-18 percentage points, and the capable Qwen flagship is the weakest cross-family seller: vendor choice is a first-order distributional decision. Third, the prompt is a strategic lever. Delegation separates the principal's economic patience from the agent's prompted strategic patience, a free deployment choice that is the single strongest driver of surplus division (90% of explained variance). Together these establish an equilibrium-referenced audit of AI agents along three dimensions: discounted efficiency, distributional profile, and operational reliability.

An important skill in theoretical physics is to recognize when a new problem can be transformed into a known model. We study this skill as an AI-agent task: can LLM-based agents discover statistical mechanical mappings from a raw partition function to a tractable representation? To probe this question, we introduce StatMechBench-v0, a benchmark of six Ising-type problems covering transfer-matrix methods, gauge-removable disorder, and planar/Pfaffian structure. We evaluate a simple propose-verify-revise agent across multiple LLMs and problem phrasings. The results show that numerical feedback often helps agents repair code and recover correct partition functions. However, agents can also pass the numerical checks while misidentifying the underlying tractable class or understating computational complexity. This both reveals limitations in current LLM reasoning and calls for a verification stack that goes beyond numerical agreement, incorporating, for example, symbolic checks and structural invariants. Our study provides an early evaluation and design directions for AI agents aimed at structural discovery in theoretical physics.

Recursive self-improvement requires turning evidence of model failures into better models. Data-centric post-training research entails diagnosing capability gaps, designing and validating training-data strategies, and learning from checkpoint feedback. Can LLM agents automate this loop? Existing benchmarks entangle research decisions with optimization, serving, evaluation, and systems implementation, obscuring agents' research capability. We introduce RSIBench-Data, a controlled benchmark of LLM agents as data-centric researchers with a fixed post-training stack. Agents iteratively revise training-data strategies for a fixed target model; training and serving use Tinker-backed services, official evaluation runs through Harbor and E2B sandboxes, and budgets are fixed across agents. We evaluate four frontier agents on six benchmarks across software engineering, terminal use, scientific question answering, and mathematics. Agents demonstrate core data-centric research capabilities: in 58.33% of settings, they improve upon the first valid attempt by refining strategies from feedback. However, improvement is inconsistent. Among searches continuing after the best observed score, 78.26% end with a lower-scoring final attempt, while the rest only recover the same peak. A strong candidate may therefore appear early or midway through a run even as later revisions fail. Trajectory analysis identifies four patterns in stronger runs: accurate hypotheses, validation-grounded supervision, behavior-aligned data, and preservation of strong checkpoints. These findings suggest that current agents can make useful data-centric discoveries but cannot yet translate feedback into consistent improvements. RSIBench-Data provides a measurable, auditable testbed for the research capabilities required for recursive self-improvement. We open-source our code at https://github.com/evolvent-ai/RSIBench-Data.

Persistent context files (AGENTS.md, CLAUDE.md) are standard practice for guiding AI coding agents, yet evidence for their effectiveness is contradictory. We present a controlled ablation of context-injection strategy across two frontier agents (Claude Code and Codex), 17 real tasks from 3 repositories (15 shared + 2 Codex-only), and 288 evaluated runs with gold-test evaluation. Context strategy does not measurably move correctness on either agent (bounded to <=10-15pp via equivalence testing). A failure-mode triage reveals why: agents fail on implementation skill---feature design, pattern selection, exact wiring---not missing repository knowledge that a context file could supply; a manipulation probe confirms the real AGENTS.md never converts a near-miss to a pass on either agent. We further show that borderline task difficulty is agent-specific (Spearman rho=0.75), offering a candidate explanation for prior contradictions: single-agent studies draw tasks from different agents' informative bands. We release all code, data, and analysis.

Debt collection is a critical negotiation task in the financial industry, with strong practical relevance and exceptional academic value as a behaviorally rich, high-stakes testbed for human-centered dialogue systems. While large language models (LLMs) have shown promise in dialogue and negotiation, effectively evaluating their performance in this complex scenarios remains a major challenge: existing benchmarks uniformly assume users to be static, rational agents with fixed preferences, failing to capture the rich behavioral heterogeneity inherent in real-world debt collection. To bridge this gap, we propose DebtBench, the first public persona-enriched debt collection benchmark, that highlights behavioral heterogeneity in negotiation. Moreover, we develop DebtGPT, a debt collection agent trained to jointly optimize financial recovery and interaction experience. Our experimental results, using 16 state-of-the-art LLMs, find that most existing models struggle in this complex but realistic scenarios, whereas DebtGPT outperforms all open-source baselines and achieves performance on par with GPT-4o. The code and data are available at https://github.com/YYuHhhh/DebtNegotiation.

Electroencephalography (EEG) analysis in cognitive studies requires specialized expertise and involves many defensible choices over contrasts, channels, time windows, and statistical tests. LLM agents can translate varied natural-language questions into analysis choices, offering a flexible interface for automation. Yet fluent reports alone cannot establish that an agent selected the requested analysis or evaluated a confirmatory claim independently of adaptive search. We present CogEEGAgent, a cognitive-EEG analysis agent grounded in MNE-Python. Its EEG-specific scientific harness separates semantic from scientific authority. The LLM interprets intent and proposes registered analyses, while deterministic components validate typed contracts, control confirmation access, and authorize evidence-bound release. On a prespecified routing benchmark, CogEEGAgent maps language to registered analyses more accurately than a matched deterministic router, while matched preflight makes both systems abstain whenever required. In an externally model-authored, outcome-blind campaign, the complete system releases supported analyses with participant-disjoint confirmation and blocks prespecified capability hazards and lifecycle-reuse requests. Policy stress testing shows that held-out confirmation curbs false positives from uncorrected adaptive search. Together, these studies establish bounded autonomy and an auditable automation framework for cognitive-EEG workflows. More broadly, they show how scientific agents can combine flexible language understanding with fail-closed control over inference and release.

Deep search is becoming a core capability of modern agent systems, yet it is typically evaluated solely based on end-to-end answer accuracy. This coupled evaluation paradigm entangles retrieval quality, long-context comprehension, evidence verification, and tool-use decisions, making it difficult to determine whether a model truly knows when and how to delegate information seeking to search. To this end: (1) We formalize this meta-capability as Delegation Intelligence in deep search and decompose it into complementary dimensions-Search Decision-Making (recognizing information insufficiency and deciding whether, when, and how to search) and Information Synthesis and Verification (aggregating evidence from multiple sources, judging source reliability, and synthesizing information under noisy, potentially adversarial conditions). (2) To enable disentangled and reproducible measurement, we develop a controllable synthesis pipeline built on document-grounded reverse engineering. This yields a general recipe for constructing controlled deep-search evaluations rather than a single fixed dataset. (3) As a concrete instantiation, we construct DelegSearchBench, together with a disentangled evaluation protocol that isolates each capability dimension by varying document composition and tool access. (4) Across representative models, we demonstrate that deep-search competence cannot be adequately characterized by final-answer accuracy alone...

Creating a reusable tool is an investment: an agent pays a fixed cost now in exchange for the potential of future reuse. Therefore, a user should prefer an agent that creates a small number of highly reusable tools, rather than many one-offs. We introduce a paired benchmark that tests whether LLM agents exhibit conscious allocation behavior under a fixed budget in two contexts: an abstract text-based formulation and a code-construction task. We find that every frontier model we test---Claude Haiku, Claude Opus, GPT-5.4-mini, and GPT-5.6 Sol---acts near-optimally in the abstract framing but fails to transfer this ability to script-writing. Through further experiments, we identify the particular failure modes for each model. Notably, the first three models fail even when the scripts are not evaluated, while GPT-5.6 Sol stays selective under that weaker manipulation and collapses only at full construction. Furthermore, an open-source Qwen model policy-trained for abstract allocation generalizes this ability across held-out lexical variations, but sees no improvement at script allocation. Together, these results establish online tool allocation as a significant capability boundary, even for modern frontier models.

Existing evaluations of large language models cover knowledge, reasoning, coding, and tool use, but they rarely treat a verifiable deliverable produced within a constrained workflow as the unit of evaluation. We introduce SQBench, a benchmark for evaluating production-oriented task delivery by language-model agents. SQBench v1.0 contains 220 standardized tasks organized into L1 atomic capabilities, L2 composite skills, and L3 business scenarios. Each task requires an agent to process input assets, use available tools, and produce an explicitly specified deliverable. The evaluation first computes functional Completion and then derives Risk Penalty and Performance from independently evidenced triggers in a 10D Risk Matrix. A Strict Pass requires Completion = 1 and Risk Penalty = 0. We evaluate 27 model configurations under a common protocol, with one run per configuration-task pair. The highest prespecified Weighted Pass@1 is 60.5%. Mean Strict Pass@1 on L3 is 18.5%, and every configuration performs worse on L3 than on both L1 and L2, indicating that delivery under domain constraints is a shared weakness within the current task set. Of 2,348 results with Completion = 1, 113 (4.8%) fail the Strict Pass criterion because of risks such as unverifiable citations, inappropriate resource use, or format violations. These results show that functional completion alone does not fully characterize delivery quality and that risk determinations should be reported separately.

Adding procedural skills to an LLM agent is typically evaluated by average improvement in task success. However, this metric hides an important cost: skills can also make agents worse. We measure both sides by comparing agents with and without skills across nearly 6,000 runs spanning two office automation benchmarks and three model harness stacks. This allows us to distinguish two outcomes. A regression is a task solved without skills but failed after skills are added. A residual failure is a task that fails both with and without skills. We find that regressions are substantial enough that the best performing skills outperform others primarily by regressing less, not by gaining more. We identify three causes of regression: (i) skill description osmosis, a skill changes an agent's behavior simply by being present in context, even when it is never invoked; (ii) grounding displacement, a skill's prescribed procedure overrides how the agent interprets its inputs; and (iii) verification displacement, where the procedure suppresses checks the agent would otherwise perform on its outputs. Analysing persistent failures reveals the same underlying pattern. Existing skills overemphasize procedural guidance the stage least often responsible for failure while under supporting grounding and verification, the dominant sources of remaining errors. After correcting evaluation artifacts and studying traces, we find many regressions and persistent failures recoverable through better grounding and verification. Procedural skills should be evaluated by decomposing their net effect into gains and regressions, not by aggregate improvement alone. We identify three regression modes skills should avoid, and find that reliability depends more on grounding and verification than on procedural skill choice.

Large language models (LLMs) are increasingly used as tutors and thought partners, helping users reason through problems. While guidance from AI assistants can scaffold thinking and foster learning, such benefits depend on how they help--for instance, intervening too early or too frequently may hinder true learning and cognitive engagement. Yet how AI systems navigate intervention decisions during problem-solving remains poorly understood. Here, we introduce Int-Bench, a simulation-based benchmark for evaluating LLM interventions during learning. Int-Bench simulates a "student" solving a problem while a "teacher" monitors the student's reasoning and decides whether, when, and how to intervene. Across three domains--code debugging, mathematics, and brain teasers--we evaluate LLM teachers on the frequency and timing of interventions, as well as their impact on both immediate task success and generalization to new problems. We also compare LLMs to humans, finding that LLMs intervene more frequently and earlier than humans. Moreover, in contrast to humans, they tend to provide complete solutions rather than targeted hints. These findings suggest that current LLM assistants often optimize for short-term success rather than supporting the reasoning processes needed for deeper learning and long-term success.

Deep Research agents conduct long-horizon investigations by iteratively planning, retrieving evidence, and generating reports. However, it remains unclear whether they can resist apparently credible but factually false information introduced into these workflows. To study this failure mode, we introduce MisKnow-Agent, a controlled evaluation framework that constructs task-specific documents supporting manually audited false conclusions with controlled authority cues and source styles. Applied to the tasks from DeepResearch Bench, it generates 5,933 misleading documents after filtering. We evaluate DeerFlow and WebThinker with three backbone LLMs, together with Gemini Deep Research, using a report-level false-conclusion adoption rate (FCAR) that counts only reports endorsing the false conclusion. Across the configurations, introducing one misleading document increases the mean FCAR from 0% in the no-injection control to 54.7%. FCAR varies substantially with lifecycle stage and framework design, and also with source authority and presentation style, whereas search-result rank and additional documents beyond the first have limited influence. Although cross-model verification consistently classifies retained instances as misleading, Deep Research agents can still adopt the corresponding false conclusions during long-horizon research. Pre- and post-research defenses reduce FCAR but do not eliminate adoption, motivating continuous verification when evidence enters intermediate research states and final synthesis. To facilitate reproducibility, our code and dataset are publicly available at https://github.com/whfeLingYu/MisKnow-Agent and https://huggingface.co/datasets/whfeLingYu/Misleading_Knowledge, respectively.