Evolutionary Algorithms

Long-context dialogue systems suffer from state inertia, where models over-attend to history and fail to adapt to evolving intents. We demonstrate that standard alignment methods like DPO and even recent long-context optimization techniques struggle to resolve this without incurring a severe contextual alignment tax--a substantial perplexity surge caused by disrupting pre-trained priors. To address this, we propose DZ-TiDPO, a minimally invasive framework that synergizes conflict-aware optimization (during training) with a structural temporal attention bias. This design effectively decouples state updating from general linguistic modeling. Experiments on Multi-Session Chat and our new Inertia Challenge (IC-Bench) show DZ-TiDPO preserves structural coherence while resolving inter-turn conflicts. Crucially, our framework supports dual inference strategies: a negligible-latency static mode for general robustness and a precision-focused dynamic mode for micro-semantic conflicts. Furthermore, our scaling analysis reveals a capacity-stability trade-off, confirming that highly capable mid-sized models (7B) can efficiently internalize temporal alignment. Code and data are available at: https://github.com/lyj20071013/DZ-TiDPO.

Progress in computational science depends on complex numerical workflows that must faithfully encode physical laws, yet translating conceptual insight into reliable code remains a major bottleneck. Although large language models can generate isolated code fragments, they lack the structured reasoning required to design, verify, and iteratively refine complete scientific pipelines. Here we introduce ATHENA, an agentic framework explicitly designed to emulate scientific research modeled as a knowledge-driven contextual bandit process. Its core loop separates conceptual policy from numerical realization through expert-derived conceptual scaffolding, enabling principled diagnosis, reformulation, and repair of computational strategies. Across scientific computing and scientific machine learning tasks, ATHENA autonomously derives and correctly applies exact analytical solutions, constructs stable numerical solvers, diagnoses ill-posed formulations, and orchestrates hybrid symbolic-numeric workflows. Quantitatively, ATHENA matches and frequently surpasses the accuracy of expert-authored reference solutions reported in the literature on canonical benchmarks. By reframing computation as an object of agentic reasoning, our framework enables autonomous orchestration of heterogeneous algorithms across scientific domains.

We present an algorithm for computing all evolutionarily stable strategies in nondegenerate normal-form games with three or more players.

Large Language Model (LLM) Agents, often trained with Reinforcement Learning (RL), are constrained by a dependency on human-curated data, limiting scalability and tethering AI to human knowledge. Existing self-evolution frameworks offer an alternative but are typically restricted by the model's inherent capabilities and single-round interactions, hindering the development of complex curricula involving tool use or dynamic reasoning. We introduce Agent0, a fully autonomous framework that evolves high-performing agents without external data through multi-step co-evolution and seamless tool integration. Agent0 establishes a symbiotic competition between two agents initialized from the same base LLM: a curriculum agent that proposes increasingly challenging frontier tasks, and an executor agent that learns to solve them. We integrate external tools to enhance the executor's problem-solving capacity; this improvement, in turn, pressures the curriculum agent to construct more complex, tool-aware tasks. Through this iterative process, Agent0 establishes a self-reinforcing cycle that continuously produces high-quality curricula. Empirically, Agent0 substantially boosts reasoning capabilities, improving the Qwen3-8B-Base model by 18% on mathematical reasoning and 24% on general reasoning benchmarks. Code is available at https://github.com/aiming-lab/Agent0.

We analyze an algorithm for assigning weights prior to scalarization in discrete multi-objective problems arising from data analysis. The algorithm evolves weights (interpreted as the relevance of features) by a replicator-type dynamic on the standard simplex, with update indices computed from a normalized data matrix. We prove that the resulting sequence converges globally to a unique interior equilibrium, yielding non-degenerate limiting weights.

Fine-tuning large language models (LLMs) for downstream tasks is an essential stage of modern AI deployment. Reinforcement learning (RL) has emerged as the dominant fine-tuning paradigm, underpinning many state-of-the-art LLMs. In contrast, evolution strategies (ES) has largely been overlooked due to the widespread belief that it does not scale to modern model sizes. This paper overturns this assumption by demonstrating the first successful application of ES to full-parameter fine-tuning of LLMs at the billion-parameter scale, without dimensionality reduction. ES can indeed search over extremely high-dimensional parameter spaces and outperform established RL implementations across multiple axes, including improved tolerance to long-horizon and delayed rewards, robustness across diverse base LLMs, reduced susceptibility to reward hacking, and improved training stability. These findings suggest that ES is not merely a viable alternative to RL, but a fundamentally different and powerful backpropagation-free post-training paradigm that opens a new direction for LLM fine-tuning beyond current RL-based approaches.

Large language models (LLMs) are increasingly integrated with evolutionary computation to support optimization tasks. This survey primarily focuses on evolutionary optimization, i.e., optimization based on evolutionary computation. For brevity, we use the term optimization throughout to denote this scope. However, existing surveys typically examine isolated roles of LLMs and do not provide a unified view that connects optimization modeling with optimization solving. To address this gap, we systematically review recent developments through a workflow-oriented framework. First, we organize the literature into two primary stages: LLMs for optimization modeling and LLMs for optimization solving (in this survey, the terms optimization modeling and optimization solving are used as concise forms of optimization problem modeling and optimization problem solving, respectively). Second, we divide the solving stage into three paradigms according to the role of the LLM: stand-alone optimizers, low-level components embedded in optimization algorithms, and high-level managers for algorithm selection and generation. Third, we analyze representative methods, identify their technical limitations, and clarify their relationships with traditional optimization approaches. We further substantiate this taxonomy through benchmark systematization, baseline comparisons, and practitioner-oriented guidance, and we review interdisciplinary applications across the natural sciences, engineering, and machine learning. Based on the resulting analysis, we identify research directions toward dynamic, self-evolving, and agentic optimization ecosystems. An up-to-date collection of related literature is maintained at https://github.com/ishmael233/LLM4OPT.

Artificial learning systems are graduating from passive learners to increasingly autonomous agents, lending pragmatic urgency to the question of what constitutes agency. Reinforcement learning (RL) offers arguably the most explicit formulation of agent-environment interaction, built on three core tenets: the environment as a Markov decision process, learning as policy optimization, and the agent as a maximizer of scalar reward. Recent work has called to revise these tenets: reconceptualizing learning as adaptation rather than optimization, broadening goals beyond scalar reward, and noting the absence of a formal theory of the agent in a formalism that so heavily emphasizes the environment. We argue that the artificial life community is uniquely positioned to illuminate this critique and concretize an alternative. We draw on open-ended novelty search as a complementary model of adaptation and goal-directed behavior beyond reward optimization, and ground such evolutionary dynamics in thermodynamic theories of origin-of-life and agency, toward a more biologically faithful and formally grounded account of what it is to be an adaptive agent.

Large Language Models have achieved remarkable performance across a large number of tasks, but face critical deployment and usage barriers due to substantial computational requirements. Model compression methods, which aim to reduce model size while preserving its capacity, are an important means to mitigate these issues. Promising approaches along these lines, such as structured pruning, typically require costly manual hyperparameter exploration or rely on local heuristics that may run the risk of ignoring better solutions. In this work we introduce GeLaCo, an evolutionary approach to LLM compression via layer collapse. Our approach supports an efficient exploration of the compression solution space via population-based search and a novel layer collapse formulation based on parametrized weight merging, with a fitness function based on similarity over residual updates and language modeling KL divergence. GeLaCo also supports both single and multi-objective evolutionary compression search, establishing the first Pareto front estimation along compression and quality axes. We evaluate GeLaCo solutions via both perplexity-based and generative evaluations over foundational and instruction-tuned models, outperforming state-of-the-art alternatives.

Online alignment of large language models (LLMs) is dominated by reinforcement learning from human feedback (RLHF) with gradient-based optimizers such as PPO or GRPO. While effective, these pipelines require backpropagation through long rollouts, gradient synchronization across devices, and careful hyperparameter tuning, all of which become increasingly costly at scale. We present ESSA (Evolutionary Strategies for Scalable Alignment), a gradient-free online alignment stage that follows supervised fine-tuning (SFT) and replaces the gradient loop with inference-only black-box search. ESSA optimizes only the singular values of low-rank adaptation (LoRA) factors after a short SFT warm-start, restricting the search to a compact, task-aligned subspace where evolutionary search is practical even for 72B-parameter models. Because the loop is inference-only, ESSA is compatible with INT4/INT8 weight quantization and reduces inter-GPU communication to a few bytes per iteration. Across instruction following (IFEval), preference-based assistant tuning (HelpSteer2, HH-RLHF), and mathematical reasoning (GSM8K, MATH500), ESSA matches or exceeds LoRA-GRPO in the reported LoRA comparisons; on GSM8K it also outperforms Online DPO and PPO, while remaining competitive with both methods on IFEval. At scale, ESSA reaches a fixed MATH500 accuracy on Qwen2.5-32B/PRM800K up to 7.8x faster than LoRA-GRPO on 128 GPUs.

Consensus-based optimization (CBO) has established itself as an efficient gradient-free optimization scheme, with attractive mathematical properties, such as mean-field convergence results for non-convex loss functions. In this work, we study CBO in the context of closed-box adversarial attacks, which are imperceptible input perturbations that aim to fool a classifier, without accessing its gradient. Our contribution is to establish a connection between the so-called consensus hopping as introduced by Riedl et al. and natural evolution strategies (NES) commonly applied in the context of adversarial attacks and to rigorously relate both methods to gradient-based optimization schemes. Beyond that, we provide a comprehensive experimental study that shows that despite the conceptual similarities, CBO can outperform NES and other evolutionary strategies in certain scenarios.

The serious games between humans and AI have only just begun. Evolutionary Game Theory (EGT) models the competitive and cooperative strategies of biological entities. EGT could help predict the potential evolutionary equilibrium of humans and AI. The objective of this work was to examine EGT models relevant to human-AI interaction, evolution, and co-evolution. Of thirteen EGT models considered, three were examined: the Hawk-Dove Game, Iterated Prisoner's Dilemma, and the War of Attrition. This selection was based on the widespread acceptance and clear relevance of these models to potential human-AI evolutionary dynamics and co-evolutionary trajectories. The Hawk-Dove Game predicts balanced mixed-strategy equilibria based on the costs of conflict. Iterated Prisoner's Dilemma suggests that repeated interaction may lead to cognitive co-evolution. The War of Attrition suggests that competition for resources may result in strategic co-evolution, asymmetric equilibria, and conventions on sharing resources. Each model was examined from the perspective of human and AI decision-making, from psychological and biological perspectives, and from an AI viewpoint. AI is being shaped by human input and is evolving in response to it. So too, neuroplasticity allows the human brain to evolve in response to stimuli. If humans and AI converge in future, what might be the result of human neuroplasticity combined with an ever-evolving AI? There are profound ethical and cognitive implications. EGT may provide a suitable framework to understand and predict human-AI interaction, evolution, and co-evolution. However, future research should extend beyond EGT and explore additional frameworks, empirical validation methods, and interdisciplinary perspectives. In the spirit of further exploration, an illustrative computational simulation is provided.

Federated learning seeks to foster collaboration among distributed clients while preserving the privacy of their local data. Traditional federated learning methods typically assume a fixed setting, where participating clients, client data, and learning objectives remain unchanged. However, in real-world scenarios, a federation may evolve over time, with changes in both its client composition and target label space. In this evolving federated setting, conventional round-wise model aggregation becomes inflexible, as each federation update requires repeated communication, repeated local computation, and synchronized participation from all accumulated clients. To address this limitation, we propose CA-MMDS, a continual multiple-model distillation framework for federated continual segmentation with asynchronous clients and evolving label spaces. Instead of repeatedly aggregating model parameters from all clients, CA-MMDS maintains a server-side archive of client models and updates the global model through proxy-based distillation from multiple archived local models. When new clients join or existing clients evolve, only the newly added or updated local models need to be uploaded, while unchanged clients can remain offline and continue to contribute through their archived models. This design substantially reduces communication and computation costs while enabling flexible asynchronous cooperation among evolving clients. Using multi-class 3D abdominal CT segmentation as an application task, we demonstrate that CA-MMDS efficiently incorporates evolving client knowledge while achieving competitive segmentation performance.

Multi-task optimization is typically characterized by a fixed and finite set of tasks. The present paper relaxes this condition by considering a non-fixed and potentially infinite set of optimization tasks defined in a parameterized, continuous and bounded task space. We refer to this unique problem setting as parametric multi-task optimization (PMTO). Assuming the bounds of the task parameters to be ($\boldsymbolθ_l$, $\boldsymbolθ_u$), a novel ($\boldsymbolθ_l$, $\boldsymbolθ_u$)-PMTO algorithm is crafted to operate in two complementary modes. In an offline optimization mode, a joint search over solution and task spaces is carried out with the creation of two approximation models: (1) for mapping points in a unified solution space to the objective spaces of all tasks, which provably accelerates convergence by acting as a conduit for inter-task knowledge transfers, and (2) for probabilistically mapping tasks to their corresponding solutions, which facilitates evolutionary exploration of under-explored regions of the task space. In the online mode, the derived models enable direct optimization of any task within the bounds without the need to search from scratch. This outcome is validated on both synthetic test problems and practical case studies, with the significant real-world applicability of PMTO shown towards fast reconfiguration of robot controllers under changing task conditions. The potential of PMTO to vastly speedup the search for solutions to minimax optimization problems is also demonstrated through an example in robust engineering design.

While guided decoding, especially value-guided methods, has emerged as a cost-effective alternative for controlling language model outputs without re-training models, its effectiveness is limited by the accuracy of the value function. We identify that this inaccuracy stems from a core distributional gap: existing methods train static value functions on trajectories sampled exclusively from the base policy, which inherently confines their training to a narrow and suboptimal view of the potential output space. We propose Iterative Value Refinement, a evolutionary framework designed to narrow this gap. It employs Value Exploration to provide a more comprehensive and robust training signal, complemented by Iterative Self-Refinement, which uses the improved value function from one iteration to guide the generation of higher-quality data for the next. Extensive experiments on text summarization, multi-turn dialogue, and instruction following demonstrate the effectiveness of our framework in aligning language models. Our approach not only achieves alignment but also significantly reduces computational costs by leveraging principled value function optimization for efficient and effective control.

Since antiquity, musical scales have been explained by harmony rather than melody. This view relies on the mathematically designed scales of a few traditions, and was never directly tested. Testing it requires cross-cultural data and a method that judges theories by what they get wrong as well as right. We provide both, modelling scale evolution across 1,314 scales from 96 countries. A Melody model explains the near-universal preference for step-sizes of 1-3 semitones, and matches independent data from melodies, singing, and psychoacoustics. Harmony does far less: it explains the music-theoretic scales, but in those measured from performance it adds only a weak bias towards fourths, fifths, and octaves. Harmony's importance has been overstated, likely due to the historical focus on music-theoretic rather than measured scales. Melody is the primary driver of global scale structure; harmonic constraints are less impactful and mainly reflect musicological theory over musical performance.

Much of Western classical music relies on instruments based on acoustic resonance, which produce harmonic or quasi-harmonic sounds. In contrast, since the mid-twentieth century, popular music has increasingly been produced in recording studios, where it is not bound by the constraints of harmonic sounds. In this study, we use modified MPEG-7 features to explore and characterise the evolution of noise and inharmonicity in popular music since 1961. We place this evolution in the context of other broad categories of music, including Western classical piano music, orchestral music, and musique concrète. We introduce new features that distinguish between inharmonicity caused by noise and that resulting from interactions between discrete partials. Our analysis reveals that the history of popular music since 1961 can be divided into three phases. From 1961 to 1972, inharmonicity in popular music, initially only slightly higher than in orchestral music, increased significantly. Between 1972 and 1986, this rise in inharmonicity was accompanied by an increase in noise, but since 1986, both inharmonicity and noise have moderately decreased. In recent years (up to 2020), popular music has remained much more inharmonic than popular music from the 1960s or orchestral music involving acoustic resonance instruments. However, it has become less noisy, with noise levels comparable to those of orchestral music. We relate these trends to the evolution of music production techniques. In particular, the use of multi-tracking may explain the higher inharmonicity in popular music compared to orchestral music. We illustrate these trends with analyses of key artists and tracks.

Molecular optimization, the process of designing molecules with desirable properties, represents a critical challenge in drug discovery. The recent advancements in large language models (LLMs) have opened new opportunities for their integration with traditional molecular optimization algorithms to improve performance. In this work, we propose Molecular Language Model powered Evolutionary Algorithm (Mol-E), an evolutionary algorithm that relies on the generative capabilities of LLMs trained on molecules and molecular properties. Scientific Contribution. Mol-E obtains the highest aggregate Top-10 AUC among the comparable full-23-task results considered here, scoring 17.500 in the task-agnostic regime, in which the oracle is treated strictly as a black box, and 20.551 in the task-informed regime, in which the optimizer receives a fixed semantic description of the objective. Mol-E also improves over the evaluated baselines on multi-property optimization with docking against DRD2, MK2, and AChE.

Phylogenetics is a branch of computational biology that studies the evolutionary relationships among biological entities. Its long history and numerous applications notwithstanding, inference of phylogenetic trees from sequence data remains challenging: the high complexity of tree space poses a significant obstacle for the current combinatorial and probabilistic techniques. In this paper, we adopt the framework of generative flow networks (GFlowNets) to tackle two core problems in phylogenetics: parsimony-based and Bayesian phylogenetic inference. Because GFlowNets are well-suited for sampling complex combinatorial structures, they are a natural choice for exploring and sampling from the multimodal posterior distribution over tree topologies and evolutionary distances. We demonstrate that our amortized posterior sampler, PhyloGFN, produces diverse and high-quality evolutionary hypotheses on real benchmark datasets. PhyloGFN is competitive with prior works in marginal likelihood estimation and achieves a closer fit to the target distribution than state-of-the-art variational inference methods. Our code is available at https://github.com/zmy1116/phylogfn.

In recent years, with the increase of social investment in scientific research, the number of research results in various fields has increased significantly. Cross-disciplinary research results have gradually become an emerging frontier research direction. There is a certain dependence between a large number of research results. It is difficult to effectively analyze today's scientific research results when looking at a single research field in isolation. How to effectively use the huge number of scientific papers to help researchers becomes a challenge. This paper introduces the research status at home and abroad in terms of domain information mining and topic evolution law of scientific and technological papers from three aspects: the semantic feature representation learning of scientific and technological papers, the field information mining of scientific and technological papers, and the mining and prediction of research topic evolution rules of scientific and technological papers.

According to E.T. Jaynes and E.P. Wigner, entropy is an anthropomorphic concept in the sense that in a physical system correspond many thermodynamic systems. The physical system can be examined from many points of view each time examining different variables and calculating entropy differently. In this paper we discuss how this concept may be applied in information entropy; how Shannon's definition of entropy can fit in Jayne's and Wigner's statement. This is achieved by generalizing Shannon's notion of information entropy and this is the main contribution of the paper. Then we discuss how entropy under these considerations may be used for the comparison of password complexity and as a measure of diversity useful in the analysis of the behavior of genetic algorithms.

In general, we can not use algebraic or enumerative methods to optimize a quality control (QC) procedure so as to detect the critical random and systematic analytical errors with stated probabilities, while the probability for false rejection is minimum. Genetic algorithms (GAs) offer an alternative, as they do not require knowledge of the objective function to be optimized and search through large parameter spaces quickly. To explore the application of GAs in statistical QC, we have developed an interactive GAs based computer program that designs a novel near optimal QC procedure, given an analytical process. The program uses the deterministic crowding algorithm. An illustrative application of the program suggests that it has the potential to design QC procedures that are significantly better than 45 alternative ones that are used in the clinical laboratories.

Neuroevolution of Augmenting Topologies (NEAT) and its advanced version, Evolvable-Substrate HyperNEAT (ES-HyperNEAT), have shown great potential in developing neural networks. However, their effectiveness heavily depends on the selection of hyperparameters. This study investigates the optimization of ES-HyperNEAT hyperparameters using the Tree-structured Parzen Estimator (TPE) on the MNIST classification task, exploring a search space of over 3 billion potential combinations. TPE effectively navigates this vast space, significantly outperforming random search in terms of mean, median, and best accuracy. During the validation process, the best hyperparameter configuration found by TPE achieves an accuracy of 29.00% on MNIST, surpassing previous studies while using a smaller population size and fewer generations. The transferability of the optimized hyperparameters is explored in logic operations and Fashion-MNIST tasks, revealing successful transfer to the more complex Fashion-MNIST problem but limited to simpler logic operations. This study emphasizes a method to unlock the full potential of neuroevolutionary algorithms and provides insights into the hyperparameters' transferability across tasks of varying complexity.

Robotic cellular warehousing systems (RCWS) give rise to multi-agent pickup and delivery (MAPD) processes in which robots sequentially collect multiple stock-keeping units (SKUs) for each order. Unlike classical MAPD formulations that assume static tasks, real warehouse operations often involve dynamic order evolution, where new SKUs may be appended to an order while it is being executed. Motivated by this practical requirement, this letter formulates the Dynamic-MAPD problem considering internal order evolution for the first time. Building on the token passing (TP) mechanism, we propose two event-triggered online replanning algorithms. The two strategies target different robot-resource configurations, depending on whether additional robotic resources are available for cooperative assistance. The first, Dynamic-TP, enables an event-triggered dynamic response by allowing robots to replan from their current execution states through priority-aware token acquisition after order updates. The second, Cooperative-TP, further enables reserved robots to assist newly added SKUs while preserving the original order ownership. Simulation results demonstrate that the proposed methods significantly reduce order flowtime compared with static and non-cooperative baselines, thereby improving the order fulfillment efficiency in RCWS.

Embodied agents are increasingly built as systems around foundation models, where performance depends not only on model weights but also on the skills, context, action interfaces, and execution harness surrounding the model. While supervised fine-tuning and reinforcement learning can adapt agents to new environments, they require additional data, rewards, and training runs; meanwhile, many train-free code-centric approaches rely on programmable robot APIs that may be unavailable in fixed-interface settings. We propose SHAPER, a self-evolving framework for train-free embodied adaptation that keeps model parameters frozen and improves the non-parametric agent system by evolving reusable skills and a context-code harness through target-environment rollouts. In SHAPER, the same frozen model can serve as both planner and optimizer, refining its external skills and context-code harness without parameter updates. We evaluate SHAPER on VLABench and ESI-Bench, covering embodied agents with different low-level action interfaces, and compare against pure execution, supervised fine-tuning, and test-time-scaling baselines such as verifier-free selection and voting. Our results suggest that skill-and-harness optimization is a practical route to self-evolving embodied agents when model training is expensive, unavailable, or undesirable.

The convergence of large language models and agents is catalyzing a new era of scientific discovery: Agentic Science. However, common agent infrastructure is repeatedly rebuilt across scientific fields (loop fragmentation) and useful evidence and experience are lost in long-horizon research (loop discontinuity), bringing obstacles to Agentic Science at Scale. We introduce EvoMaster, a foundational evolving agent framework for Agentic Science at Scale. EvoMaster handles loop fragmentation and loop discontinuity by implementing Loop Research in which external evidence persists and improves later decisions. Through three nested loops, Execution, Exploration and Evolution (E$^3$), loop research connects research within runs, across experiments and across studies. Across ten benchmarks spanning scientific research, coding and reasoning, EvoMaster achieves the best score among four agents using GPT-5.4, reaching a mean score of 58.02%, and outperforms the strongest competing agent Codex(40.29%) while costing 35.6% less. These results show that a shared loop-research foundation can support diverse scientific agents at scale.

Quantitative strategy optimization remains largely manual, requiring domain experts to identify weak signals, tune risk-control rules, and repeatedly validate iterative revisions. Large language models can accelerate this process, but directly relying on them to rewrite trading strategies often introduces hallucinated edits, strategy drift, and backtest overfitting. We propose EVOQUANT, a self-Evolving Verifier-guided framework for strategy Optimization in Quantitative trading. Our method utilizes LLMs to deeply diagnose performance bottlenecks, generates semantically controlled candidate edits, selects the best strategy through a multi-stage verification pipeline, and distills optimization experience into reusable knowledge for continual self-improvement. We evaluate our method using seven representative strategies: four from the A-share market and three from the Crypto market. Experimental results show that our method significantly improves the Sharpe ratio across all tested strategies: the average test Sharpe increases from -0.298 to 0.538, and the best-performing strategy achieves a 199% relative improvement. Ablation studies and stress tests under stricter conditions further validate the effectiveness and robustness of the framework. Overall, this work transforms quantitative strategy optimization from costly manual trial and error into an automated and verifiable iterative paradigm, offering a new path for applying large language models to financial strategy research.

Cumulative culture enables humans to preserve, reuse, and extend knowledge and skills across experiences and generations. Inspired by this principle, we introduce \textit{SimSkill}, a self-evolving agent built around the Simulation of Urban MObility (SUMO) traffic simulator. SimSkill continually identifies capability gaps, generates and solves environment-grounded tasks, verifies solutions through an action--critic loop, and consolidates experience into episodic, procedural, and semantic memory. Through autonomous exploration, it builds a library of reusable skills and knowledge spanning major stages of the traffic-simulation workflow. We evaluate SimSkill on two held-out benchmarks across three backbone LLMs, with each result independently verified. It improves verified success by up to 25 percentage points, and ablations show complementary contributions from procedural and semantic memory. Its benefits remain backbone- and budget-dependent, as memory does not improve every model or uniformly reduce inference cost. More broadly, SimSkill illustrates a natural-language-centered design paradigm for LLM-based agent systems. Its high-level control logic, operating principles, and accumulated knowledge are expressed in natural language, while an LLM integrates them with executable tools and code to realize precise and reproducible execution. All code and experimental data are publicly available at https://github.com/qiliuchn/SimSkill-V1.