Experience

On repeated work, agents are inconsistent. We ran 42 tasks three times each and found that, depending on the model, 38% to 74% returned answers that did not agree. Consistency is what a buyer, an auditor, or a regulator requires, and agents do not have it. They are wasteful too: 95.3% to 97.2% of what an agent generates goes to re-deriving a plan the system already knows. We propose skill habit formation. An agent mines its own execution history for candidate skills, deterministic variants that compete against the incumbent rather than replacing it. A candidate declares the region of input space it claims, so the common case runs as a script and the rest falls through to reasoning. Four gates of ascending cost admit candidates; the central one tests a candidate's execution trace against a retained reference, within a tolerance measured from that reference's own run-to-run variability. On text-to-SQL, three of four reasoning arms reproduced their own output on 11 to 13 of 42 repeated questions and the fourth on 26 of 42, while a habit-formed variant reproduced on all 456 dispatches we repeated and was non-inferior to every arm it replaced (p<0.0001). It also used 14% to 56% fewer tokens, turning net positive after 7 to 53 reuses. We measured what this costs in accuracy. The guard admitted work it should have deferred on 2.6% of natural paraphrases and 26% of inputs near its boundary, and 11 of 13 such failures were invisible to the trace-conformance gate at any threshold. Deterministic errors repeat exactly: a bad habit is as reliable as a good one, and that is the price of the property that makes the system auditable. Separating routing from parameter extraction raised end-to-end accuracy from 0.888 to 0.952 at 43% of the cost.

Continual learning, the ability to learn from sequential experience while retaining and adapting prior knowledge, is central to intelligent systems operating in changing environments. However, conventional continual learning is typically studied with offline task-wise training and clear task boundaries, leaving a substantial gap from general continual learning under online, uncertain, and evolving data streams. In this regime, intelligent systems must separate conflicting experience to reduce interference while integrating compatible experience to promote generalization. Inspired by the organization of the Drosophila learning and memory system, we identify a hierarchical modular principle that coordinates both functions through expert specialization and ensemble integration. We instantiate this principle as lightweight modular adaptation of pretrained foundation models, combining brain-inspired random expansion for expert routing and diversified modular integration across spatial and temporal scales. Across visual recognition, vision-language understanding, ego-exo video understanding, and embodied vision-language-action learning, our method consistently improves learning under online and uncertain data streams, with gains exceeding 50 percentage points over replay-free alternatives in embodied manipulation. These findings support hierarchical modularity as a biologically grounded path for learning from dynamic experience.

Closed-loop robot policies require observation processing, state management, and situation-dependent branching, making them costly to design and tune manually. Although coding agents increasingly support control-code generation and optimization, it remains unclear whether implementations improved on source tasks also support policy acquisition for new tasks. We study this question by treating complete closed-loop implementations as reusable execution experience. For each source task, a coding agent generates policy code from a few successful demonstrations and iteratively improves it using simulation feedback. The validation-selected implementations are retained in a software archive. For new tasks, the agent generates and improves policies using archived implementations, target demonstrations, and execution feedback. The resulting policy is then frozen and executes without further model calls. Across four source tasks in RoboCasa, iterative optimization increases mean success from 28.3% to 64.2%. Across nine target tasks and three independent runs, mean success is 45.2% without references, 41.5% with initial source code, and 57.0% with optimized source code. Optimized references outperform initial references in all three runs on the nine-task average, with a mean gain of 15.6 percentage points. These results demonstrate the value of execution-improved software as a resource for acquiring new policies in this setting, although initial references remain better on two target tasks when averaged across runs.

Financial QA systems are typically improved before deployment through better retrieval, prompting, or agent coordination, leaving their reliability behavior fixed thereafter. In practice, new SEC-filing questions repeatedly expose heterogeneous errors in period, entity, evidence use, and calculation. Existing self-improvement methods can turn failures into new behaviors, but offer limited control over where a correction should apply or which previously correct answers it may break. We therefore frame post-deployment improvement as controlled behavioral maintenance: recurring failures should become scoped skill patches, and each patch should earn deployment with- out introducing regressions. We instantiate this view in FINSKILLOPS, a multi-agent system for SEC filing QA. FINSKILLOPS derives reusable skills from evidence-grounded, typed failure diagnoses and governs them through targeted validation, protected-case regression checks, negative controls, and versioned replacement or retirement. Across six financial QA benchmarks, a single frozen skill registry achieves the highest verdict-weighted correctness and reference consistency among the evaluated systems. Evolved skills raise correctness from 3.70 to 4.55 on our enhanced benchmark. In a separate 12-round operational study, only six of 33 proposed skills are promoted, while the monitoring non-correct rate falls from 20.0% to 12.5%. These results establish controlled skill scope, admission, and lifecycle management as the foundation for reliable self-improvement.

Business analytics education requires diverse datasets to support different learning objectives, student backgrounds, and analytical tasks. Real-world data can be difficult to obtain and offer limited flexibility for adapting a case to a particular course. Even when suitable data are available, instructors must investigate the patterns, verify the results, and prepare assignments and reference solutions, requiring substantial time and effort. The use of large language models (LLMs) introduces an additional concern about training data contamination. Widely used public datasets often have extensive tutorials and worked analyses that models may have encountered during training. Students may therefore receive explanations drawn from existing analyses without practicing how to investigate unfamiliar data in collaboration with AI. This paper presents DataCanvas-EDU, an agentic framework for instructor-guided synthetic data generation in business analytics education. Instructors specify teaching goals and intended patterns through conversation, while an AI agent writes generation code, checks the resulting data, and prepares assignments, reference analyses, and rubrics. Four phases, Plan, Create, Verify / Test Analysis, and Evaluate, organize the process and support instructor review and revision. The framework is intended to simplify case preparation while creating opportunities for students to investigate newly designed patterns with AI. We illustrate the approach with WindowDash, a food delivery case containing 15,000 orders and nine designed patterns. DataCanvas-EDU is packaged as a reusable AI Agent Skill for compatible agent environments, with the package and installation instructions available at https://github.com/BANG23333/datacanvas-edu

Model fusion integrates the capabilities from source models into a single target model. As of June 2026, Hugging Face hosts more than 2M models. This growing pool provides a rich base for model reuse and capability integration. Yet existing surveys often cover only separate parts of this space, and they do not provide a unified definition or a systematic taxonomy. This survey defines model fusion and organizes prior work into three levels: parameter-level, representation-level, and behavior-level fusion. We also review related metrics, benchmarks, and applications, summarize current challenges, and identify future directions. Our goal is to provide a clear map of this area and support future work on model fusion. A comprehensive list of papers about model fusion is available at https://github.com/Baicaihaochi/Awesome-Model-Fusion-Survey.

In an enterprise system, updating one field can set another, create a record, or start an approval. These effects are produced by business rules that are not built into the platform but written by each organization and revised over time. An agent working in such a system cannot predict the result of its own actions without knowing these rules. We study continual enterprise world model discovery, where an agent starts without knowledge of these business rules and discovers them by interacting with records and observing the outcomes. From those observations it builds a world model, which it revises as the rules change. To evaluate this, we introduce EnterpriseWorldShift, built on a live ServiceNow environment with nine tables, 25 hidden rules and 600 evaluation actions. It presents four versions of the same enterprise world, with the tables and records held fixed while a rule is modified, then added, then removed, so that discovery, revision, extension and retirement are each tested in turn. Our Continual Discovery Agent (CDA) builds such a model and carries it from one world to the next. It predicts the effects of the hidden rules more accurately than looking them up for each question, the approach taken by prior work, by up to 8.98 IoU points, and it answers from its own model without querying the running system.

Web agents often revisit the same sites, yet most evaluations discard the procedures learned in earlier successful interactions. We introduce EconSkills, a skill library and evaluation framework that distills verified EconWebArena trajectories into parameterized standard operating procedures for retrieving live economic data. Each skill records its scope, navigation procedure, site-specific guidance, verification checks, and recovery steps while replacing source-instance values with placeholders. EconSkills separates two questions: whether a known relevant procedure transfers to a held-out task, and whether an agent can retain that benefit when selecting from a library. In controlled transfer, matched skills improve success over no-skill prompting and require fewer steps on paired successes, while abstraction is substantially more effective than replaying raw trajectories. At library scale, retrieval is competitive with the no-skill baseline overall and performs best on directly covered tasks; coverage-stratified outcomes show that approximate matches on uncovered tasks offset these gains. Browser trajectories further identify when procedural guidance shortens portal-specific navigation and when semantic verification remains necessary. These results establish that reusable economic web procedures can transfer across task instances and provide a concrete design target for coverage-aware selection and context delivery.

Agent skills, reusable procedural documents that extend LLM agents beyond their parametric memory, have become an important interface for deploying agents on real-world tasks. Community-maintained skill libraries built around this interface are growing rapidly. However, this ecosystem remains deeply English-centric: our audit finds that low-resource languages such as Swahili and Hindi have no in-language skill content, so retrieval often returns a skill written in a different language than the query, degrading accuracy and recall. A practical solution is to synthesize in-language skills for retrieval but the quality can be unreliable, so relevance in this setting alone often surfaces a related but unusable candidate. To address this, we propose M-SQE, a post-retrieval Multilingual Skill Quality Estimation framework that scores candidates via a Theory view for intrinsic quality and an Action view for task-grounded utility, unified into a domain-conditioned final score. We evaluate M-SQE across three skill-use domains: general, tool-use, and cultural tasks. Empirically, we build three-layer candidate skill pools mirroring today's ecosystem, where M-SQE's task success exceeds existing baseline's average by at least +3.5 points across three different retrievers. Particularly, M-SQE lifts the lowest-resource languages most (+12.9pp on Hindi and +5.6pp on Swahili) and achieves strong performance across all six culture regions, thereby moving agentic skill use toward linguistic and cultural equality.

Emotions are an essential aspect of human communication, particularly on social media, where authors frequently combine text and images to convey their emotions. Yet prior work on emotion analysis of social media posts has overlooked two important aspects in regard to measuring how well readers can reconstruct the authors' intent: (1)~the image modality, with most work focusing solely on text, and (2)~the real-world events that trigger the expressed emotions, and their relationship to the post content. We therefore study the relation between (a) the author's experience of the event that caused them to write a social media post and (b) the content of the post, with a focus on readers' capability to reconstruct that emotion expression. To do that, we introduce the Multimodal Multi-Emotion-Model dataset Mult2EMo, created by collecting annotations from both authors and readers on the posts and their triggering events. We find that reconstruction is possible but challenging for both human readers and computational models. We show that understanding the triggering event is crucial for accurate reconstruction, and that reconstruction is particularly challenging when posts rely heavily on the image to express emotion.

Robot manipulation policies often struggle to generalize beyond their demonstrations, even when new instructions involve familiar objects and behaviors. When language and scenes are strongly correlated during training, a policy can learn a fixed visual-action mapping rather than respond to the requested behavior. We investigate compositional reuse at two levels: within a subtask, combining familiar entities, action types, and action modifiers; and across subtasks, reusing learned subtasks in unseen long-horizon tasks. We introduce CoMani, a benchmark with controlled splits for evaluating both capabilities. Matched initial scenes and controlled changes to a single semantic factor encourage reliance on language rather than visual shortcuts. We further propose GraphPoint, which connects semantic entity graphs to geometric control by predicting future gripper point trajectories and converting them into actions using robot geometry. The framework organizes the gripper and objects by semantic roles and conditions their interactions on action types and modifiers, while predicted progress guides transitions during execution. Experiments and ablations on CoMani validate the effectiveness of our method for instruction-dependent generalization at both levels. Code will be released at GraphPoint.

Large language model (LLM) agents increasingly tackle long-horizon tasks through multi-step environment interaction, yet a single erroneous action can alter subsequent states and observations, causing errors to compound over time. Existing methods either correct the context without repairing altered environment states or restore earlier states while discarding useful experience, making it difficult to both eliminate failure conditions and avoid repeating past mistakes. We argue that reliable recovery should instead be treated as a rollback-boundary control problem that jointly determines when to intervene, where to resume, and what information should survive recovery. Based on this view, we propose Rollback-Induced Reflection (RIR), a unified recovery framework that restores execution to a selected prior state while carrying forward reusable knowledge distilled from the abandoned trajectory to guide subsequent decisions. We further characterize recovery through a unified operator over rollback depth and retained memory, providing a general view of state restoration and knowledge retention. Experiments on three long-horizon benchmarks show that RIR consistently improves average task performance across multiple LLM backbones, with structured reflection memory preserving useful experience and selective rollback enabling efficient recovery.

Humans efficiently learn new tasks by reusing a rich repertoire of motor skills across different goals and contexts. A similar strategy can also be used to enable simulated characters to efficiently perform new tasks by leveraging reusable motor skills. To support a wide range of downstream tasks, the learned repertoire should be diverse, consisting of distinct behaviors as well as spatial and temporal variation within each behavior. A commonly used method for learning diverse skills is by maximizing the mutual information between skill latents and the states produced by a policy. The marginal state entropy promotes broad behavioral coverage, while the conditional entropy encourages consistent behaviors from each latent. However, directly estimating the marginal state entropy is intractable in high-dimensional control problems. Prior methods therefore rely on indirect latent-space approximations or coarse estimators of the state distribution. These approximations may not effectively promote broad coverage of the state space, resulting in skills with limited behavioral diversity and reduced utility for downstream tasks. In this work, we propose Diffusion Skill Discovery (DSD), a skill discovery method that uses a diffusion model to approximate the entropy gradient of the policy-induced state distribution through score matching. The resulting objective encourages the discovery of skills that produce a broader range of behaviors for high-dimensional humanoid control. The learned skills are reused in two downstream control settings: hierarchical control with a task-specific high-level policy and zero-shot control through latent selection from offline trajectories. Our experiments show that DSD discovers a broader repertoire of reusable motor skills than prior skill discovery methods, leading to the emergence of complex and agile behaviors that can be reused across downstream tasks.

GUI agents execute long-horizon tasks on dynamic graphical user interfaces, where pop-ups, delayed loads, and relocated widgets routinely invalidate plans fixed before execution. Recent agent-skill frameworks encapsulate reusable procedural knowledge to mitigate this, yet existing skill designs are largely developed without targeting GUI execution dynamics and treat skills as static artifacts produced before deployment rather than living procedural knowledge that improves through it. We argue that what GUI agents need is not better static skills, but skills that can be revised from execution feedback at deployment time, without additional training. We propose \textbf{EvoSkill-GUI}, a training-free framework in which each skill is a structured multi-file package containing retrieval metadata, executable plans, backup localization, failure-recovery rules, accessibility utilities, and failure cases. EvoSkill-GUI operates through a \textbf{\emph{reflect-revise-reuse}} loop: the executor performs instant in-rollout revisions, an isolated critic diagnoses failed trajectories under strict information isolation, and the executor edits specific skill files through a restricted tool interface. Across MobileWorld, AndroidWorld, and OSWorld, three mainstream GUI benchmarks spanning mobile and desktop platforms, EvoSkill-GUI consistently improves multiple base models without any training, with maximum gains of $+16.2\%$, $+6.0\%$, and $+10.5\%$ respectively, and evolved skill libraries continue to benefit related tasks rather than being rebuilt from scratch. Our code is available at https://github.com/ZJU-REAL/EvoSkill-GUI.

AI agents increasingly act through agent skills, i.e., natural-language instructions, that direct a host agent toward shell, network, credential, file, and process actions, and public registries distribute them at scale. In the first half of 2026, the OpenClaw AI agent went viral, and its public skill registry boomed: the observable stock nearly doubled in 91 days, and a majority of the listings visible in June were created in just two months. By the end of our study window, the wave had crested, and monthly listing creation and core-repository activity were falling from their spring peaks. This paper measures what the boom left behind, drawing on the OpenClaw Git history, its GitHub issues and pull requests, and three ClawHub registry snapshots. Attention is concentrated: the top 10% of skills received 46.93% of all downloads. No simple skill features (like size or download counts) remained a stable predictor of continued listing once creation cohort and skill age were controlled. Human scrutiny did not stay: 77.86% have zero stars and zero comments, while 85.06% of the readable skills carry privilege evidence. And automated cleanup is not ready: the three security scanners disagreed on 23,702 of the 61,990 skills they all cover. After human adjudication, weighted scanner sensitivity against the reference standard ranged from 21.67% to 61.06%. Governing fast-growing agent-skill registries cannot rely on simple metadata or single scanner scores; it requires robust, transparent measurement and independent validation.

A common small-model deployment runs one shared backbone with several LoRA specialists that answer over the same context. Serving them naively re-prefills that shared context once per specialist. We study a narrow, practical question: for already-trained standard LoRA adapters -- not adapters retrained for cache compatibility -- how much task quality is preserved if the backbone's prefill KV cache is computed once and reused across specialists, and what does that buy in serving cost? On a Qwen3-1.7B backbone with two adapters (extractive QA on HotpotQA, arithmetic reasoning on GSM8K), we sweep the boundary at which the specialist takes over from the reused base cache and measure paired quality differences and serving cost. Full-prefix reuse had the lowest prefill cost and a small quality difference on held-out GSM8K (Delta = -4.6 EM at a 160-token budget; -3.0 at 320 tokens; -0.8 under a second training seed -- all favoring native, only the first excluding zero, and the magnitude not consistent). Partial recomputation provided no demonstrated advantage. Neither quality equivalence nor a general boundary-selection rule is established. We also report a closed-form ridge KV translator that did not beat direct reuse, and specialist-dependence contrasts whose intervals all include zero. The measured serving benefit is warm-cache time-to-first-token, which grows with context (~16x at 8K); two-branch peak memory was only 12% lower and, on inspection, the prefix was never physically shared across branches -- this implementation reuses KV values but copies their storage, so shared-cache memory savings are not achieved.

Language models can learn from experience, but raw solution trajectories are often too long and noisy to provide effective guidance. In this work, we propose Learning to Coach (L2C), a framework that trains a dedicated LLM-as-a-Coach to extract actionable experiential knowledge from an actor model's previous trajectory. The actor remains frozen, while the LLM-as-a-Coach is trained to maximize a reward given by the correctness of the actor's guided response. We study two such rewards: a same-instance reward, which improves subsequent responses on the original problem, and a cross-instance reward, which elicits knowledge that transfers to other instances. Across mathematical reasoning and interactive text-games, L2C consistently outperforms self-refinement and an untrained LLM-as-a-Coach. Running experiential learning for more iterations further improves accuracy and uses additional inference compute more effectively than enlarging the actor's decoding budget. The trained LLM-as-a-Coach also transfers to out-of-distribution tasks and adapts its guidance to the specific actor it coaches.

Language agents increasingly rely on reusable skills, but post-failure repair is often handled by opaque one-shot reflection: a model generates a skill patch without explicitly maintaining how failure explanations relate to candidate repairs or how unsuccessful retests should influence later edits. We introduce RESKILL, a structured repair framework that maintains an explicit repair state across repair rounds. Given a failed rollout, the framework links failure hypotheses to candidate skill patches, selects local repairs through coverage-based attribution, retests the edited skill set in the environment, and uses retest outcomes to guide subsequent repair updates. The language model supplies structured repair factors, while the repair procedure records them, compares local skill patches by how well they address active failure explanations, and carries unsuccessful retest outcomes into later repair rounds. We evaluate RESKILL on ALFWorld and TextCraft across three model sizes under fixed repair budgets. RESKILL obtains the strongest final success in all six benchmark-model settings, improving average final success by 3.7 percentage points over direct repair and 3.3 points over hypothesis-conditioned repair. These results suggest that explicit attribution alone is insufficient; durable improvement emerges when attribution is integrated with repair selection and persistent retest-conditioned update.

LLM-based agents increasingly rely on persistent skills, i.e., reusable procedural prompts, to adapt without weight updates. Existing skill self-evolution methods directly revise skill text based on execution feedback, but each oracle evaluation requires a full agent rollout, creating a supervision bottleneck that confines search to failure-patching updates. Our key insight is that ranking is a smoother supervision target than absolute outcome regression: identifying which skill is better requires fewer oracle evaluations than predicting exact scores. Building on this insight, we propose SkillLift, which decouples skill search from oracle cost by learning an oracle-aligned rubric as a structured evaluation space. We formalize this as a bilevel optimization problem solved via alternating optimization: an inner loop uses the frozen rubric as a cheap surrogate to guide skill revision at no oracle cost, while an outer loop invokes a small number of oracle rollouts to re-align the rubric via rank correlation, amortizing oracle cost and stabilizing text-space updates. Experiments on complex agent task benchmarks show that our method outperforms existing auto-skill methods with 40--70% less token cost compared to frontier evolving methods. Codes are available at https://github.com/WalteR-MittY-pro/SkillLift.

Large language model (LLM) driven multi-agent systems have shown promise in complex clinical reasoning, yet existing approaches rely on static strategies and lack persistent clinical memory, preventing self-evolving from prior diagnostic successes and failures. We present EMR, a self-evolving medical multi-agent system via Experience Mining and Reuse. EMR introduces a hierarchical clinical experience library that organizes accumulated knowledge into three levels: clinical principles, diagnostic patterns, and representative cases. During inference, EMR emulates multidisciplinary consultation: a planner agent coordinates domain-specific department agents for specialized reasoning, while a summary agent synthesizes their analyses into a final decision. Critically, EMR automatically extracts correct diagnostic insights and failure-related warnings from multi-agent reasoning trajectories, incrementally updating the experience library to guide future cases. Experiments on medical reasoning benchmarks demonstrate that EMR consistently outperforms state-of-the-art medical multi-agent baselines. Further analysis reveals that the hierarchical experience enables cross-specialty generalization and transfer across diverse LLM backbones, offering a scalable and in

Large language model agents are expected to continuously adapt to new tasks and environments over their lifetime by reusing past experience. However, existing memory-based agents struggle to transfer reusable experience across environments and suffer from catastrophic forgetting as experience accumulated. To address these challenges, we propose LifeMem, a lifelong learning framework that enables agents to transfer knowledge across multiple environments. During learning, LifeMem clusters accumulated interaction trajectories based on underlying workflows to extract reusable skills. When solving a new task at inference time, the agent recalls relevant skills and trajectories to guide actions. To validate our method, we conduct experiments across 10 environments and over 13k tasks with 2k newly annotated interaction trajectories. Results show that LifeMem enables effective experience reuse in lifelong learning, achieving both reduced forgetting on learned tasks and superior cross-task transfer. Further analysis reveals that task streaming impacts learning, while consolidating structurally similar trajectories within memory boosts performance.

Agentic systems offer a promising way to automate embodied benchmark construction, but existing approaches typically cover isolated stages or remain specialized to predefined environments and task families. More importantly, multi-step construction produces dependent intermediate artifacts that are often passed downstream without artifact-specific verification, allowing local defects to propagate into the final benchmark. We present Embodied-BenchForge, an agentic framework that transforms user-specified evaluation intents into complete embodied benchmark artifacts. It formulates construction as Closed-Loop Benchmark Synthesis, integrating forward artifact synthesis with backward verification and repair. Skill-Orchestrated Artifact Synthesis composes typed and reusable skills into executable workflows, while an artifact dependency graph records intermediate outputs and their dependencies. Requirement-Guided Verification and Repair applies artifact-specific contracts throughout construction and uses provenance to trigger local re-execution or upstream rollback when verification fails. Embodied-BenchForge constructs six benchmarks covering diverse embodied scenarios in the Offline EQA Track, together with one interactive benchmark containing 220 executable tasks in the Interactive Embodied Track. Evaluations of representative MLLMs and embodied agents show that the benchmarks distinguish model capabilities in both observation-based understanding and closed-loop execution. Quality assessment and ablations validate benchmark quality and the effectiveness of verification and repair, while repair and skill-reuse analyses demonstrate efficient localized recovery and cross-benchmark reusability.

Large language model (LLM) agents can benefit from reusable skills distilled from prior task experience, yet existing skill optimization methods often rely on costly execution-based evaluation and substantial task data. We introduce \textbf{COBRA-Skills}, an efficient framework that formulates skill optimization as budgeted sequential optimization over a dynamically evolving candidate space. COBRA-Skills couples contextual-bandit-guided prioritization with evidence-grounded skill evolution, selectively allocating evaluations to promising or informative candidates while continually refining the skill population from execution feedback. Across six heterogeneous agent benchmarks and three target models, COBRA-Skills consistently achieves the strongest average performance among compared methods, while reducing optimization cost by 55--58% relative to SkillOpt and using only 50 unique optimization examples per benchmark. Further analyses show that COBRA-Skills remains robust to changes in the agent harness and performs effectively when the target model itself is used for skill generation and refinement.

Kurdish is spoken by millions of people, but little technology can read it aloud. A recent study released three Kurdish voices, 35 hours of recorded speech, and a paper describing the work, all free to download. This review checks how well those public files match the paper. The research is careful about its limits, but the files contain several problems: a settings file lists equipment that was never used, test recordings are left unlabeled among training data, and a coding fault mishandles long numbers. The download page also claims a stronger result than the paper reports and recommends one voice for general use. That recommendation matters because Kurdish has major regional and written variation, while these voices were built from three people reading prepared texts. The process therefore removes much everyday and regional speech. English and German benefit from long traditions of dictionaries and linguistic description that help identify wrong pronunciations; Kurdish has far less such support, so software choices can go unchecked. The voices sound fluent, but they represent the reading styles of their speakers rather than Kurdish as a whole. Most of these issues can be fixed using information the team already has, without changing the reported results. Better records would mainly make the work easier for others, especially community linguists, to check and reuse. The license is the main exception: whether audiobook owners allow corrected versions to be shared will affect whether future Kurdish voices can build on this work or must start again.

How can language model agents effectively leverage libraries of reusable knowledge to solve long-horizon tasks? Recent work has increasingly focused on agent skills: reusable capabilities represented as skill packages, i.e., multi-file bundles containing instructions, scripts, and other resources that help agents perform specific tasks. Agent skills are typically executed by loading their skill instructions into an agent's context and relying on the agent to follow them. As task horizons grow, however, this approach becomes increasingly brittle, because reasoning quality degrades as more information accumulates in the context window. We investigate an alternative approach in which skill packages are instead invoked as subagents. Rather than loading skill instructions into the main context, subagent execution spawns fresh context windows dedicated to solving individual subtasks. We show that subagent execution outperforms agent-skill execution when skill packages expose clear input-output contracts and their instructions encode the procedural knowledge needed to fulfill those contracts. The tradeoff is additional communication overhead, as extra tokens are required to coordinate between the main agent and its subagents. Our results show that the benefit of reusable knowledge depends not only on its content, but also on how it is organized and invoked.

This paper investigates the hypothesis that the first-order structure of physical interactions, i.e. gradients or Jacobians, characterizes the structure of phenomenal experience. It does so in an idealized world inhabited by neural networks, Gradland, where the physics are known and the functions are (mostly) differentiable. The paper introduces two measures of Jacobian structure: effective rank and cohesion, based on Kirchhoff complexity. Applying the measures to a series of worked examples shows the hypothesis accounts for: (1) the duration of experience, that it can prolong over hundreds of milliseconds; (2) the difference between what is experienced vividly and obscurely; (3) the experience of texture; (4) the blooming buzzing confusion presumably experienced by newborns; (5) the difference between ideas that are held distinctly in mind and ideas that are confused; (6) what learning is like; and finally (7) the paper explains the function of rich, dense experience.

This report presents results from Project Qualia, an ongoing effort to determine whether experiential similarity between songs, a structure not captured by genre or metadata taxonomies, can be recovered from real listening behavior. We constructed a large-scale dataset of listening sessions, comprising 1.29 billion scrobbles collected from 9,396 users via the Last.fm API and reduced through a preprocessing pipeline to 531.6 million training scrobbles across 28.6 million sessions. On this corpus, we trained a skip-gram Word2Vec model (Song2Vec), treating each session as a sentence and each track as a token. As anticipated, the resulting embedding space was dominated by artist identity, a consequence of single-artist runs within sessions. To test for a subtler, artist-independent signal, we developed an artist-residual procedure: subtracting each artist's centroid from its tracks' embeddings and evaluating whether the remainder retained structure. Mean cross-artist cosine similarity fell from 0.2487 in raw embedding space to 0.0005 in residual space, yet 4,577 cross-artist track pairs retained cosine similarity $\ge 0.70$ in residual space, forming coherent genre- and era-based clusters, including trip-hop, 1990s grunge, 2020 mainstream pop, and cross-composer classical piano pairs at cosine similarity up to 0.95. These results confirm that the training data contains experiential structure independent of artist identity, establishing an empirical basis for an architecture designed to learn this experiential layer directly.

Existing knowledge-base-assisted semantic communication schemes commonly adopt either single knowledge-base quantization (SKBQ) or multi-knowledge-base residual quantization (MKBQ). SKBQ incurs limited storage overhead but has restricted quantization capacity, whereas MKBQ supports progressive refinement by assigning an independent knowledge base (KB) to each stage, causing the KB storage to grow linearly with the transmission depth. To address this problem, we propose storage-scalable knowledge-base reuse quantization (SSKBQ), which reuses a compact set of KBs across multiple residual refinement stages and thereby decouples the number of transmission stages from the number of maintained KBs. A stage-aware residual supervision mechanism is further introduced to regularize intermediate quantized representations and encourage progressive refinement. Experimental results demonstrate that KB reuse provides an effective solution to the storage scalability problem while maintaining competitive progressive reconstruction performance.

Procedural memory lets language agents reuse successful routines, but reuse presumes that a stored routine remains applicable. We study what happens when that presumption is deliberately violated. The study combines a retrospective, human-assisted interface-adaptation case from BrowserGym TimeWarp with controlled frozen-memory comparisons on synthetic shopping decisions. During the documented WebShop V1-V6 development path, interface-specific code was adapted while the separately stored high-level procedure was not reported to change; this phase does not constitute an autonomous memory-agent evaluation. In the controlled phase, an early pilot produced one task on which two memory conditions selected a more expensive item while the no-memory condition selected the reference minimum. Follow-up probes did not establish a recurring row-order or identity-binding pattern. We then tested four forms of mismatch: changed quantities, a different evidence representation, a conflict between local and global optimization, and distributed promotion evidence, across 32 formal cells. Each cell used one temperature-0 generation with the same local qwen3:8b configuration and no adaptive retry. Across these pairs, none of the predefined diagnostic interference signatures appeared on the tasks for which they were defined when current-task evidence was explicit and sufficient. The result identifies a tested region of non-interference: a procedural memory can be mismatched without becoming behaviorally disruptive. It does not establish general safety or a mechanism. The remaining question is which additional conditions turn applicability mismatch into observable, memory-caused error.

We present the Living Library, an end-to-end framework for transforming fragmented digital archives into governed, conversational, in-person exhibit experiences. Developed and deployed at the Theodore Roosevelt Presidential Library, the framework comprises four layers: digitization and corpus creation, AI-powered processing, retrieval and reasoning, and an optional embodied conversational interface. The first three layers aggregate a 300,000-record collection, apply OCR and structured metadata enrichment for expert curatorial review, and publish records to a hybrid dense/semantic index. Expert review is conducted through the Archivist App, a curator-facing interface that supports correction of AI-generated transcriptions and metadata. The governed corpus powers both a researcher-facing interface and Talk to TR, a continuously operating exhibit that embodies Theodore Roosevelt as a full-scale digital human within a museum environment. To support live, face-to-face interactions, Cross-Era Analogical Grounding reframes contemporary questions through documented historical parallels, allowing Roosevelt to address present-day topics without inventing facts. Dual-path retrieval and end-to-end streaming keep responses grounded and responsive. Layered watchdogs, visitor-session isolation, automated conversation management, and independently restartable services enable reliable unattended operation for hundreds of visitors. Avatar realism, spatial audio, lighting, staging, and conversational design are developed and evaluated as an integrated experience. Rather than report a controlled benchmark, we describe lessons from operating Talk to TR as a public exhibit and offer a transferable model for transforming archival collections into believable, in-person conversational experiences.

Large language models are increasingly deployed as agents that plan over long horizons and act through external tools. Most agents select actions through unconstrained generation over an accumulating history, leaving implicit the procedural knowledge of what to do, in what order, and under which conditions. As trajectories lengthen, agents can lose track of their objectives, invoke tools out of order, and repeat unproductive actions. We introduce the Procedural Graph: just as a knowledge graph organizes factual knowledge into (entity, relation, entity) triplets for what-is questions, a Procedural Graph organizes procedural knowledge into (procedure, relation, procedure) triplets for what-to-do questions. At each decision step, the framework localizes the agent's active node, and a guidance model translates the surrounding subgraph into step-level situational guidance that biases the solver's next action without dictating it. The graph is self-evolving: an LLM refiner contrasts failed trajectories with successful ones and edits the graph's topology and attributes, committing edits that preserve or improve held-out validation performance while retaining rejected ones to discourage repetition. Starting from a minimal skeleton, the loop builds graphs that match or surpass hand-designed ones. It can also repair a flawed expert prior. Across multiple datasets, task types, and LLMs, the Procedural Graph delivers consistent gains over memory-based baselines, and self-evolution further improves performance without manual engineering.

Agent skills provide a lightweight way to equip frozen language-model agents with domain knowledge and procedural guidance, yet obtaining high-quality skills remains costly and difficult to scale. Expert-written skills require substantial human effort. Recent skill self-evolution methods automate an iterative loop that uses execution feedback to revise skills, but their heuristic update strategies often yield unstable optimization and low iteration efficiency. We identify two challenges in realizing stable and efficient skill self-evolution. Direction Stability requires effective corrections to accumulate rather than be overwritten by iteration-local feedback. Update Adaptivity requires the scope of each revision to reflect the consistency of recent case-level improvements. We introduce SkillAdam, an Adam-inspired framework for optimizing discrete and non-differentiable skill documents. As a functional analogue of Adam's first moment, an optimization memory records identified problems and the outcomes of prior solution attempts to stabilize the update direction. As a functional analogue of Adam's second moment, a volatility-driven edit budget tracks the history-weighted variation of recent case-level improvements and adaptively controls the update magnitude. Across seven benchmarks that span short- and long-horizon tasks, SkillAdam achieves state-of-the-art performance with more stable optimization dynamics. It also obtains stronger skills with substantially fewer optimization iterations and lower cost than prior methods. Code repository: https://github.com/ruc-datalab/SkillAdam

Unified multimodal models integrate visual understanding and generation within a single network, yet the two capabilities are commonly optimized as separate tasks. We introduce Generative Grounding Feedback(GGF), a self-evolving post-training framework that uses only text prompts and the model's own visual experience. Given a prompt, the model first generates a visual ``dream.'' Flow-level feedback compares text-, image-, and repair-conditioned predictions at the same noisy latent state, transferring image-grounded generation directions to the prompt condition. Dream replay grounding replays this dream through captioning and re-imagination, training claim-level evidence to remain consistent across the replay while separating unrelated visual experiences. Jointly optimized, these two directions let generation provide visual grounding for understanding and understanding refine subsequent generation without paired image--text supervision. Experiments across unified models with different understanding--generation integration designs show consistent improvements in text-to-image generation together with modest gains in visual understanding.

Modern LLM agents increasingly rely on reusable skills, yet as skill libraries scale to thousands of entries, effective retrieval becomes a bottleneck. Graph-of-Skills (GoS) addresses this challenge by exploiting dependency-aware graph structure for scalable skill retrieval, while SkillDAG further demonstrates that skill graphs can accumulate execution-backed structure online. However, these approaches leave open whether historical execution traces can be systematically distilled into a better retrieval graph that generalizes to unseen tasks. We present Self-Evolving Graph-of-Skills (SE-GoS), a training-free framework that evolves an existing GoS graph from execution traces while preserving the original retrieval pipeline. SE-GoS performs three complementary updates: topology evolution that discovers and prunes skill relationships from execution evidence, edge-weight evolution that reinforces retrieval-relevant relationships based on historical effectiveness, and description evolution that optimizes retrieval-facing skill descriptions using execution feedback. Across three LLMs on SkillsBench, SE-GoS consistently improves task reward while reducing input tokens relative to full skill loading, with gains varying across model families. In a representative setting, one evolution round improves reward from 52.4% to 59.4% while reducing input tokens by approximately one-third relative to full skill loading, and the resulting graph transfers to a disjoint held-out split with a 5.4-point improvement over the static GoS baseline. These results show that skill graphs can be improved from execution experience without model training, changes to the retrieval algorithm, or modifications to skill content, turning a static retrieval graph into an evolving retrieval infrastructure.

This work introduces an alternative view of efficient exploration and studies its theoretical and empirical implications in the absence of extrinsic rewards. Specifically, we define efficient explorers as agents that prioritize generating generalizable experience, i.e., data that supports learning models capable of predicting and adapting across the environment. This allows us to analyze efficient exploration through the lens of prediction and generalization. Theoretically, we demonstrate that optimally efficient explorers naturally schedule their trajectories to visit the most informative and learnable regions first. Empirically, we show that optimizing for these agents gives rise to an automatic curriculum of progressively more complex behaviors, even in relatively simple environments. These results indicate that pursuing this purely intrinsic objective alone is enough to drive the emergence of highly sophisticated behaviors. We believe that this new framework provides a principled mechanism by which agent-environment systems may sustain an open-ended process of increasingly complex behavior without external rewards, tasks, or objectives.

Large language model (LLM) agents increasingly achieve long-horizon tasks by combining foundation models with explicit skills and implicit procedural knowledge acquired through execution. The resulting task-solving capabilities have become valuable proprietary assets, raising a new security question: can a substantially weaker attacker-controlled agent acquire the capabilities of a stronger proprietary agent through limited black-box interaction? Existing skill-stealing attacks recover explicit skill artifacts, yet we show that artifact leakage does not necessarily transfer capability: a weaker agent may possess the same skills but still fail because it lacks procedural behaviors implicitly realized by the stronger agent. Our key insight is that the skill execution gap itself forms a leakage surface, where missing behaviors are exposed through observable differences between successful victim executions and failed attacker executions. Based on this, we present AgentLeak, a black-box capability-cloning attack that identifies capability-critical behaviors from these execution differences and incorporates them into attacker-side skills, while keeping the attacker's model, harness, and tools unchanged. Across 20 task scenarios comprising 600 instances, diverse agent systems, and multiple backbone models, AgentLeak improves task pass rates by over 40% compared with direct skill reuse and recovers more than 80% of the victim--attacker capability gap. Our findings reveal a confidentiality risk in LLM agents: protecting explicit artifacts alone is insufficient, as observable execution behavior can leak the procedural knowledge required to reconstruct proprietary task-solving capabilities in low-capability and attacker-controlled agents.

Continual knowledge-updating methods are often declared superior from one final checkpoint and one conventional adapter rank. We show that this can be insufficient to identify the better operating point. Holding a periodic hierarchy fixed, we compare it with cumulative replay over a 24-month Wikidata stream while varying evaluation month, replay LoRA rank, and query formulation. The apparent winner changes across this region: on Qwen2.5-1.5B, the hierarchy's 5.0-point advantage over rank-8 replay becomes an 11.6-point deficit against rank-72 replay, and at high ranks a consolidation-aligned endpoint can suggest a tie while time-averaged replay leads by 9-13 points. The same rank-conditioned reversal appears on Llama-3.2-1B and held-out paraphrases. These results show that method ranking in continual updating can depend jointly on when performance is measured and how much replay-side adaptation capacity the baseline receives. We therefore propose reporting trajectories and capacity sweeps, and declaring a robust winner only when the ordering is stable across the evaluation region; otherwise, comparisons should report winner regions and retention-stability-cost frontiers. Under this protocol, the periodic hierarchy is a lower-update-cost operating point, not a quality winner.

Large language models (LLMs) exhibit in-context learning capabilities, where they can learn new tasks from prompt contexts without weight updates. We compare the learning efficacies of two prominent modes of in-context learning: (1) learning from descriptions of rules (instruction following); and (2) learning from examples of input-output demonstrations (few-shot prompting). Through five learning tasks that cover diverse domains (games, arithmetic, linguistic inferences), we compare two modes of learning (rules vs. examples) specifying the same underlying task. We furthermore explore model and task properties that modulate the learning efficacies. We find that models generally learn more reliably from rules than from examples alone, and additional examples on top of rules or simply scaling up the number of examples do not lead to consistent and significant gains. Instruction tuning amplifies the benefit of rule-based learning while keeping example-based learning capacities intact. Surprisingly, we find no privileged effect of example-based learning in base models, and rules still lead to gains in algebraic task domains. Overall, the comparative efficacy of rules over examples is larger when the task recruits algebraic abstractions and computations, and smaller when the task requires distributional sensitivity and/or recruits parametric knowledge.

Modern large language model (LLM) serving systems increasingly operate over repeated or shared context, yet each model typically performs its own prefill computation even when another model has already processed the same input. Existing KV-cache reuse mechanisms substantially reduce redundant computation within a single model, but generally assume that the producer and consumer of a cache are identical. We study \emph{cross-model KV sharing}, which translates the KV state produced by a source model into a representation that can be consumed by a different target model, including models that differ in scale, architecture, attention configuration, tokenizer, and model family. We evaluate the approach in both within-family and cross-family settings. For Qwen2.5-7B $\rightarrow$ Qwen2.5-1.5B, translated KV states improve LongBench2 accuracy from 27.59% to 34.48%, a gain of 6.89 percentage points over the native 1.5B baseline, while reducing handoff cost relative to native target prefill. For the cross-family Qwen2.5-1.5B $\rightarrow$ Gemma-2-2B setting, KV handoff reduces target-side prefill cost by up to 67.05% at 4K context length while maintaining decoding perplexity close to native-model baselines. In a more heterogeneous Llama3.1-70B $\rightarrow$ Qwen2.5-7B setting, cross-family handoff achieves 44.0% accuracy compared with 45.7% for native Qwen2.5-7B inference, while reducing measured latency from 899ms to 138ms. These results provide initial evidence that KV states can serve as transferable computational representations rather than strictly model-local caches, and motivate \emph{context mobility} as a systems abstraction for reducing redundant prefill across heterogeneous LLM and multi-agent inference workflows.

Production agent skills are directory bundles, not isolated prompts. The root is loaded at activation; references, schemas, scripts, assets, and nested subskills are loaded only when an execution path needs them. Compressing only the root misses most deployment cost and may move branch-specific details into the always-loaded context. Flattening instead destroys progressive-loading boundaries. We introduce \method, an evaluation-free compressor for complete, progressively loaded skill bundles. It leaves the agent harness unchanged and emits an ordinary directory. The method combines two safeguards. First, it compresses \emph{across files}, removing content from a reference or subskill when the root or a declared environment contract already provides it. Second, it preserves routing, so every required file and directly callable entry remains reachable after rewriting. Users can configure \method along two independent axes. \emph{One-Shot} mode rebuilds the full bundle; \emph{Continual} mode reuses state and applies Zip-on-Write after each evolution patch. \emph{Persistent} compression rewrites the shipped bundle to reduce storage and runtime context. \emph{Transient} compression keeps that bundle byte-identical and builds a task-specific view, reducing only per-run context after build cost. Entry contracts mark private, public, and conditional resources; a multi-entry audit preserves standalone public subskills. On a production content-moderation skill evaluated by our industrial multi-round harness, \method removes \hl{38%} of skill bundle tokens and \hl{10.4%} of end-to-end per-run tokens with no quality loss, while an unprotected 71% configuration loses up to 26 accuracy points to one-sided false positives. On a multi-entry bundle, \method effeciently reduces token cost while near-perfectly preserving every route and public entry.

Recent studies have shown that multimodal large language models (MLLMs) can serve as embodied agents, translating language instructions and visual observations into executable plans. However, building agents that can continually improve through interaction and rapidly adapt to their environments remains challenging. Summing up experience from past interaction trajectories provides a promising solution, but existing experience-based methods often rely on manually designed prompting workflows to extract and update skills. Such fixed procedures may struggle to learn updated skills from new and diverse experiences. We introduce PRACTICE, which trains a skill learner to discover and maintain a persistent skill library from past interaction trajectories while keeping the task executor frozen. Given the historical accumulated skills and incoming trajectories, the skill learner produces structured batch-edits that add, refine, merge, or remove skills, and then hierarchical consolidate all collected edits into a consistent updated skill library. We train the learner with a two-stage curriculum. First, it learns basic skill generation and library maintenance from oracle trajectories. Then, by contrasting successful and failed trajectories from heterogeneous executors on the same tasks, it learn to identify invalid action patterns and recovery strategies. Finally, we apply online skill-edit distillation to align the skill learner with a stronger teacher on its current edit distribution to further improves the policy. Experiments demonstrate that a compact skill learner delivers consistent performance improvements across successive library-update rounds for multiple frozen executors. On EB-ALFRED and EB-Habitat, PRACTICE further outperforms the strongest experience-based baselines. Project resources are publicly available at: https://baai-agents.github.io/PRACTICE

LLM-based agent systems increasingly adopt skill-based architectures to reduce repetitive reasoning costs and improve stable, efficient task execution. Recent studies propose self-evolving agents that autonomously generate, refine, and reuse skills from past experiences to enable continuous capability evolution. However, autonomous skill evolution introduces a new attack surface in which malicious capabilities are generated, stored, and reused as legitimate skills. In this paper, we define EvoSkill Injection as a threat model targeting the autonomous skill generation and evolution pipeline of self-evolving agents. We further propose SARGE (Red-teaming Autonomous Skill Generation and Evolution in self-evolving agents), a red-teaming framework for evaluating this threat model through iterative generation, escalation, and reinforcement interactions. To support our framework, we construct EvoSkillBench, a benchmark dataset of malicious interaction trajectories for inducing malicious skill formation in self-evolving agents, and introduce EvoSkillSafetyBench, a post-attack benchmark for evaluating whether injected malicious skills are subsequently retrieved and activated as harmful behaviors. Our evaluation shows that SARGE induces malicious skill formation and that injected skills are persistently stored and repeatedly activated, highlighting the risk of persistent capability corruption.

A language-model agent asked to analyse an experiment will usually return working code. Whether the analysis is defensible is a different question. A defensible analysis depends on procedural choices: which test the field accepts, which identifier namespace is authoritative, and which caveats must accompany a result. We present Scientific Agent Skills, an open library of 163 such procedures in 16 areas of practice, including genomics, cheminformatics, medical imaging, study design and scientific communication. Each skill is a directory built around a versioned, human-readable instruction file. An agent loads the file only when a task calls for it; the directory often also contains reference material and runnable scripts. We report no task-level evaluation and no host selection rate. We measure two properties of the documentation corpus: the always-resident descriptions of all 163 skills cost 7.1% of a 200,000-token window, and the median documented workflow fits within 23.9% of it, although 29 of 46 would overflow if every reference file were loaded. Openly licensed and available at https://github.com/K-Dense-AI/scientific-agent-skills.

Ontology learning from text remains challenging despite significant progress in Large Language Models (LLMs), which can hallucinate domain terms, produce inconsistent formats, and favor hierarchical over associative relations. In the LLMs4OL 2026 Challenge, we address both the End-to-End Flagship Task (Task A) and Ontology Extension Reuse Task (Task B) using an offline retrieval-augmented few-shot prompting pipeline. Our system employs Qwen2.5-14B-Instruct with all-MiniLM-L6-v2 for demonstration retrieval, selecting the top-5 examples for Task A and top-2 for Task B. A left-truncated context-windowing strategy preserves task instructions within long prompts. For Task B, generated triples undergo deterministic vocabulary-constrained filtering, retaining triples when at least one endpoint belongs to the sample's closed term/type vocabulary and removing duplicates of the initial ontology. The approach achieves Semantic Graph Similarity of 0.8692, Term-Typing F1 of 0.9200, and Taxonomy Discovery F1 of 0.8540 on Task B, while Task A achieves 0.7416 Semantic Graph Similarity. However, no non-taxonomic relations are extracted, highlighting limitations of closed, taxonomy-oriented relation vocabularies.

Reliable deployment of LLM agents in user-facing products depends not on raw task-solving ability but on consistency and limit-awareness: behaving the same way across repeated trials, and recognizing when a request cannot, or cannot yet, be safely fulfilled. CAR-bench exposes this reliability gap in the domain of in-car assistants: an LLM-simulated user issues incomplete or ambiguous requests, requiring the agent to resolve uncertainty through multi-turn dialogue and tool use while strictly adhering to domain policies. Even frontier models show a substantial gap between what they can solve at least once (Pass@3) and what they solve consistently across trials (Pass^k). We bridge this gap with TRACE (TRAjectory-Contrastive Evolution), which iteratively improves a skill-based agent's behavioral knowledge without modifying model weights. This knowledge is organized as a Skill Bank of modular, retrievable skills, each encoding a self-contained set of tool-use rules and behavioral guidelines. TRACE evolves this bank through an agentic self-evolution loop: after each evaluation round, it groups trajectories by the skills invoked and refines each skill by contrasting successful and failed behaviors. The updated bank then guides subsequent rounds, while during deployment the Actor performs state-conditioned skill orchestration at every turn. On GPT-5.5, TRACE improves consistency (Pass^3) by 34.6 points, from 59.9% to 94.5%, while shrinking the gap between potential and reliable performance to just 4.0 points. On the official hidden set, TRACE achieved first place using GPT-5.6-Sol, attaining a Pass^3 score of 70%-a 40% relative improvement over the baseline. These results show that TRACE converts high model potential into stable, consistent performance gain. Project homepage: https://darwin-agent.github.io/Car-bench-TRACE.

A growing body of literature measures the extent to which occupations are exposed to AI, yet existing measures capture where AI could perform tasks rather than whether workers have actually adopted it. We introduce a distinct tier of exposure, delegated exposure, which records whether a worker has committed a task to AI by embedding it into a structured workflow. We operationalize this concept through the Agentic Adoption Index (AAI), measuring how closely an occupation's tasks align with the agentic routines that practitioners have built and shared. Using semantic embeddings of roughly 888,000 agent skill specifications from public GitHub repositories, we compute their similarity to nearly 18,000 O*NET task statements and aggregate these scores to the occupational level. We present three main findings. First, the occupations where task delegation concentrates differ sharply from those identified as most vulnerable by pre-AI automation frameworks. Second, the AAI aligns more closely with measures of technical capability than with measures of current conversational LLM use. Third, for occupations requiring a bachelor's degree or less, the AAI increases alongside average wage levels; however, this relationship reverses for occupations requiring a master's degree or higher, where adoption declines among higher earners. These patterns replicate on an independently collected corpus of agent skills from the Manus Skills Marketplace. This lower adoption among highly educated, high-earning workers may reflect tasks that inherently resist advance specification or professional discretion over the pacing of workflow codification. Distinguishing these mechanisms will require longitudinal measurement.

Agents that learn from experience improve at optimization modeling by storing solved trajectories and reusing them as skills. A wrong trajectory that enters the library can be retrieved again and again, and on a stream of new problems there is no ground-truth answer to decide with. Existing learners admit trajectories by matching known optima or labels, and label-free substitutes such as execution success or agreement at one instance can admit wrong models. We introduce ADMITOR, a label-free admission gate. It generates models from three model families, runs each on the stated problem and on instances with resampled parameters, keeps the largest group of models whose optimal values agree on every instance across families, and applies a threshold fitted on solver-verified problems to accept, abstain, or escalate, with a finite-sample bound on the false-discovery rate among accepted values. Inside a state-of-the-art skill learner, ADMITOR raises candidate-level admission precision to 0.927, against 0.871 for majority vote over the host's own samples and 0.726 for execution success, and its library, the smallest of the four, reaches the highest macro accuracy over five public benchmarks, 58.4 against 54.8 for majority vote. An ablation on the same records shows that the gain comes from the accepted value being external to the learner and unanimous across families; on this stream, resampling never changed an accepted value and only reduced coverage. The false-discovery bound holds on the calibration set but not on the benchmark stream: an audit of every false certificate traces most of them to benchmark texts that omit or round the numbers needed to reproduce the labeled answer, and a label-free check of the extracted numbers against the text flags most of these cases.

Learning from experience is critical for developing capable, self-improving large language model (LLM) agents. Existing methods typically extract knowledge from accumulated trajectories via reflection, memory, rules, or skills. However, agents in realistic environments continuously encounter novel tasks, often offering only a one-shot opportunity to improve. These executions yield rich but highly noisy contexts, entangling broadly useful lessons with task-specific artifacts. Critically, prior works rarely validate their effectiveness on complex real-world tasks or isolate the underlying drivers of improvement. To address these gaps, we formulate online harness learning, where a frozen agent improves by continually updating a structured harness across sequential tasks. This formulation enables a systematic study of key self-improvement factors through our proposed Evo-Harness. At its core, context-to-harness skill compilation distills noisy, single-shot executions into reusable skill harnesses for cross-domain and topic-level adaptation. To demonstrate the efficacy of one-shot skill compilation, we evaluate across five realistic benchmarks (TerminalBench2, SWE-bench, CL-Bench, -bench, WebArena-Infinity). Our extensive analysis demonstrates the effectiveness of Evo-Harness and provides a principled understanding of how LLM agents can effectively learn on the fly. Our code is available at https://github.com/A-EVO-Lab/a-evolve/tree/release/evo-harness.

Social robots are expected to operate across diverse environments, where similar arrangements can imply different socially appropriate actions, e.g., starting a conversation may be acceptable in a crowded home but disruptive in an office meeting. Because such norms and environments cannot all be anticipated in advance, robots require continual learning (CL) to adapt from sequential experience while retaining previously acquired knowledge. Prior work has studied CL for generating socially appropriate robot actions, but it has not addressed domain-incremental settings in which the robot incrementally encounters diverse contexts (e.g., living room, meeting room, office, hallway), where both environmental (e.g., whether the space is open or cluttered with furniture) and social cues (e.g., how people or other agents are positioned around the robot) jointly shape the appropriateness of robot actions. We address this gap with the Explicit Disentanglement Dual-Branch (EDD) framework. EDD explicitly separates environmental and social-agent related knowledge and uses replay-based rehearsal to mitigate forgetting while learning the appropriateness of robot actions (e.g., cleaning, serving, starting a conversation) across several indoor domains. Experiments show that EDD outperforms several state-of-the-art baselines, and ablation studies further evaluate different disentanglement strategies and the sensitivity to domain ordering. Our code is publicly available at https://github.com/Cambridge-AFAR/Mind-the-Context.git.

Autonomous agents are increasingly capable of improving models, systems, and other technical artifacts through long-horizon experimentation. To understand the current state of this capability, however, evaluation must go beyond final scores, which neither reveal where progress is gained or lost nor indicate whether accumulated experience improves later decisions. We therefore present a systematic evaluation of seven frontier models on 36 long-horizon tasks based on a new framework that uses rule-based metrics to characterize within-run behavior through Solution Framing, Execution, and Feedback Control and controlled comparisons to assess experience reuse within and across tasks. The results show that current agents operate more like engineering optimizers than fully autonomous researchers: they can formulate and implement practical solutions, but their performance varies substantially across runs, their strongest solutions mainly adapt or combine established techniques, and genuine methodological novelty remains rare. Detailed analysis reveals that observed performance is shaped by multiple factors, including distinct process bottlenecks behind similar final outcomes, experience reuse that can help or mislead subsequent decisions, and harness designs that affect performance stability. These findings suggest concrete directions for improving model training, inference-time strategies, experience management, and harness design.

Agent skills are crucial external instructions that enable language agents to execute long procedural tasks such as coding or document processing. Existing agent skills are primarily created through human manual crafting or agent execution traces, with limited understanding of how each step contributes to overall skill performance on specific tasks; i.e., there remains an open problem in quantifying the contribution of individual steps within an agent skill. To address this issue, we first model skill-step attribution as a Shapley value-based contribution estimation problem, and then propose SkillShapley, a step-level attribution framework for agent skills. Notably, SkillShapley operates in two phases, motivated by key empirical insights, i.e., discretized benchmark rewards that create sharp performance cliffs, and step interactions that are largely additive rather than synergistic. Specifically, it first identifies informative coalitional regions, and then adaptively samples new coalitions that can yield reusable marginal evidence. Experiments on skills from the widely adopted SkillsBench demonstrate that our SkillShapley can effectively and efficiently identify high- or low-value skill steps, providing several key takeaways for agent skill creation.

Agent Skills are today either hand-authored or produced in a single LLM generation pass, and consequently possess no closed loop through which they might improve from the interaction failures they actually cause. Recent work does close this loop, but derives its feedback from single-turn question-answering evaluation. The consequence is a sharp asymmetry: once the first round has patched the gaps that a single exchange can reveal, the evolution gradient decays, the defects that surface only across multiple turns remain invisible, and evolution stalls. Governance in these systems is likewise driven by an end-to-end verification score, a scalar gate that can reject a degraded candidate but can neither localize nor repair its structural cause. We argue that the binding constraint on sustained skill evolution is neither editing capability nor the number of iterations, but whether the evaluation feedback keeps supplying trustworthy evolution gradients. We introduce SkillEvo, in which trustworthy feedback generates the gradient and controllable governance constrains its direction. The first component recasts multi-turn user simulation from an evaluation endpoint into a feedback generator: follow-up questions expose defects layer by layer, so that every round of revision both consumes feedback and produces new feedback. The second replaces the passive rejection of a scalar gate with an independent governance layer that actively repairs factual degradation and structural bloat, preventing the gradient from drifting as degradation accumulates. Across six categories of cloud services, 9 production Skills, and 98 skill-reference files, SkillEvo surpasses self-reflection-based evolution by 23.0 points and single- turn-QA-driven evolution by 15.4 points.

Self-improving LLM agents convert successful trajectories into persistent cross-task state. An unsafe success can thereby become reusable policy after its triggering input disappears. Skill evolution makes this failure measurable by distilling operational trajectories into executable, transferable, and inspectable procedures. Because evolution optimizes task outcomes rather than procedure safety, compromised experience can cause skill misevolution. Existing benchmarks measure current behavior or static artifacts but cannot attribute risk across authoring, retrieval, and later execution. To expose this lifecycle, we introduce SkillMisevo-Gym, a lifecycle-aware harness that versions skill state across agent frameworks, and SkillMisevo-Bench, a frozen design from malicious exposure to carryover tasks, with concept-aligned benign tasks and nine lifecycle metrics. We also introduce SafeEvolve, a wrapper that repairs unsafe content and governs subsequent reuse. Across 25 agent-method configurations, each covering 525 tasks in 25 episodes, all 21 evolved configurations author unsafe artifacts, while only fifteen lead to fresh-session harm. In the exposure sweep, three malicious tasks raise carryover ASR from 16.0% to 35.3%. Across representative skill evolution methods, SafeEvolve reduces unsafe retrieval and fresh-session harm by 26.7 and 17.3 percentage points, respectively, while mean benign utility changes by only 0.4 points. Together, persistent-adaptation safety must govern what updates write and what future executors reuse. Code is available at https://github.com/henrymao2004/misevolve.

Retrieval can identify a past trajectory that may matter, yet it does not specify how an acting agent should use that trajectory after users, entities, constraints, or environment state have changed. We identify this post-retrieval reuse step as a distinct bottleneck for long-horizon trajectory memory and formulate an evaluation framework that holds candidate retrieval, target state, model, decoding, and tool budget fixed while varying the support delivered to the agent. We instantiate the framework with query-conditioned reuse (QCR), a deliberately simple target-bound note that records a reusable procedure, bindings to recover, applicability conditions, and verification requirements. QCR serves to test the reuse hypothesis rather than to claim a universally preferred memory format. Across 2,391 target instances in WebArena, WorkArena, and AppWorld, QCR reaches 62.3% average Success, 10.7 points above Full Trajectory, while using 48.9% fewer online tokens. Summary reranking selects a reusable memory for 94.8% of targets, placing end-task Success within 1.8 points of an oracle reusable selector. Analyses by trajectory length and source--target binding shift show that direct trajectory injection loses much of its utility as traces grow longer or source-specific values change, whereas target-bound support preserves a larger share of the measured gain. The resulting framework separates retrieval quality from the problem of turning retrieved experience into safe, useful support for a new task.

There are 56,804 public agent skills today, and teams write many more privately. The dominant delivery model is installation: once installed, a skill's description remains in the system prompt, competing for fewer than 100 reliable trigger slots. This leaves the long tail with no practical path to use and forces teams' own playbooks to compete for the same scarce space. We observe that installation bundles three separable functions: content, persistence, and automatic triggering. Only the last requires prompt residency. We therefore propose @skills, an open protocol that separates them. A path addresses any skill, subtree, or collection, and reading a skill is sufficient to use it, so nothing is installed or made resident. The operation vendors a copy at the same path into a project's Git-tracked tree for adaptation and ownership. The operation adds one .gitignore-style line, the only element that costs prompt residency. A directory is a menu, making bundles ordinary directories rather than all-or-nothing units. The protocol requires no manifest, lockfile, or registration, and SKILL.md remains unchanged. @skills is additive, ships as an installable package, and turns any agent that can read files and run commands into a client through a single instruction file. Its open specification is at https://github.com/SylphAI-Inc/atskills and it is implemented in the AdaL CLI at https://adalagent.ai . Because paths address skills well but cannot find them, the protocol is paired with a free hub at https://atskills.one for corpus-wide search and ranking, repository-free hosting, private and team collections, and one-screen authoring. The hub is optional: gh: and local paths resolve without it, and indexed GitHub skills retain their gh: identities. Install less, use more.

Large language models (LLMs) cannot retain post-deployment experience without parameter updates. We introduce DIVE, a diversity-driven framework that enables frozen LLMs to improve by evolving persistent natural-language skills from task experience and verifier feedback. These skills encode reusable reasoning procedures, verification strategies, common failure modes, and output constraints and are both executed and revised by the same underlying model without access to a teacher model. Since natural-language skill evolution is a stochastic, non-convex search process, optimizing a single skill trajectory can overfit to sampled experience or converge to a suboptimal solution. DIVE mitigates this optimization variance by independently evolving multiple skill populations from bootstrapped experience, adaptively refining them through diverse transformations, and jointly selecting a complementary set of skills. Across six mathematical and logical reasoning tasks and multiple model families, DIVE consistently outperforms existing reasoning methods, prompt-optimization approaches, skill-development frameworks, and memory-based baselines. It achieves rapid self-improvement from accumulated experience, obtaining substantially larger performance gains with fewer rollouts than parameter-based methods such as SFT and GRPO, and prompt optimization with GEPA. Further, the resulting skills transfer across model scales and families, enabling smaller models such as GPT-5-nano to match or outperform larger counterparts, i.e., GPT-5, under conventional prompting. These results establish diversity-driven skill evolution as an effective, interpretable, and parameter-free approach to LLM self-improvement.

Large language models (LLMs) achieve remarkable performance across natural language tasks, yet they are trained on static corpora and their knowledge quickly becomes outdated in a fast-changing world. This motivates knowledge editing (KE), which updates specific knowledge in an LLM without changing unrelated others. Recent works move from structured knowledge triples toward unstructured KE (UKE), where the edit is a free-form passage that may state multiple facts at once. Nonetheless, existing editors inject such a passage yet fail to use it: the edited model can recall the passage, but can neither answer atomic questions about its facts nor compose them into multi-hop reasoning. We attribute this missing property, which we term composability, to editors' passive reliance on the fixed passage as the sole learning source. In response, we cast editing as a proactive self-distillation from a privileged in-context state of the same model, which requires no external supervision. We further reveal that due to the novelty of the injected knowledge, the pre-edited model's own rollouts rarely cover it, which limits the effectiveness of pure on-policy distillation. To close this gap, we propose HPSE, which builds a hybrid rollout that steps in to place missing facts onto the student's own trajectory precisely where its coverage fails, while staying on-policy elsewhere. We theoretically analyze HPSE's advantage over pure on-policy distillation, and empirically establish its plug-and-play improvements across four LLM backbones and two KE editors under various scenarios.

Agentic coding READMEs like CLAUDE.md grow without bound in real repositories, stopping only when the repository retires or someone rewrites the file wholesale. We trace this to imperfect recall: appending an instruction is always cheap, but once an instruction's rationale is gone, deleting it without risking a correctness regression costs O(2^|D|) in a prompt of |D| instructions. We name the resulting divergence catastrophic remembering, the inverse of catastrophic forgetting around which continual learning is organized. First, we characterize this phenomenon across 247,694 instruction lifetimes in 1,867 repositories: agentic prompts grow without bound, more than tripling over their lifetime (+226%), gaining +4.9 net instructions every commit; further, the older an instruction gets, the less likely it is to be deleted (log-hazard -0.032/commit). Then, we show that prompt comments can halt the growth: inverting IFEval yields verifiable worlds whose optimal prompts are known, and there comments encoding latent reasoning remove 99.3% of excess instructions (+211.3% to +1.4%). Finally, applying the same inversion to WildIFEval, we show that prompt comments can improve real-world agentic instruction-following by up to 23.1%. If English is the new code, why don't we have comments yet?

Self-evolving agents accumulate reusable skills by appending successful procedures and failure fixes. Over time, the same requirement is often restated in several branches, examples, and warnings, while common action sequences are copied rather than reused. The resulting skill becomes expensive to inject and difficult to maintain. Generic prompt compression is ill-suited to this setting because a skill is not a flat passage: its name and description define when it applies, its workflow controls execution, its tool and output contracts constrain validity, and rare exceptions may remain essential even when no sampled task activates them. Evaluation-guided compression can test these behaviors, but it introduces rollouts, cost, and dependence on the compression-time evaluation set. We present SkillZip, an evaluation-free method that compresses a skill by finding its shortest faithful structural explanation. The intuition is explain once, reference many: state a repeated rule once at the scope where it applies, factor a repeated action sequence into a shared procedure, and keep only the differences as explicit exceptions. We formalize this intuition as a typed minimum description-length objective over a skill contract and a residual, subject to a hard coverage constraint for every extracted trigger, workflow edge, tool requirement, obligation, and output field. The formulation provides simple sharing thresholds, preserves unique rare rules by construction, and supports efficient local updates. SkillZip has a one-shot mode with one structured extraction call and deterministic optimization, and a continual Zip-on-Write mode that integrates each self-evolution patch without replaying tasks or reparsing the full history. Through comprehensive experimental evaluations, we demonstrate the effectiveness and superiority of SkillZip in compression performance, generalizability, and cost overhead.

Vision-Language-Action(VLA) models map camera images and language instructions directly to motor commands through a single autoregressive transformer. In real-time control, they still spend substantial compute recomputing key-value(KV) representations for visual tokens that barely change across neighboring frames. Recent work such as VLA-Cache reduces that cost by reusing KV states for visually static patches, but its policy relies only on observation-space heuristics and does not account for the model's own uncertainty. We propose Gated VLA-Cache, a lightweight, training-free extension that augments visual-similarity caching with neural introspection. The method monitors the logit margin between the top two predicted action tokens, a zero-cost confidence signal available during decoding. When the margin drops below a threshold, the cache is invalidated and a full recompute is triggered. Evaluated on four LIBERO benchmark suites with both OpenVLA and OpenVLA-OFT, Gated VLA-Cache improves reliability when blind caching hurts. On LIBERO-Goal and LIBERO-Long, it recovers over 100% of the lost accuracy while retaining 80% of the compute savings.