Searcharxiv⌕ Search

arXiv subjects

Earl T. Barr

Publications and source records attributed to Earl T. Barr.

At least 19 recordsLinked to original sources

TerminalWorld: Benchmarking Agents on Real-World Terminal Tasks

We introduce TerminalWorld, a scalable data engine that automatically reverse-engineers high-fidelity evaluation tasks from "in-the-wild" terminal recordings. Processing 80,870 terminal recordings, the engine yields a full benchmark of 1,530 validated tasks, spanning 18 real-world categories, ranging from short everyday operations to workflows exceeding 50 steps, and covering 1,280 unique commands. From these, we curate a Verified subset of 200 diverse, manually reviewed tasks. Comprehensive benchmarking on TerminalWorld-Verified across eight frontier models and six agents reveals that current systems still struggle with authentic terminal workflows, achieving a maximum pass rate of only 62.5%. Moreover, TerminalWorld captures real-world terminal capabilities distinct from existing expert-curated benchmarks (e.g., Terminal-Bench), with only a weak correlation to their scores (Pearson r=0.20). The automated engine makes TerminalWorld authentic and scalable by construction, enabling it to evaluate agents in real-world terminal environments as developer practices evolve.

cs.AI↗

Cadence: Strategic Guidance for Coding Agents

Runtime monitors are increasingly used to improve the reliability of LLM-based coding agents by inspecting execution trajectories and delivering corrective guidance upon detecting misbehavior. However, their effectiveness remains limited by static guidance triggering schemes. Existing monitors rely either on fixed inspection intervals, missing timely guidance during severe misbehaviors while incurring unnecessary overhead during healthy execution, or on rigid heuristic rules, failing to detect complex reasoning errors. To address these limitations, we propose Cadence, a dynamic monitoring framework that adaptively schedules inspections and delivers guidance according to the agent's real-time execution health. Cadence consists of two core modules: a two-tier intervention module and an inspection scheduler. The intervention module delivers advisory-level guidance for normal executions and minor lapses, while providing replacement-level guidance for severe misbehaviors. Driven by the intervention level, the scheduler adjusts the inspection frequency by tightening supervision after replacement-level guidance and relaxing it after advisory-level guidance. Evaluated on 300 SWE-bench Lite tasks across two distinct agents, mini-swe-agent and Moatless, Cadence achieves the highest resolve rate among all evaluated monitors. Specifically, Cadence outperforms vanilla agents by 25.33\% (+76 resolved tasks) on mini-swe-agent and 15.67\% (+47 resolved tasks) on Moatless, while maintaining competitive token efficiency compared to state-of-the-art baselines.

cs.SE↗

Back to the Future: Regressing Readability Features from LLMs

Code readability supports software understanding, review, and maintenance. As AI coding agents become more widely used, readable code helps both humans review agent-generated output and agents maintain code within limited context windows. Yet existing readability models do not consistently agree with human judgements across datasets. We introduce BTTF (Back To The Future), pairing linear regression with information-theoretic features. Rather than using language models as black-box judges, BTTF uses them as measurement instruments. It combines traditional code measurements, embedding-based semantic organisation, and causal-LM predictability to capture source structure, semantic coherence, and contextual surprise. We evaluate BTTF on five publicly available human-rated readability benchmarks and our manually annotated MBJP development set, totalling 1,100 samples. BTTF improves average Spearman correlation over the strongest baseline by up to 0.101. Across 13 controlled readability-reducing transformations on production code, it achieves response rates of 93.8% in Java and 87.6% in Python, outperforming Dorn by 14% and 19.2%. It performs best among local models and trails the strongest cloud-LLM response by 1.7% in Java and 5.1% in Python. The direct-LLM baselines require an average of 1,083 prompt tokens per evaluation while remaining opaque. BTTF combines state-of-the-art benchmark performance and strong transformation robustness with feature-granular interpretability: every feature coefficient matches its theoretical directional effect across all evaluated model configurations.

cs.SE↗

OpenRUA: Robot-Use Agents Are Zero-Shot Visuomotor Policies

Coding agents are extending their reach into the physical world by writing and executing robot control programs. One might expect the agents to use the existing mature software stack that engineers have developed over decades to access sensors and control motion. Yet prior work primarily engineers complex custom harnesses to orchestrate agents for robot use, particularly by prescribing specialized workflows and providing bespoke interfaces. This raises the question: "Is such additional harness engineering necessary?" We introduce OpenRUA, a zero-abstraction harness that bypasses bespoke abstraction layers by providing off-the-shelf coding agents with only terminal access to the robot's native software interface ROS 2. OpenRUA employs a minimalist workspace-as-harness design, only offering ROS 2 documentation and basic tools while leaving the coding agent to organize its own work without orchestrating any agentic workflow. Within this workspace, OpenRUA recasts perception as file I/O and manipulation as coding. With Claude Code powered by Claude Opus 5, OpenRUA achieves success rates of 99.0% on CaP-Bench and 87.0% on LIBERO-PRO, demonstrating that an off-the-shelf coding agent can serve as a zero-shot visuomotor policy through the robot's native interface, without bespoke primitives or task-specific training. Under this minimalist design, further analysis reveals striking emergent behaviors of coding agents: (1) For perception, the agent spontaneously writes programs that process raw sensory inputs and derive metric measurements in 96.80% of episodes. (2) For manipulation, the agent spontaneously builds motion-control clients (e.g., gripper control) in 95.87% of episodes and closed-loop control programs (e.g., adjusting motion based on sensor feedback) in 50.13% of episodes. Our code is available at https://github.com/terminalworld/OpenRUA.

cs.RO↗

Reusing Battle-tested Code: A Systematic Literature Review of Automated Software Transplantation

Software reuse remains a fundamental challenge in software engineering, often caught between the security risks of external libraries and the reliability concerns of generative AI. Automated software transplantation offers a compelling alternative by enabling the direct extraction and integration, into new host systems, of battle-tested logic (code that has already survived the rigors of production, possibly over many years of extensive in-production use). This paper presents a systematic literature review to evaluate the current state-of-the-art and identify the remaining barriers preventing this technology from becoming a mainstream engineering practice. We identified 27 distinct automated software transplantation approaches. Our analysis reveals that while the adaptation of code is highly automated, significant bottlenecks remain in the manual prerequisite steps of organ location and test suite augmentation. We conclude with a research roadmap that suggests how these manual obstacles might be overcome, including the strategic use of Large Language Models to assist in identifying and verifying high-value, production-hardened logic for transplantation.

cs.SE↗

A-JIT: Agentic Just-In-Time Software Construction

Traditional software delivery assumes a static paradigm: code is constructed prior to execution and deployed as a fixed artifact. We present Agentic Just-In-Time Software Construction (A-JIT), a paradigm that replaces static binaries with dynamic, software systems that can perpetually evolve to meet changing demands. In A-JIT, an application is an integrated assembly comprising code, a runtime harness, and an embedded AI agent that continuously observes system usage and live execution traces. Much like a traditional JIT compiler specializes machine code to runtime execution paths, A-JIT specializes software logic, workflows, and tool interfaces to meet the specific needs of the end-user. By integrating synthesis directly into the ambient application lifecycle, A-JIT enables applications to dynamically construct missing implementations, generate new capabilities on the fly, and continuously adapt to end-user behavior. We demonstrate how this model supports trace-driven human-AI co-construction and opens a new design space for adaptive, self-evolving software.

cs.SE↗

Evaluating Inference-Time Defenses Against Package Hallucination in LLM-Generated Code

LLMs are increasingly used for code generation, yet they frequently hallucinate non-existent software packages, creating exploitable entry points into the software supply chain. We make four contributions to this problem. First, we show that prior evaluation methodologies systematically inflate hallucination rates by misclassifying standard-library modules as hallucinations in some languages. For Python, the overestimation reaches 9.4 percentage points. Second, we evaluate seven inference-time defenses for mitigating package hallucinations, including five guided decoding strategies (Greedy, Contrastive, DoLa, Nudging, and Active Layer-Contrastive Decoding), an iterative self-refinement approach (Self-Refine), and a Retrieval-Augmented Generation (RAG)-based defense.. Across eight models spanning five families and four programming languages (Python, JavaScript, Ruby, Rust), RAG reduces the package hallucination rate (PHR) in 18 of 32 model--language configurations. Third, we introduce Package Utility (PU) to assess whether defenses preserve valid and task-relevant recommendations. Among strategies evaluated, Greedy decoding provides the strongest average mitigation--utility trade-off. Fourth, we stress-test all strategies under adversarial prompts seeded with fabricated package names and find that PHR surges by up to 45 percentage points relative to standard prompts, with Ruby consistently the most vulnerable language (80.9--95.2\%). Under adversarial conditions, RAG and Self-Refine outperform all decoding-only strategies, indicating that robust defense requires either external grounding or iterative self-verification when prompts are actively hostile. Our results recast package hallucination as both a measurement problem and a decoding-time control problem, and they demonstrate that the choice of defense must be matched to the threat model and recommendation utility.

cs.SE↗

Failure as a Process: An Anatomy of CLI Coding Agent Trajectories

Large language model (LLM) coding agents are increasingly deployed to autonomously perform software engineering tasks in terminal-based environments, making their reliability a growing concern. Existing empirical studies investigate why coding agents fail, yet they largely treat failure as a final outcome rather than a temporal process, providing limited insight into how failures emerge, evolve, and become unrecoverable. We present the first large-scale empirical study of CLI coding-agent failure trajectories, introducing a process-oriented framework that analyzes failure through its onset, evolution, and recovery across execution trajectories. We first collect 3,843 execution trajectories generated by seven frontier models across three coding-agent scaffolds (OpenHands, MiniSWE, and Terminus2) on Terminal-Bench, then carefully filter them to obtain 1,794 complete and valid trajectories for manual annotation (over 63,000 execution steps), from which we derive 14 findings spanning failure occurrence, root causes, recovery, and cross-system consistency. Our findings show that coding-agent failures are predominantly driven by epistemic errors, typically begin within the first few execution steps, and often remain hidden until recovery is no longer possible, suggesting that improving coding-agent reliability requires earlier validation and intervention rather than relying solely on final-outcome evaluation.

cs.SE↗

Independent Test Generation for RESTful APIs

Modern REST API testing relies on brittle sequences of calls to build system state. These multi-step tests suffer from non-determinism, poor scalability, and a "reachability tax" where a single failed setup step invalidates the entire test. We introduce TECTON, which breaks this cycle by replacing implicit state construction with explicit state synthesis of both the request payload and mock data it depends on. TECTON achieves this through two complementary mechanisms: it generates diverse, valid payloads directly, and it augments existing test mocks with realistic data so those payloads have valid system state to reference. Both mechanisms apply combinatorial testing to a new domain: the nested property space of Abstract Data Types (ADTs). TECTON decomposes complex API requests into primitive components to unleash LLMs on the more tractable subtasks of identifying equivalence classes of these primitives and generating representative values for them. It then uses LLMs to extract and inject state values via test mocks, enabling payloads to reference valid state. It recomposes these values into covering combinations to directly produce high-coverage test payloads. On standard RESTful benchmarks, TECTON achieves 70% average line coverage - a 20% absolute increase over sequence-based generators. It exposes 2x more runtime errors than any prior tool, including assertions and data constraint failures. TECTON's shift from sequencing API calls to synthesized payloads advances the state of the art in automated API validation.

cs.SE↗

Hallucination Inspector: A Fact-Checking Judge for API Migration

Large Language Models (LLMs) are increasingly deployed in automated software engineering for tasks such as API migration. While LLMs are able to identify migration patterns, they often make mistakes and fail to produce correct glue code to invoke the new API in place of the old one. We call this issue Scaffolding Hallucination, a failure mode where models generate incorrect calling contexts by inventing Phantom Symbols -- such as imaginary imports, constructors, and constants -- that do not exist in the API specification. In this paper, we show that standard metrics cannot be relied upon to detect these instances of hallucination. We propose Hallucination Inspector, a static analysis tool to detect Scaffolding Hallucination in LLM-generated code. Our approach includes a lightweight evaluation framework that verifies symbols extracted from the abstract syntax tree against a knowledge base derived directly from software documentation for the API. A preliminary evaluation on Android API migrations demonstrates that our approach successfully identifies hallucinations and significantly reduces false positives compared to standard metrics and probabilistic judges

cs.SE↗

Learned or Memorized ? Quantifying Memorization Advantage in Code LLMs

The lack of transparency about code datasets used to train large language models (LLMs) makes it difficult to detect, evaluate, and mitigate data leakage. We present a perturbation-based method to quantify memorization advantage in code LLMs, defined as the performance gap between likely seen and unseen inputs. We evaluate 8 open-source code LLMs on 19 benchmarks across four task families: code generation, code understanding, vulnerability detection, and bug fixing. Sensitivity patterns vary widely across models and tasks. For example, StarCoder reaches high sensitivity on some benchmarks (up to 0.8), while QwenCoder remains lower (mostly below 0.4), suggesting differences in generalization behavior. Task categories also differ: code summarization tends to show low sensitivity, whereas test generation is substantially higher. We then analyze two widely discussed benchmarks, CVEFixes and Defects4J, often suspected of leakage. Contrary to common concerns, both show low memorization advantage across models: CVEFixes remains below 0.1, and Defects4J is lower than other program repair benchmarks. These results suggest that, for these datasets, models may rely more on learned generalization than direct memorization. Overall, our findings provide evidence that memorization risk is highly task- and model-dependent, and highlight the need for stronger evaluation protocols, especially in security-focused settings.

cs.SE↗

Compressing Code Context for LLM-based Issue Resolution

Large Language Models (LLMs) are now capable of resolving real-world GitHub issues. However, current approaches overapproximate the code context and suffer from two compounding problems: the prohibitive cost of processing massive inputs, and low effectiveness as noise floods the context window and distracts the model from the bug-fixing signal. Existing compression techniques fail to resolve this tension: generic compressors compromise the semantic integrity of code, while code-specific tools lack awareness of code structure and task context to preserve essential patch ingredients. To address this, we propose a novel framework consisting of two components. First, Oracle-guided Code Distillation (OCD), a context distillation algorithm that combines genetic search and delta debugging to systematically reduce code contexts to their minimal sufficient subsequence - retaining only the ingredients required for a successful fix. We use this distilled data to fine-tune SWEzze, a lightweight model that learns to compress code context at inference time, filtering noise and combating distraction while preserving fix ingredients. Evaluated on SWE-bench Verified across three frontier LLMs, SWEzze maintains a stable compression rate of about 6 times across models, reduces the total token budget by 51.8%-71.3% relative to the uncompressed setting, improves issue resolution rates by 5.0%-9.2%, and delivers the best overall balance among effectiveness, compression ratio, and latency compared with state-of-the-art context compression baselines.

cs.SE↗

How Robustly do LLMs Understand Execution Semantics?

LLMs demonstrate remarkable reasoning capabilities, yet whether they utilize internal world models or rely on sophisticated pattern matching remains open. We study LLMs through the lens of robustness of their code understanding using a standard program-output prediction task. Our results reveal a stark divergence in model behavior: while open-source reasoning models (DeepSeek-R1 family) maintain stable, albeit somewhat lower accuracies (38% to 67%) under code transformations & input perturbations, the frontier model GPT-5.2 exhibits significant brittleness. Despite achieving a near-perfect score of 99% on the original, unperturbed CRUXEval benchmark, perturbed inputs trigger accuracy declines between 20% and 24%. In addition, we find that many models perform much worse at predicting behavior on perturbed inputs that raise exceptions, and that prediction performance depends on the kind of exception. We study remedies to address this deficiency in exception prediction, and evaluate the effect of these remedies on the ability to predict non-exception behaviors. Our findings both point to limitations in the way all models understand code, and establish the value of using perturbation to evaluate code models.

cs.SE↗

ContextBench: A Benchmark for Context Retrieval in Coding Agents

LLM-based coding agents have shown strong performance on automated issue resolution benchmarks, yet existing evaluations largely focus on final task success, providing limited insight into how agents retrieve and use code context during problem solving. We introduce ContextBench, a process-oriented evaluation of context retrieval in coding agents. ContextBench consists of 1,136 issue-resolution tasks from 66 repositories across eight programming languages, each augmented with human-annotated gold contexts. We further implement an automated evaluation framework that tracks agent trajectories and measures context recall, precision, and efficiency throughout issue resolution. Using ContextBench, we evaluate four frontier LLMs and five coding agents. Our results show that sophisticated agent scaffolding yields only marginal gains in context retrieval ("The Bitter Lesson" of coding agents), LLMs consistently favor recall over precision, and substantial gaps exist between explored and utilized context. ContextBench augments existing end-to-end benchmarks with intermediate gold-context metrics that unbox the issue-resolution process. These contexts offer valuable intermediate signals for guiding LLM reasoning in software tasks.

cs.LG↗

Forecasting Developer Environments with GenAI: A Research Perspective

Generative Artificial Intelligence (GenAI) models are achieving remarkable performance in various tasks, including code generation, testing, code review, and program repair. The ability to increase the level of abstraction away from writing code has the potential to change the Human-AI interaction within the integrated development environment (IDE). To explore the impact of GenAI on IDEs, 33 experts from the Software Engineering, Artificial Intelligence, and Human-Computer Interaction domains gathered to discuss challenges and opportunities at Shonan Meeting 222, a four-day intensive research meeting. Four themes emerged as areas of interest for researchers and practitioners.

cs.SE↗

Excision Score: Evaluating Edits with Surgical Precision

Many tasks revolve around editing a document, whether code or text. We formulate the revision similarity problem to unify a wide range of machine learning evaluation problems whose goal is to assess a revision to an existing document. We observe that revisions usually change only a small portion of an existing document, so the existing document and its immediate revisions share a majority of their content. We formulate five adequacy criteria for revision similarity measures, designed to align them with human judgement. We show that popular pairwise measures, like BLEU, fail to meet these criteria, because their scores are dominated by the shared content. They report high similarity between two revisions when humans would assess them as quite different. This is a fundamental flaw we address. We propose a novel static measure, Excision Score (ES), which computes longest common subsequence (LCS) to remove content shared by an existing document with the ground truth and predicted revisions, before comparing only the remaining divergent regions. This is analogous to a surgeon creating a sterile field to focus on the work area. We use approximation to speed the standard cubic LCS computation to quadratic. In code-editing evaluation, where static measures are often used as a cheap proxy for passing tests, we demonstrate that ES surpasses existing measures. When aligned with test execution on HumanEvalFix, ES improves over its nearest competitor, SARI, by 12% Pearson correlation and by >21% over standard measures like BLEU. The key criterion is invariance to shared context; when we perturb HumanEvalFix with increased shared context, ES' improvement over SARI increases to 20% and >30% over standard measures. ES also handles other corner cases that other measures do not, such as correctly aligning moved code blocks, and appropriately rewarding matching insertions or deletions.

cs.LG↗

Detecting Semantic Clones of Unseen Functionality

Semantic code clone detection is the task of detecting whether two snippets of code implement the same functionality (e.g., Sort Array). Recently, many neural models achieved near-perfect performance on this task. These models seek to make inferences based on their training data. Consequently, they better detect clones similar to those they have seen during training and may struggle to detect those they have not. Developers seeking clones are, of course, interested in both types of clones. We confirm this claim through a literature review, identifying three practical clone detection tasks in which the model's goal is to detect clones of a functionality even if it was trained on clones of different functionalities. In light of this finding, we re-evaluate six state-of-the-art models, including both task-specific models and generative LLMs, on the task of detecting clones of unseen functionality. Our experiments reveal a drop in F1 of up to 48% (average 31%) for task-specific models. LLMs perform on par with task-specific models without explicit training for clone detection, but generalize better to unseen functionalities, where F1 drops up to 5% (average 3%) instead. We propose and evaluate the use of contrastive learning to improve the performance of existing models on clones of unseen functionality. We draw inspiration from the computer vision and natural language processing fields where contrastive learning excels at measuring similarity between two objects, even if they come from classes unseen during training. We replace the final classifier of the task-specific models with a contrastive classifier, while for the generative LLMs we propose contrastive in-context learning, guiding the LLMs to focus on the differences between clones and non-clones. The F1 on clones of unseen functionality is improved by up to 26% (average 9%) for task-specific models and up to 5% (average 3%) for LLMs.

cs.SE↗

Memorization or Interpolation ? Detecting LLM Memorization through Input Perturbation Analysis

While Large Language Models (LLMs) achieve remarkable performance through training on massive datasets, they can exhibit concerning behaviors such as verbatim reproduction of training data rather than true generalization. This memorization phenomenon raises significant concerns about data privacy, intellectual property rights, and the reliability of model evaluations. This paper introduces PEARL, a novel approach for detecting memorization in LLMs. PEARL assesses how sensitive an LLM's performance is to input perturbations, enabling memorization detection without requiring access to the model's internals. We investigate how input perturbations affect the consistency of outputs, enabling us to distinguish between true generalization and memorization. Our findings, following extensive experiments on the Pythia open model, provide a robust framework for identifying when the model simply regurgitates learned information. Applied on the GPT 4o models, the PEARL framework not only identified cases of memorization of classic texts from the Bible or common code from HumanEval but also demonstrated that it can provide supporting evidence that some data, such as from the New York Times news articles, were likely part of the training data of a given model.

cs.CL↗