Daily Papers

AI agents have the potential to significantly alter the cybersecurity landscape. Here, we introduce the first framework to capture offensive and defensive cyber-capabilities in evolving real-world systems. Instantiating this framework with BountyBench, we set up 25 systems with complex, real-world codebases. To capture the vulnerability lifecycle, we define three task types: Detect (detecting a new vulnerability), Exploit (exploiting a given vulnerability), and Patch (patching a given vulnerability). For Detect, we construct a new success indicator, which is general across vulnerability types and provides localized evaluation. We manually set up the environment for each system, including installing packages, setting up server(s), and hydrating database(s). We add 40 bug bounties, which are vulnerabilities with monetary awards from \10 to 30,485, covering 9 of the OWASP Top 10 Risks. To modulate task difficulty, we devise a new strategy based on information to guide detection, interpolating from identifying a zero day to exploiting a given vulnerability. We evaluate 10 agents: Claude Code, OpenAI Codex CLI with o3-high and o4-mini, and custom agents with o3-high, GPT-4.1, Gemini 2.5 Pro Preview, Claude 3.7 Sonnet Thinking, Qwen3 235B A22B, Llama 4 Maverick, and DeepSeek-R1. Given up to three attempts, the top-performing agents are Codex CLI: o3-high (12.5% on Detect, mapping to \3,720; 90% on Patch, mapping to 14,152), Custom Agent: Claude 3.7 Sonnet Thinking (67.5% on Exploit), and Codex CLI: o4-mini (90% on Patch, mapping to \$14,422). Codex CLI: o3-high, Codex CLI: o4-mini, and Claude Code are more capable at defense, achieving higher Patch scores of 90%, 90%, and 87.5%, compared to Exploit scores of 47.5%, 32.5%, and 57.5% respectively; while the custom agents are relatively balanced between offense and defense, achieving Exploit scores of 17.5-67.5% and Patch scores of 25-60%.

Static coding benchmarks are fixed at release. Over time their tasks leak into training data, and stronger agents eventually saturate them. We argue that a benchmark should instead aim to be self-scaling: its tasks and oracles are created at evaluation time, so no content exists in advance to leak, and its difficulty is set by the participating agents rather than fixed at release. We introduce SWE-Duel, an arena in which two coding agents evaluate each other over mirrored Red-Blue turns. In each turn, the Red agent works on a real repository across three context-isolated sessions. It proposes and implements a new feature with feature tests, then embeds a subtle bug in that feature along with sealed bug-triggering tests. The resulting pull request is admitted only if (1) it passes the feature tests and the repository's existing tests while failing exactly the bug tests, and (2) a fresh session of the same agent can repair the bug blind, which rules out unsolvable tasks. The Blue agent then reviews the pull request without being told a bug exists. It wins only if its revision passes all three test suites, and then the two agents swap roles. We run a tournament among 11 model-harness participants on 15 repositories in five languages for 1,500 turns. The leaders win in different ways. GPT-5.5 with Codex CLI ranks first as the strongest solver, repairing 98.3% of the defects it reviews. Claude Opus 4.8 with mini-swe-agent ranks second as the strongest task author, with a 27.3% Red win rate. Finally, within one model family, tasks written by the newest version defeat a fixed-reference solver far more often than those of older versions (42.9% vs. 4.0%), giving preliminary evidence that task difficulty scales with task authoring capability.

How well can general-purpose multimodal models turn visual understanding and reasoning into embodied manipulation via executable code? We introduce CodeActionBench, a benchmark of 25 manipulation tasks that evaluates this capability through agentic Code-as-Policy. Without task-specific fine-tuning, demonstrations, external specialist perception or grasp modules, privileged scene state, or predefined task policies, agents should select visual evidence, form task-relevant 3D estimates, construct manipulation targets, and iteratively execute and revise their policies. A shared robot API provides RGB observations, calibrated geometric operations, robot feedback, and bounded motion, leaving task-dependent decisions to the evaluated agent. Fixed task instances, resource budgets, and a hidden physical-outcome verifier support controlled comparisons across models and harness configurations. Extensive evaluations across nine configurations and 675 attempts achieve success rates ranging from 2.7% to 73.3%. The strongest configuration, GPT-6 Astra with Codex CLI, solves 22 of 25 tasks at least once in three attempts, demonstrating the best performance while still leaving substantial room for improvement. Trajectory analyses reveal difficulties in spatial alignment, object retention, and completion judgment, including task failures despite successfully completed motions. CodeActionBench provides a controlled testbed for measuring how general-purpose models translate their capabilities into manipulation behavior and for examining typical failure scenarios in that process.

LLM agents combine models with execution harnesses, so agentic benchmark performance reflects their joint behavior. We argue that harnesses should therefore be evaluated through their interactions with underlying models. We develop a framework combining five complementary measures: (i) Valid Completion Rate (VCR), how reliably agents finish normally and deliver evaluable results, irrespective of task correctness; (ii) Model-Harness Fit, how well a harness supports task achievement through valid completion for a given model; (iii) Cross-Model Fit, a harness's overall suitability across evaluated models; (iv) Interaction Residuals, pairing-specific execution strengths and weaknesses hidden by component-level summaries; and (v) Token Load, the token demands of agent execution, including failed attempts. We evaluate four general-purpose harnesses (OpenCode, Codex CLI, Pi CLI, and DeepAgents) with four core models, plus two fixed native model-harness systems, on six benchmarks covering context management, tool use, long-context planning, multimodal interaction, and artifact handoff. Including native references and stress conditions, 111 evaluation settings yield 52,168 trajectories. Harness rankings vary across models and benchmarks; task performance and valid completion can favor different pairings, and similar task scores can accompany distinct execution profiles. OpenCode ranks first in cross-model performance, while DeepSeek V4 Flash Preview maintains consistently high Mean VCR across harnesses on the shared four-benchmark panel. These findings support evaluating model-harness interactions, training transferable harness-use capabilities in models, and harness designs adapted to the model and task.

As production code evolves, the test suite must co-evolve to remain effective. Existing benchmarks for test evolution operate at method-level granularity with pre-paired inputs, bypassing the task of locating affected tests from the full project and excluding the need for new tests entirely. We present TEBench, the first project-level benchmark for test evolution. Given a project repository and a code-changing commit, TEBench requires systems to autonomously identify tests requiring modification, determine where new tests are needed, and produce the corresponding test patch. We construct TEBench through a four-stage pipeline over Defects4J projects, curating 314 task instances from 10 projects with developer-written ground truth. Each instance is annotated with one or more of three evolution types: Test-Breaking (tests that fail), Test-Stale (tests that pass but no longer meaningfully validate updated behavior), and Test-Missing (new tests needed for introduced behavior). We evaluate seven configurations spanning three industrial agent frameworks (Claude Code, Codex CLI, OpenCode) and six base models, alongside a heuristic baseline. All seven configurations converge on an identification F1 of 45.7% to 49.4%, revealing a shared performance ceiling across both frameworks and base models. Test-Stale is the most challenging type, averaging F1 around 36%, since configurations rely on execution failure signals and lack proactive semantic reasoning. On the update task, configurations produce highly executable test modifications whose surface form diverges substantially from ground truth. Trajectory analysis reveals a reactive "execute-fail-fix" loop that succeeds for breaking tests but structurally cannot address stale or missing tests. TEBench is available at https://github.com/iSEngLab/TEBench with a leaderboard at https://tebench-leadership.vercel.app.

Coding agents often finish a session by returning one squashed patch that mixes feature implementation, bug fixes, refactorings, tests, and configuration edits. While the final code may be correct, collapsing unrelated edits into one patch removes the history structure needed for review, selective revert, and later maintenance. We study retrospective commit-history reconstruction: given a completed squashed change, an agent groups its hunks into commits and materializes a replayable commit sequence. We formalize the task as hunk-to-commit partitioning with a replay requirement and build AtomicCommitBench, containing 800 real consecutive-commit episodes from 10 Python projects. Because multiple decompositions may be reasonable, we evaluate outputs using complementary metrics: PPAR for replay validity, ARI for reference-based grouping quality, and TCR for failure containment on scoreable modified-test episodes. Natural retrospective reconstruction proves substantially harder than replay checking or synthetic tangling. Although nearly all non-random methods achieve replay validity (PPAR >= 0.988), grouping quality ranges from 0.03 to 0.46 ARI. Matched synthetic composites are much easier than real same-author squashed diffs (+0.333 ARI). In our evaluation, the GPT-5.4 setup driven by Codex CLI (0.46 ARI) and the GLM-5 setup driven by Claude Code (0.43 ARI) outperform MiniMax (0.31) and Kimi (0.29). Qualitative analysis identifies same-file lumping and support-hunk drift as recurring failure modes. Dependency-Aware Commit Evidence (DACE) improves the lower-scoring setups by 0.05 to 0.08 ARI, indicating that dependency cues and hunk-role information help agents avoid locality-driven grouping errors. AtomicCommitBench enables evaluation of the commit histories produced by coding agents alongside the final code.

Existing tool-use benchmarks for LLM agents are overwhelmingly linear: our analysis of six benchmarks shows 55 to 100% of instances are simple chains of 2 to 5 steps. We introduce The Amazing Agent Race (AAR), a benchmark featuring directed acyclic graph (DAG) puzzles (or "legs") with fork-merge tool chains. We release 1,400 instances across two variants: sequential (800 legs) and compositional (600 DAG legs). Agents must navigate Wikipedia, execute multi-step tool chains, and aggregate results into a verifiable answer. Legs are procedurally generated from Wikipedia seeds across four difficulty levels with live-API validation. Three complementary metrics (finish-line accuracy, pit-stop visit rate, and roadblock completion rate) separately diagnose navigation, tool-use, and arithmetic failures. Evaluating three agent frameworks on 1,400 legs, the best achieves only 37.2% accuracy. Navigation errors dominate (27 to 52% of trials) while tool-use errors remain below 17%, and agent architecture matters as much as model scale (Claude Code matches Codex CLI at 37% with 6x fewer tokens). The compositional structure of AAR reveals that agents fail not at calling tools but at navigating to the right pages, a blind spot invisible to linear benchmarks. The project page can be accessed at: https://minnesotanlp.github.io/the-amazing-agent-race

Harnesses are now central to coding-agent performance, mediating how models interact with tools and execution environments. Yet harness engineering remains a manual craft, because automating it faces a heterogeneous action space across editable components, voluminous trajectories that bury actionable signal, and edits whose effect is hard to attribute. We introduce Agentic Harness Engineering (AHE), a closed loop that addresses these challenges through three matched observability pillars: (1) component observability gives every editable harness component a file-level representation so the action space is explicit and revertible; (2) experience observability distills millions of raw trajectory tokens into a layered, drill-down evidence corpus that an evolving agent can actually consume; and (3) decision observability pairs every edit with a self-declared prediction, later verified against the next round's task-level outcomes. Together, these pillars turn every edit into a falsifiable contract, so harness evolution proceeds autonomously without collapsing into trial-and-error. Empirically, ten AHE iterations lift pass@1 on Terminal-Bench 2 from 69.7% to 77.0%, surpassing the human-designed harness Codex-CLI (71.9%) and the self-evolving baselines ACE and TF-GRPO. The frozen harness transfers without re-evolution: on SWE-bench-verified it tops aggregate success at 12% fewer tokens than the seed, and on Terminal-Bench 2 it yields +5.1 to +10.1pp cross-family gains across three alternate model families, indicating the evolved components encode general engineering experience rather than benchmark-specific tuning. Ablations localize the gain to tools, middleware, and long-term memory rather than the system prompt, suggesting factual harness structure transfers while prose-level strategy does not.

Large language models can generate plausible quantum code, but it is unclear whether they can reliably target the specific software development kit (SDK) version requested by the user. We study this problem as API drift and introduce quantum-api-drift, a benchmark for measuring version fidelity, defined here as execution success on the requested SDK version, cross-version compatibility, failure modes, and documentation-guided repair in LLM-generated quantum SDK code. We instantiate the benchmark with Qiskit, a representative quantum SDK that underwent substantial interface changes across v0.43, v1.3, and v2.0. We evaluate 17 models on 50 tasks with 3 samples per prompt, yielding 450 generated samples and 1,350 executions per model. Sixteen models are tested in a matched REST API setting with a 1024-token output cap, while GPT-5.4 (Codex CLI) is reported separately as a reference configuration. Across the 16 matched REST models, diagonal Pass@1 ranges from 0.02 to 0.85. Claude Opus 4.7 is strongest on v0.43 and v2.0, while Grok 4.20 is strongest on v1.3 at 0.513. Error profiles differ systematically by model strength: weaker models fail mainly with broken imports, while stronger models more often reach deprecation-level failures. Documentation-guided repair succeeds for 0.19 to 0.59 of repair attempts overall and is consistently much more effective for migration to v2.0 than to v1.3. The benchmark artifacts are publicly available at https://github.com/arasyi/quantum-api-drift. These results show that version alignment is a distinct evaluation axis for quantum code generation and that API drift remains only partly recoverable even with migration guidance.

Developers increasingly delegate real maintenance work to product-grade coding agents, and many state tasks in their native language, in the style of a customer request rather than a curated English issue. We introduce RuBench 1.0, a benchmark of 25 tasks mined from recent fix commits in five live open-source repositories (aiohttp, aiogram, Laravel, NestJS, Fastify), each specified natively in Russian -- written from scratch, not translated -- and judged by the upstream maintainer's regression tests, which we withhold from release. All fix commits postdate the training-data cutoffs of every evaluated model. Round 1 evaluates Claude Code with Opus 4.8, Sonnet 5, and Haiku 4.5, and Codex CLI with GPT-5.5 (3 independent runs each; pass@1 with task-level uncertainty); the best configuration resolves 78.7% of tasks. Auditing full trajectories of an hors-concours configuration (Claude Code + Fable 5), we caught the product silently substituting the model on 20% of tasks via an official safeguard fallback -- evidence that the deployed product, not the model, is the unit actually measured. Version 2 adds Round 2: seven further configurations on the same frozen set under a per-configuration freshness gate -- the Russian-market agents SourceCraft CLI (ds, legacy) and Koda CLI (koda-pro), Antigravity with Gemini 3.1 Pro and 3.5 Flash, and Codex CLI with GPT-5.6 Sol and Luna. SourceCraft's flagship resolves 68.1% (N=23), above GPT-5.5 and both Gemini rows. A tool-call contamination re-audit of all 437 Round-2 trajectories finds the Russian and Gemini columns clean (0/293 cells) while flagging systematic oracle-hunting in the GPT-5.6 family (8/69 and 13/75 cells), including one case of mining a prior round's artifacts from the run machine's disk; honest scores are published alongside raw ones. We release statements, metadata, trajectories, and diffs; oracles are withheld with a SHA-256 manifest.

Front-end development accumulates change after change at the repository level, weaving complex cross-file dependencies that current LLM coding agents tuned for single-shot tasks cannot reliably track across multiple iterations, leading to functional regressions and code that resists maintenance. We argue the missing piece is design knowledge: architectural principles, module responsibilities, and structural constraints that developers lean on to keep code readable, maintainable, and evolvable as a system scales. To operationalize this, we propose WebDesignIter, a framework built around a persistent knowledge graph (WebAppArchKG) that fuses repository structure with design knowledge and keeps both in sync across development cycles. WebDesignIter works in two stages: design-informed planning pulls historical context and architectural overviews from WebAppArchKG to produce an implementation plan with corresponding test scripts, and design-aware generation executes that plan through targeted diff-based patches, validated by sandbox execution and automatic syntax repair. On Web-Bench, WebDesignIter delivers an average Pass@2 gain of 9.55 percentage points across nine foundation models over existing baselines. More importantly, WebDesignIter outperforms every general-purpose coding agent Claude Code, OpenHands, SWE-Agent, Codex CLI on every model configuration, posting the highest Pass@1 and Pass@2 while consuming 2530 fewer input tokens. Ablation singles out design knowledge as the most impactful component: stripping it drops Pass@1 by 11.40 percentage points, a degradation far larger than removing code-graph retrieval, patch-based generation, or sandbox verification, confirming that design knowledge provides a fundamentally more efficient and reliable path to repository-level code generation.

Coding-agent evaluation has progressed from resolving individual issues to carrying out long-horizon development, yet task completion is still largely assessed within a single codebase. In software ecosystems, many features and bug fixes require coordinated changes across multiple repositories. We introduce WideSWE to evaluate coding agents on such cross-repository tasks. Mining and reviewing changes across 103 software ecosystems yields 120 real-world tasks, balanced between 60 bug fixes and 60 features. We derive prompts from related issues and pull requests. We systematically review and adapt hidden tests to support diverse correct implementations while preserving required behavior and regression checks. Across seven agent configurations, full task success ranges from 10.83% to 42.50%, with the configuration pairing Codex CLI with GPT-5.6-sol achieving the highest rate. Trajectories show agents failing to identify necessary changes, recognizing changes but leaving them unfinished, or modifying the required repositories without fully satisfying the request. To examine whether working on one repository at a time can alleviate these difficulties, we compare it with joint execution under identical prompts. Independent execution mainly recovers omitted work and is less effective at correcting previously attempted but unsuccessful implementations. Joint execution can use information from related repositories to guide implementation and verification. Code is available at https://github.com/ZJU-ACES-ISE/WideSWE.

Multi-agent coding systems are designed to tackle complex software engineering tasks through collaboration. However, existing evaluations typically assume configured organizations are followed faithfully, whereas reality differs. This behavioral gap, combined with differences in underlying system components, prevents clear attribution of observed gains. To this end, we introduce OpenCollab, a multi-agent coding framework that provides a unified infrastructure for programmable collaboration and controllable runtime. Specifically, OpenCollab unifies organization design, enforces experimental control on a shared runtime, and tracks execution through fine-grained event streams. On this basis, we define Adherence to quantify whether the declared organization is actually realized. Our experiments reveal that agents collaborate very differently across configurations: changing any single dimension shifts Adherence, from 47.2% to as high as 97.2%. Furthermore, extensive agentic coding benchmarks show that a two-coder workflow built on OpenCollab establishes new SOTA performance compared to the mainstream harnesses such as Mini-SWE-agent, Codex CLI, and Claude Code, showing that a well-designed organization can outperform strong existing harnesses, while OpenCollab's single-agent configuration uses the fewest tokens across all evaluated suites. OpenCollab establishes a unified multi-agent infrastructure for easy programmable collaboration and controlled causal evaluation.

Coding agent harnesses mediate tool use and authorize actions, yet their security mechanisms and runtime effects remain incompletely characterized. We present HarnessSecurity, the first systematic empirical study and benchmark of open- and closed-source coding agent harnesses. First, we derive a ten-mechanism taxonomy and then assess 400 harness-mechanism cells using independent ratings by researchers and large language model (LLM) judges. We find that about half of confirmed mechanism implementations are opt-in, while closed-source harnesses exhibit substantial evidence gaps. Second, we introduce HarnessSecurity-Bench, a benchmark of 23 tasks across five attack surfaces without sacrificing legitimate task requirements. Using separate deterministic oracles to measure task utility and attack effects with security setting comparisons, we evaluate nine mechanisms across six leading harnesses: Claude Code, Codex CLI, Gemini CLI, gptme, Qwen Code, and GitHub Copilot. Under a controlled LLM baseline GLM-5.2, we conduct 2,500 trials, recording 81,155 tool calls and over 2.2 billion tokens. Enabling auto-approve increases utility and raises attack success from 29.2% to 95.6%. Network isolation and read-only mode reduce attack effects with substantial utility losses, while command allowlisting and command denylisting reduce attack effects with a small utility loss and a utility gain, respectively. Task-level cases show that restrictions on a shared capability can obstruct both legitimate and malicious operations, and that allowed tools or commands can leave unauthorized operations reachable through alternative execution paths. Harness providers should make security settings verifiable, test alternative execution paths to protected operations, and assess attack effects alongside task utility and execution costs.

AI tools and agents are reshaping how researchers work, from proving theorems to training neural networks. Yet for many, it remains unclear how these tools fit into everyday research practice. This paper is a practical guide to AI-assisted research in mathematics and machine learning: We discuss how researchers can use modern AI systems productively, where these systems help most, and what kinds of guardrails are needed to use them responsibly. It is organized into three parts: (I) a five-level taxonomy of AI integration, (II) an open-source framework that, through a set of methodological rules formulated as agent prompts, turns CLI coding agents (e.g., Claude Code, Codex CLI, OpenCode) into autonomous research assistants, and (III) case studies from deep learning and mathematics. The framework runs inside a sandboxed container, works with any frontier LLM through existing CLI agents, is simple enough to install and use within minutes, and scales from personal-laptop prototyping to multi-node, multi-GPU experimentation across compute clusters. In practice, our longest autonomous session ran for over 20 hours, dispatching independent experiments across multiple nodes without human intervention. We stress that our framework is not intended to replace the researcher in the loop, but to augment them. Our code is publicly available at https://github.com/ZIB-IOL/The-Agentic-Researcher.

Modern coding agents expose multiple tool surfaces -- IDE primitives, bash, and Model Context Protocol (MCP) code-execution -- and the field has shipped three contradictory claims about which one matters. We run the missing crossed comparison: an integrity-clean three-arm ablation (baseline / bash_only / code_only) on synthetic computation tasks and SWE-bench Mini modification tasks, holding model, harness, and prompts fixed, with two agents (Claude Code, OpenAI Codex CLI) so the comparison spans both regime and agent-design axes. Across the four resulting (regime, agent) cells, restricting the agent to a single execute_code MCP tool is cheaper than -- or statistically tied with -- its cheapest tool-rich rival in three cells (significantly on Artifact/Claude and SWE-bench/Codex; directionally on Artifact/Codex), with pass rates statistically tied within each cell. The lone exception is SWE-bench/Claude, where code_only is directionally costlier (+14.4%, not significant); a conditional-cost analysis localizes that gap to failure-cost on doomed-run trajectories, not a per-edit tax on successful runs. Two implications: the cheapest tool surface is jointly determined by task regime and agent design rather than by either axis alone, and the headline cost signal lives in cache-adjusted cost -- not pass rate, which is invariant across surfaces at the model sizes we evaluate. The benchmark harness, task suite, and analysis code are available at https://github.com/hyang0129/onlycodes.

OpenAI's Codex, a GPT-3 like model trained on a large code corpus, has made headlines in and outside of academia. Given a short user-provided description, it is capable of synthesizing code snippets that are syntactically and semantically valid in most cases. In this work, we want to investigate whether Codex is able to localize and fix bugs, a task of central interest in the field of automated program repair. Our initial evaluation uses the multi-language QuixBugs benchmark (40 bugs in both Python and Java). We find that, despite not being trained for APR, Codex is surprisingly effective, and competitive with recent state of the art techniques. Our results also show that Codex is slightly more successful at repairing Python than Java.

Clinical research over electronic health records depends on derived variables whose definitions sit outside the schema, in item codes, time windows, missingness rules, and output-grain choices that change cohorts and downstream conclusions. We introduce M4Bench, 28 clinical derivation tasks across 15 task families on MIMIC-IV and eICU, to test whether AI coding agents can reproduce such derivations and how clinician-reviewed agent skills change their results. Its controls decompose the value of skills into specification completeness, delivery format, and schema familiarity. In a Docker-isolated Codex campaign (GPT 5.4 Mini and GPT 5.5), skills raise reward by 0.136 over the no-skill baseline, with 26 of 28 tasks improving. Specification completeness accounts for most of this gain: on an eight-task control subset, an operational spec alone raises reward by 0.162, and skills add only 0.024 on top, indistinguishable from zero. On a ten-task subset, decoy skills score like no skill and raw reference SQL nearly matches targeted skills, arguing against generic context or the skill format as the explanation. Schema obfuscation and restructuring preserve 86% and 70% of the skill gain (GPT 5.4 Mini, 26 MIMIC-IV tasks), a descriptive robustness check rather than a bound on memorization. A descriptive Claude Code replication on all 28 tasks (Opus 4.7 and Sonnet 4.6, 560 runs) is directionally consistent (+0.120, 21 of 28 tasks improving). These gains are partial agreement with the reference, not completed tasks: only 25 of 280 with-skill and 13 of 280 no-skill Codex runs pass every strict check, and skills barely change which rows agents return (key F1 +0.003). Skills are released as a versioned, clinician-reviewable markdown delivery format, with a family-level reference-validity audit. M4Bench releases tasks, skills, evaluation code, and sanitized per-run evidence as an open-source benchmark.

Agents often work on complex problems that require millions of tokens of context, which necessitates compacting across sessions due to limited context windows. We develop CliffCompaction, an autocompaction technique that reduces cost by up to 50% under a bounded context while maintaining or improving performance on Terminal-Bench and achieving new levels of efficiency for test-time scaling and state-of-the-art results on KernelBench. The per-rollout savings of CliffCompaction make the performance--cost trade-off of test-time scaling more efficient, adding over 10 percentage points on Terminal-Bench for less than the cost of two full-context runs. Under parallel test-time scaling, CliffCompaction lets Kimi K2.6 match Opus 4.7, and exceed Opus 4.6 and GPT-5.3 Codex at lower cost. The key to CliffCompaction's effectiveness is that it keeps compacted information faithful by only truncating or dropping content, never rephrasing or rewriting it. We never compact a compaction---each pass operates only on original content, and prior compacted output is discarded, preventing context drift from accumulating. These properties sustain continual learning over sessions exceeding a million tokens: on KernelBench, CliffCompaction reaches CUDA kernel speedups of 2.23times after 200 steps and 3.58times after 400 steps, surpassing specialized search algorithms and trained agents despite being a general-purpose compaction technique. We open-source a scaffold-agnostic API-proxy implementation of CliffCompaction usable with Claude Code, Codex and other harnesses.

Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agents on a leaderboard, Agent4Sci asks what kinds of progress agents can make on these tasks under in-the-wild conditions, and where they fall short. Empirically, today's frontier agents advance the published best on every benchmark, with Claude Opus reaching 10.4% mean relative improvement and no run flagged for academic misconduct. These gains, however, come almost entirely from a bag of ML-engineering tricks rather than research-level insight. Trajectory analysis surfaces three recurring limitations in this setting: novelty miscalibration, resource under-utilization, and rapid collapse into local engineering sweeps. Lightweight plug-in skills for research ideation, experiment management, and scheduled checklists raise Claude Opus from 10.4% to 20.1% and Codex from 4.4% to 8.1%, while yielding more diverse, domain-relevant interventions. Agent4Sci gives the community a way to track and push frontier agents from ML engineering toward domain insight on real-world ML-for-science problems. Code and data are available on GitHub (https://github.com/maszhongming/agent4sci) and Hugging Face (https://huggingface.co/datasets/MingZhong/agent4sci).

With the success of large language models (LLMs) of code and their use as code assistants (e.g. Codex used in GitHub Copilot), techniques for introducing domain-specific knowledge in the prompt design process become important. In this work, we propose a framework called Repo-Level Prompt Generator that learns to generate example-specific prompts using prompt proposals. The prompt proposals take context from the entire repository, thereby incorporating both the structure of the repository and the context from other relevant files (e.g. imports, parent class files). Our technique doesn't require any access to the weights of the LLM, making it applicable in cases where we only have black-box access to the LLM. We conduct experiments on the task of single-line code-autocompletion using code repositories taken from Google Code archives. We demonstrate that an oracle constructed from our prompt proposals gives a remarkably high relative improvement of 36% over Codex, showing the quality of these proposals. Further, we show that when we train a model to predict a prompt proposal, we can achieve significant performance gains over Codex and other baselines. We release our code, data, and trained checkpoints at: https://github.com/shrivastavadisha/repo_level_prompt_generation.

As AI agents become increasingly capable of complex, long-horizon reasoning, rigorous and holistic evaluation is essential for measuring progress toward real-world healthcare applications. We introduce HealthAgentBench, a suite of 54 agentic healthcare tasks across 7 categories each with its unique environment. The benchmark suite spans diverse workflows throughout the patient journey and a broad range of modalities. Each task is designed to replicate an end-to-end clinical workflow: given minimal instructions, an agent must explore raw healthcare data, operate within a complex environment, and execute multi-step solutions that go beyond naive prompting. A final task success rate is reported to provide a single, interpretable metric for HealthAgentBench overall performance for each agent. Evaluating frontier agents on HealthAgentBench, we find that overall task success rate remains low, underscoring the difficulty of the suite. The strongest and the most cost effective agent, Codex GPT-5.5, achieves only approximately 42% success rate. Beyond aggregate performance, HealthAgentBench reveals nuanced strengths and weaknesses across task categories. Frontier agents show promise in automatically developing research modeling pipelines over EHR data, but medical imaging remains especially challenging, particularly for Claude Code models, while Codex GPT-5.5 shows emerging capability. Tasks that combine large search spaces with compositional reasoning requirements remain difficult for all current agents. Together, these results suggest that HealthAgentBench provides a challenging and realistic benchmark with substantial room for future progress. We release our benchmark at https://github.com/microsoft/HealthAgentBench.

Large language models (LMs) of code have recently shown tremendous promise in completing code and synthesizing code from natural language descriptions. However, the current state-of-the-art code LMs (e.g., Codex (Chen et al., 2021)) are not publicly available, leaving many questions about their model and data design decisions. We aim to fill in some of these blanks through a systematic evaluation of the largest existing models: Codex, GPT-J, GPT-Neo, GPT-NeoX-20B, and CodeParrot, across various programming languages. Although Codex itself is not open-source, we find that existing open-source models do achieve close results in some programming languages, although targeted mainly for natural language modeling. We further identify an important missing piece in the form of a large open-source model trained exclusively on a multi-lingual corpus of code. We release a new model, PolyCoder, with 2.7B parameters based on the GPT-2 architecture, which was trained on 249GB of code across 12 programming languages on a single machine. In the C programming language, PolyCoder outperforms all models including Codex. Our trained models are open-source and publicly available at https://github.com/VHellendoorn/Code-LMs, which enables future research and application in this area.

We introduce Codex, a GPT language model fine-tuned on publicly available code from GitHub, and study its Python code-writing capabilities. A distinct production version of Codex powers GitHub Copilot. On HumanEval, a new evaluation set we release to measure functional correctness for synthesizing programs from docstrings, our model solves 28.8% of the problems, while GPT-3 solves 0% and GPT-J solves 11.4%. Furthermore, we find that repeated sampling from the model is a surprisingly effective strategy for producing working solutions to difficult prompts. Using this method, we solve 70.2% of our problems with 100 samples per problem. Careful investigation of our model reveals its limitations, including difficulty with docstrings describing long chains of operations and with binding operations to variables. Finally, we discuss the potential broader impacts of deploying powerful code generation technologies, covering safety, security, and economics.

LLM agents span command-line interfaces (e.g., Codex) and agent frameworks (e.g., LangChain), integrating backend LLMs with harness code that parses model outputs, controls agent loops, and manages context. Both the harness and LLM-generated responses jointly shape an agent's execution. This architecture gives rise to bugs that cannot be readily understood by inspecting either component alone, because some bugs occur only when a particular LLM response elicits an abnormal reaction from the agent. Prior empirical studies of agent bugs have largely attributed failures either to limited model capabilities or to harness-side defects, such as outdated APIs and configuration misalignment, without characterizing these AR bugs. We conduct the first empirical study focused on agent-reactive (AR) bugs. Through manual analysis of 255 bug reports from Codex, Gemini-CLI, LangChain, and CrewAI, we construct a two-axis taxonomy covering observable symptoms and the LLM behaviors that trigger them. Our findings show that many AR bugs manifest as silent errors without well-defined test oracles, which makes detection difficult. The stochasticity of LLM responses further complicates bug reproduction. We additionally examine fixes proposed by users and implemented by developers. This analysis exposes a mismatch: users frequently advocate harness-side guardrails, whereas developers may attribute the issue to the LLM or respond slowly to user-proposed fixes. These findings point to the need for mechanisms that help users and developers understand the root causes and resolutions of AR bugs. Overall, the study highlights challenges specific to LLM agents and motivates the design of test oracles, reproduction support, and fault-localization techniques for AR bugs.

Advances in frontier coding agents have sharply increased the token costs of AI-assisted software engineering. Their token efficiency is typically evaluated on isolated tasks, whereas real-world developers use them to continuously build, update, and maintain large-scale projects. This mismatch motivates us to evaluate agents over sequences of realistic coding tasks, where we find that agents exhibit recurring environment-dependent token costs that existing methods struggle to reduce. To minimize these costs, we introduce TokenEconomizer, an external controller that cuts environment-specific token costs over a sequence of tasks. These cuts come from determining (i) which operational micro-skills to inject, (ii) which persistently irrelevant tools to prune, and (iii) when to intervene in unproductive failure loops. Across Claude Code, Codex, and GitHub Copilot CLI paired with frontier models, TokenEconomizer reduces the average token cost by 18.4-30.9% with comparable task success. Savings hold in every repository under every tested agent-model setting and generalize to diverse domains such as sequential web development and security auditing, where costs fall by 6.4-21.2% depending on task structure. Our work establishes adaptive black-box cost control as a promising direction for reducing token costs of coding agents in realistic settings.

Training language models (LMs) remains a highly human-intensive process, even as frontier language model agents become increasingly capable at software engineering and other long-horizon tasks. A central challenge is that autonomous post-training is not just a coding problem: it requires the agent to repeatedly plan iterations, construct benchmark-aligned data, run stable training jobs, evaluate checkpoints, and preserve experiment state across many hours of interaction. We present AutoTrainess, a LM agent that exposes these operations as a repository of agent-computer interfaces for planning, data preparation, training, evaluation, and logging. Rather than leaving the agent to operate in a raw CLI environment with an underspecified action space, AutoTrainess externalizes prior human experience as explicit workflows, rules, and execution constraints that guide the agent toward effective and reliable training behavior. On PostTrainBench, AutoTrainess consistently outperforms CLI-only baselines, achieving 26.94 average score with GPT-5.4 (Codex) versus 23.21 for CLI-only. It also generalizes across models and harnesses, improving DeepSeek-V4-Flash (OpenCode) from 12.13 to 19.58.

Agentic data science (ADS) systems are rapidly improving their capability to autonomously analyze, fit, and interpret data, potentially moving towards a future where agents conduct the vast majority of data-science work. However, current ADS systems use statistical tools designed to be interpretable by humans, rather than interpretable by agents. To address this, we introduce Agentic-imodels, an agentic autoresearch loop that evolves data-science tools designed to be interpretable by agents. Specifically, it develops a library of scikit-learn-compatible regressors for tabular data that are optimized for both predictive performance and a novel LLM-based interpretability metric. The metric measures a suite of LLM-graded tests that probe whether a fitted model's string representation is "simulatable" by an LLM, i.e. whether the LLM can answer questions about the model's behavior by reading its string output alone. We find that the evolved models jointly improve predictive performance and agent-facing interpretability, generalizing to new datasets and new interpretability tests. Furthermore, these evolved models improve downstream end-to-end ADS, increasing performance for Copilot CLI, Claude Code, and Codex on the BLADE benchmark by up to 73%

Large language and vision-language models increasingly power agents that act on a user's behalf through command-line interface (CLI) harnesses. However, most agent benchmarks still rely on synthetic sandboxes, short-horizon tasks, mock-service APIs, and final-answer checks, leaving open whether agents can complete realistic long-horizon work in the runtimes where they are deployed. This work presents WildClawBench, a native-runtime benchmark of 60 human-authored, bilingual, multimodal tasks spanning six thematic categories. Each task averages roughly 8 minutes of wall-clock time and over 20 tool calls, and runs inside a reproducible Docker container hosting an actual CLI agent harness (OpenClaw, Claude Code, Codex, or Hermes Agent) with access to real tools rather than mock services. Grading is hybrid, combining deterministic rule-based checks, environment-state auditing of side effects, and an LLM/VLM judge for semantic verification. Across 19 frontier models, the best, Claude Opus 4.7, reaches only 62.2% overall under OpenClaw, while every other model stays below 60%, and switching harness alone shifts a single model by up to 18 points. These results show that long-horizon, native-runtime agent evaluation remains a far-from-resolved task for current frontier models. We release the tasks, code, and containerized tooling to support reproducible evaluation.

Software engineering tools increasingly rely on LLM based agents to localize files to change to resolve a software issue. Most AI agents explore repositories linearly, that is, visiting one directory or file per step. We postulate that this is a structural mismatch for changes that span several subsystems. We compare linear sequential exploration against non-linear, domain-scoped parallel agentic exploration. Using SWE Bench Pro as initial benchmark, we focus on ansible as an exemplar. We construct an approach for persistent-session evaluation of GitHub issues anchored at a single base commit. We compare our non-linear domain-agent file traversal system against a base LLM without direct repository access, a single agent Recursive Language Model (RLM) baseline with a persistent Python REPL and an external CLI baseline using Codex 5.5 High. Domain scoped parallel agent spawning with a small Haiku-class model achieves the highest micro F1 among Haiku class models by a large margin. Domain-agents is the second highest behind only the much larger Codex 5.5 High on our own expanded benchmark including over more recent PRs from 2025 and 2026. On the original, curated, 2020 SWE-bench Pro benchmark, a larger Sonnet plain LLM baseline attains higher micro F1 by predicting few files, leading to higher precision, but at significantly lower all gold recall. We also present three additional findings. First, documentation evolution is a latent dependency unresolved by any approach. Second, naive file system access can degrade localization driven by test-file over prediction. Lastly, forced multi-agent consultation does not measurably help and raises token cost substantially.

AI agents act on behalf of user prompts, consuming external data and taking actions based on the agent context. Prior research on AI agent security has primarily focused on indirect prompt injection (IPI). Its most well-studied category is instruction injection, where attacker-controlled untrusted data is interpreted as an instruction. In response, many mitigations have been proposed to prevent instruction injection attacks. In this paper, we introduce a new category of IPI, agent data injection attacks (ADI). ADI injects malicious data disguised as trusted data, such as security-critical metadata (e.g., resource identifiers or data origins) or agent context data (e.g., tool call and response formats). As a result, agents unknowingly execute unintended actions based on attacker-controlled data. ADI has similar attack impacts as instruction injection attacks, because it causes agents to misbehave and execute unintended actions. Despite the similar impact, ADI remains underexplored and easily bypasses existing IPI defenses. We found several critical vulnerabilities in real-world agents that allow an attacker to launch various attacks: arbitrary click attacks on web agents (Claude in Chrome, Antigravity, and Nanobrowser), and remote code execution and supply-chain attacks on coding agents (Claude Code, Codex, and Gemini CLI). We evaluate ADI vulnerabilities across off-the-shelf models and AI agents, and find that ADI is effective in both standalone LLMs and AI agent settings. ADI exposes a critical gap in agent security, signifying that current AI agents do not employ a fundamental security principle: current agents do not isolate trusted data from untrusted data.

Choosing an agent system means choosing both a language model and the harness through which it acts. We ask whether a strong model, harness, or pairing stays strong when the setting changes. We evaluate 66 configurations: four configurable harnesses (OpenHands, DeepSeek Harness, PI, and openJiuwen) paired with five models on TUA-Bench, ALE-CLI, and Terminal-Bench 4, plus the native Codex-GPT and Claude Code-Claude pairings. Model rankings reverse across harnesses. On Terminal-Bench 4, Claude leads GPT by 7.94 points in OpenHands but trails it by 30.16 points in PI. For four of the five models, the best harness changes from one benchmark to another, yet some pairings hold: openJiuwen gives Kimi its highest score on all three benchmarks, by 5.61 to 11.11 points. A model's own vendor harness is not reliably its best, and higher cost does not reliably buy a higher score. On Terminal-Bench 4, GPT scores higher under PI than under DSH at less than a quarter of the cost per task. Matched trajectories suggest why fit varies. Models start almost all repairs themselves, so much depends on whether the harness hands failures back in a form the model can use. GPT does best with PI's lean scaffold, while Kimi, which often issues malformed tool calls, does best in openJiuwen. We argue that the model, the harness, and the task should be evaluated together, and we release the harness adapters, evaluation code, and all 6,204 scored trajectories at https://github.com/liyix/finding-the-right-fit and https://huggingface.co/datasets/yixuanli97/finding-the-right-fit.

We introduce EvilGenie, a benchmark for reward hacking in programming settings. We source problems from LiveCodeBench and create an environment in which agents can easily reward hack, such as by hardcoding test cases or editing the testing files. We measure reward hacking in three ways: held out unit tests, LLM judges, and test file edit detection. We verify these methods against human review and each other. We find the LLM judge to be highly effective at detecting reward hacking in unambiguous cases, and observe only minimal improvement from the use of held out test cases. In addition to testing many models using Inspect's basic\_agent scaffold, we also measure reward hacking rates for three popular proprietary coding agents: OpenAI's Codex, Anthropic's Claude Code, and Google's Gemini CLI. We observe explicit reward hacking by both Codex and Claude Code, and misaligned behavior by all three agents. Our codebase can be found at https://github.com/JonathanGabor/evilgenie_inspect .

Codex, a large language model (LLM) trained on a variety of codebases, exceeds the previous state of the art in its capacity to synthesize and generate code. Although Codex provides a plethora of benefits, models that may generate code on such scale have significant limitations, alignment problems, the potential to be misused, and the possibility to increase the rate of progress in technical fields that may themselves have destabilizing impacts or have misuse potential. Yet such safety impacts are not yet known or remain to be explored. In this paper, we outline a hazard analysis framework constructed at OpenAI to uncover hazards or safety risks that the deployment of models like Codex may impose technically, socially, politically, and economically. The analysis is informed by a novel evaluation framework that determines the capacity of advanced code generation techniques against the complexity and expressivity of specification prompts, and their capability to understand and execute them relative to human ability.

Though end-to-end neural approaches have recently been dominating NLP tasks in both performance and ease-of-use, they lack interpretability and robustness. We propose Binder, a training-free neural-symbolic framework that maps the task input to a program, which (1) allows binding a unified API of language model (LM) functionalities to a programming language (e.g., SQL, Python) to extend its grammar coverage and thus tackle more diverse questions, (2) adopts an LM as both the program parser and the underlying model called by the API during execution, and (3) requires only a few in-context exemplar annotations. Specifically, we employ GPT-3 Codex as the LM. In the parsing stage, with only a few in-context exemplars, Codex is able to identify the part of the task input that cannot be answerable by the original programming language, correctly generate API calls to prompt Codex to solve the unanswerable part, and identify where to place the API calls while being compatible with the original grammar. In the execution stage, Codex can perform versatile functionalities (e.g., commonsense QA, information extraction) given proper prompts in the API calls. Binder achieves state-of-the-art results on WikiTableQuestions and TabFact datasets, with explicit output programs that benefit human debugging. Note that previous best systems are all finetuned on tens of thousands of task-specific samples, while Binder only uses dozens of annotations as in-context exemplars without any training. Our code is available at https://github.com/HKUNLP/Binder .

Agent Skills package procedural instructions and checks for use by general-purpose agents, but loading a skill does not guarantee that every requirement remains active throughout a long tool-using trajectory. We study this problem in a production-derived, white-box code-audit workflow. Holding the task and 24 artifact checks fixed, we vary the surrounding context and classify where failures first become visible: lost requirements, editing drift, failed checking, or non-agent evaluator/runtime failures. Codex with gpt-5.4-mini passes 8/10 runs in a 10,991-character clean context but only 3/10 in both a 299,140-character relevant context and an equal-length irrelevant context. This 50-percentage-point difference is large but remains trend-level under two-sided Fisher tests (p = 0.0698). Requirement coverage nevertheless stays above 92% in both long conditions, showing that a few omissions can invalidate an otherwise complete artifact. A second task passes all clean and long runs, so the evidence does not support a universal context-length threshold. A detailed external checklist passes 10/10 runs, compared with 5/10 for a generic self-check (p = 0.0325). Coding-agent scaffolds may help by selecting a smaller working set, but they do not eliminate failures. We do not introduce context rot or a new general monitoring method; we provide a bounded failure classification and empirical case study for white-box code auditing.

To extend the scope of coding queries to more realistic settings, we propose ODEX, the first Open-Domain EXecution-based natural language (NL) to Python code generation dataset. ODEX has 945 NL-Code pairs spanning 79 diverse libraries, along with 1,707 human-written test cases for execution. Our NL-Code pairs are harvested from StackOverflow forums to encourage natural and practical coding queries. Moreover, ODEX supports four natural languages as intents, in English, Spanish, Japanese, and Russian. ODEX unveils intriguing behavioral differences among top-performing code language models (LM). While CODEX achieves better overall results, CODEGEN improves effectively via scaling -- CODEGEN 6.1B performs comparably with CODEX 12B. Both models show substantial gaps between open and closed domains, but CODEGEN gaps tend to decrease with model size while CODEX gaps increase. We release ODEX to facilitate research into open-domain problems for the code generation community.

Large language models have demonstrated the ability to generate both natural language and programming language text. Such models open up the possibility of multi-language code generation: could code generation models generalize knowledge from one language to another? Although contemporary code generation models can generate semantically correct Python code, little is known about their abilities with other languages. We propose MultiPL-E, a system for translating unit test-driven code generation benchmarks to new languages. We create the first massively multilingual code generation benchmark by using MultiPL-E to translate two popular Python code generation benchmarks to 18 additional programming languages. We use MultiPL-E to extend the HumanEval benchmark and MBPP benchmark to 18 languages that encompass a range of programming paradigms and popularity. Using these new parallel benchmarks, we evaluate the multi-language performance of three state-of-the-art code generation models: Codex, CodeGen, and InCoder. We find that Codex matches or even exceeds its performance on Python for several other languages. The range of programming languages represented in MultiPL-E allow us to explore the impact of language frequency and language features on model performance. Finally, the MultiPL-E approach of compiling code generation benchmarks to new programming languages is both scalable and extensible, making it straightforward to evaluate new models, benchmarks, and languages.

Benchmark datasets have a significant impact on accelerating research in programming language tasks. In this paper, we introduce CodeXGLUE, a benchmark dataset to foster machine learning research for program understanding and generation. CodeXGLUE includes a collection of 10 tasks across 14 datasets and a platform for model evaluation and comparison. CodeXGLUE also features three baseline systems, including the BERT-style, GPT-style, and Encoder-Decoder models, to make it easy for researchers to use the platform. The availability of such data and baselines can help the development and validation of new methods that can be applied to various program understanding and generation problems.

Security vulnerability repair is a difficult task that is in dire need of automation. Two groups of techniques have shown promise: (1) large code language models (LLMs) that have been pre-trained on source code for tasks such as code completion, and (2) automated program repair (APR) techniques that use deep learning (DL) models to automatically fix software bugs. This paper is the first to study and compare Java vulnerability repair capabilities of LLMs and DL-based APR models. The contributions include that we (1) apply and evaluate five LLMs (Codex, CodeGen, CodeT5, PLBART and InCoder), four fine-tuned LLMs, and four DL-based APR techniques on two real-world Java vulnerability benchmarks (Vul4J and VJBench), (2) design code transformations to address the training and test data overlapping threat to Codex, (3) create a new Java vulnerability repair benchmark VJBench, and its transformed version VJBench-trans and (4) evaluate LLMs and APR techniques on the transformed vulnerabilities in VJBench-trans. Our findings include that (1) existing LLMs and APR models fix very few Java vulnerabilities. Codex fixes 10.2 (20.4%), the most number of vulnerabilities. (2) Fine-tuning with general APR data improves LLMs' vulnerability-fixing capabilities. (3) Our new VJBench reveals that LLMs and APR models fail to fix many Common Weakness Enumeration (CWE) types, such as CWE-325 Missing cryptographic step and CWE-444 HTTP request smuggling. (4) Codex still fixes 8.3 transformed vulnerabilities, outperforming all the other LLMs and APR models on transformed vulnerabilities. The results call for innovations to enhance automated Java vulnerability repair such as creating larger vulnerability repair training data, tuning LLMs with such data, and applying code simplification transformation to facilitate vulnerability repair.

Recent development of large language models (LLMs) for code like CodeX and CodeT5+ demonstrates tremendous promise in achieving code intelligence. Their ability of synthesizing code that completes a program for performing a pre-defined task has been intensively tested and verified on benchmark datasets including HumanEval and MBPP. Yet, evaluation of these LLMs from more perspectives (than just program synthesis) is also anticipated, considering their broad scope of applications in software engineering. In this paper, we explore the ability of LLMs for testing programs/code. By performing thorough analyses of recent LLMs for code in program testing, we show a series of intriguing properties of these models and demonstrate how program testing ability of LLMs can be improved. Following recent work which utilizes generated test cases to enhance program synthesis, we further leverage our findings in improving the quality of the synthesized programs and show +11.77% and +4.22% higher code pass rates on HumanEval+ comparing with the GPT-3.5-turbo baseline and the recent state-of-the-art, respectively.

Large Language Models (LLMs) excel in stand-alone code tasks like HumanEval and MBPP, but struggle with handling entire code repositories. This challenge has prompted research on enhancing LLM-codebase interaction at a repository scale. Current solutions rely on similarity-based retrieval or manual tools and APIs, each with notable drawbacks. Similarity-based retrieval often has low recall in complex tasks, while manual tools and APIs are typically task-specific and require expert knowledge, reducing their generalizability across diverse code tasks and real-world applications. To mitigate these limitations, we introduce \framework, a system that integrates LLM agents with graph database interfaces extracted from code repositories. By leveraging the structural properties of graph databases and the flexibility of the graph query language, \framework enables the LLM agent to construct and execute queries, allowing for precise, code structure-aware context retrieval and code navigation. We assess \framework using three benchmarks: CrossCodeEval, SWE-bench, and EvoCodeBench. Additionally, we develop five real-world coding applications. With a unified graph database schema, \framework demonstrates competitive performance and potential in both academic and real-world environments, showcasing its versatility and efficacy in software engineering. Our application demo: https://github.com/modelscope/modelscope-agent/tree/master/apps/codexgraph_agent.

Very large language models (LLMs), such as GPT-3 and Codex have achieved state-of-the-art performance on several natural-language tasks, and show great promise also for code. A particularly exciting aspect of LLMs is their knack for few-shot and zero-shot learning: they can learn to perform a task with very few examples. Few-shotting has particular synergies in software engineering, where there are a lot of phenomena (identifier names, APIs, terminology, coding patterns) that are known to be highly project-specific. However, project-specific data can be quite limited, especially early in the history of a project; thus the few-shot learning capacity of LLMs might be very relevant. In this paper, we investigate the use few-shot training with the very large GPT (Generative Pre-trained Transformer) Codex model, and find evidence suggesting that one can significantly surpass state-of-the-art models for code-summarization, leveraging project-specific training.

We introduce a new type of programming challenge called programming puzzles, as an objective and comprehensive evaluation of program synthesis, and release an open-source dataset of Python Programming Puzzles (P3). Each puzzle is defined by a short Python program f, and the goal is to find an input which makes f return True. The puzzles are objective in that each one is specified entirely by the source code of its verifier f, so evaluating f is all that is needed to test a candidate solution. They do not require an answer key or input/output examples, nor do they depend on natural language understanding. The dataset is comprehensive in that it spans problems of a range of difficulties and domains, ranging from trivial string manipulation problems, to classic programming puzzles (e.g., Tower of Hanoi), to interview/competitive-programming problems (e.g., dynamic programming), to longstanding open problems in algorithms and mathematics (e.g., factoring). We develop baseline enumerative program synthesis, GPT-3 and Codex solvers that are capable of solving puzzles -- even without access to any reference solutions -- by learning from their own past solutions. Codex performs best, solving up to 18% of 397 test problems with a single try and 80% of the problems with 1,000 tries per problem. In a small user study, we find a positive correlation between puzzle-solving performance and coding experience, and between the puzzle difficulty for humans and AI solvers. Therefore, further improvements on P3 could have a significant impact on many program synthesis areas.

Command-line interfaces (CLIs) provide structured textual environments for system administration. Explorations have been performed using pre-trained language models (PLMs) to simulate these environments for safe interaction in high-risk environments. However, their use has been constrained to frozen, large parameter models like GPT. For smaller architectures to reach a similar level of believability, a rich dataset of CLI interactions is required. Existing public datasets focus on mapping natural-language tasks to commands, omitting crucial execution data such as exit codes, outputs, and environmental side effects, limiting their usability for behavioral modeling. We introduce a Shell Input -Output Environment (ShIOEnv), which casts command construction as a Markov Decision Process whose state is the partially built sequence and whose actions append arguments. After each action, ShIOEnv executes the candidate and returns its exit status, output, and progress toward a minimal-length behavioral objective. Due to the intractable nature of the combinatorial argument state-action space, we derive a context-free grammar from man pages to mask invalid arguments from being emitted. We explore random and proximal-policy optimization (PPO)-optimized sampling of unrestricted and grammar-masked action spaces to produce four exploration strategies. We observed that grammar masking and PPO significantly improve sample efficiency to produce a higher quality dataset (maximizing the number of arguments while minimizing redundancies). Policy-generated datasets of shell input-output behavior pairs are used to fine-tune CodeT5, where we observe 85% improvements in BLEU-4 when constraining the action space to grammar productions with an additional 26% improvement when applying PPO. The ShIOEnv environment and curated command behavior datasets are released for use in future research.

Despite the success of text retrieval in many NLP tasks, code retrieval remains a largely underexplored area. Most text retrieval systems are tailored for natural language queries, often neglecting the specific challenges of retrieving code. This gap leaves existing models unable to effectively capture the diversity of programming languages and tasks across different domains, highlighting the need for more focused research in code retrieval. To address this, we introduce CodeXEmbed, a family of large-scale code embedding models ranging from 400M to 7B parameters. Our novel training pipeline unifies multiple programming languages and transforms various code-related tasks into a common retrieval framework, enhancing model generalizability and retrieval performance. Our 7B model sets a new state-of-the-art (SOTA) in code retrieval, outperforming the previous leading model, Voyage-Code, by over 20% on CoIR benchmark. In addition to excelling in code retrieval, our models demonstrate competitive performance on the widely adopted BeIR text retrieval benchmark, offering versatility across domains. Experimental results demonstrate that improving retrieval performance significantly enhances end-to-end Retrieval-Augmented Generation (RAG) performance for code-related tasks.

Agents built on vision language models (VLMs) now compete with learned policies on robot manipulation leaderboards. We ask how far a coding agent can go when it controls a robot instead of a computer, and where it stops. Our system, HexaShell, keeps the harness thin: it only turns the agent's commands into robot motion. The agent drives a robot with two arms through one command line. It sees three RGB cameras and knows their calibration. After each failed attempt, the agent reads the log, finds the cause, and writes or fixes small programs called tools. In this way it keeps what it learns as code, as a coding agent does when it fixes a failing test. The tools may use only what the robot observes. They use triangulation to find where objects are, relative to each other and to the gripper. If two cameras see the same point, the point lies where the two camera rays meet. If a point lies on a known plane, such as the table or the top of a block, it lies where its ray meets that plane. When the head camera is not enough, the robot moves a wrist camera to get a second view. We build the tools on only ten layouts of one seed per task, then freeze them and test them on ten new layouts of the same seed. We use Codex with GPT-6-Astra at medium reasoning effort. On all 42 RoboDojo tasks, HexaShell reaches a success rate of 39.7% and a progress score of 48.0, averaged in the same way as the leaderboard. This is a new state of the art on RoboDojo. The limit is contact, not reasoning: pushing reaches 92.5% and memory tasks 65.0%, but tight insertion only 8.3%. When the agent fails, it usually sees the failure but cannot make the motion work. We open source the tools evolved so far at https://github.com/HexaFuture/HexaShell.

We introduce DS-1000, a code generation benchmark with a thousand data science problems spanning seven Python libraries, such as NumPy and Pandas. Compared to prior works, DS-1000 incorporates three core features. First, our problems reflect diverse, realistic, and practical use cases since we collected them from StackOverflow. Second, our automatic evaluation is highly specific (reliable) -- across all Codex-002-predicted solutions that our evaluation accept, only 1.8% of them are incorrect; we achieve this with multi-criteria metrics, checking both functional correctness by running test cases and surface-form constraints by restricting API usages or keywords. Finally, we proactively defend against memorization by slightly modifying our problems to be different from the original StackOverflow source; consequently, models cannot answer them correctly by memorizing the solutions from pre-training. The current best public system (Codex-002) achieves 43.3% accuracy, leaving ample room for improvement. We release our benchmark at https://ds1000-code-gen.github.io.

Statistical language modeling and translation with transformers have found many successful applications in program understanding and generation tasks, setting high benchmarks for tools in modern software development environments. The finite context window of these neural models means, however, that they will be unable to leverage the entire relevant context of large files and packages for any given task. While there are many efforts to extend the context window, we introduce an architecture-independent approach for leveraging the syntactic hierarchies of source code for incorporating entire file-level context into a fixed-length window. Using concrete syntax trees of each source file we extract syntactic hierarchies and integrate them into context window by selectively removing from view more specific, less relevant scopes for a given task. We evaluate this approach on code generation tasks and joint translation of natural language and source code in Python programming language, achieving a new state-of-the-art in code completion and summarization for Python in the CodeXGLUE benchmark. We also introduce new CodeXGLUE benchmarks for user-experience-motivated tasks: code completion with normalized literals, method body completion/code summarization conditioned on file-level context.

AI agents combine language models with external data and tools that can modify files, call APIs, or execute code. Security failures can arise when adversarial content changes an agent's tool use or when the surrounding software contains vulnerabilities such as path traversal or command injection. We study authorized white-box pre-deployment auditing, where the auditor has access to the target repository and a controlled runtime, but successful attacks must still act through the task-defined attacker interface and be confirmed by an external verifier. We present AgentXploit, a two-role auditing system that separates repository-level attack-path discovery from runtime exploitation. The Analyzer Agent traces attacker-controlled inputs to sensitive operations and records code-supported candidate attack paths; the Exploiter Agent turns these paths into concrete attacks and revises them using runtime feedback. We also introduce AgentXploit-Bench, containing 72 reproducible vulnerabilities across 12 open-source AI-agent systems and frameworks. Across three runs, AgentXploit reaches 59.3% end-to-end success, compared with 38.4% for Codex. Under a token-budget-matched comparison, Codex reaches 46.3%. On AgentDojo, where injection points are provided, the Exploiter Agent reaches 79.2% attack success versus 52.7% for AgentVigil. These results highlight repository discovery and runtime exploitation as distinct challenges in end-to-end agent security auditing.

Command line interface (CLI) agents are emerging as a practical paradigm for agent-computer interaction over evolving filesystems, executable command line programs, and online execution feedback. Recent work has used reinforcement learning (RL) to learn these interaction abilities from verifiable task feedback, yet few methods exploit the native structured attributes of CLI actions as learning signals. Beyond this underused action structure, CLI learning also couples two bottlenecks for coding agents. First, the agent must identify task-relevant evidence in a large codebase from partial observations. Second, sparse terminal rewards must be assigned to the actions that shape a long multi-turn trajectory. We study these bottlenecks through shell-driven information extraction and file editing tasks. For selective observation, we introduce σ-Reveal, an inference-time mechanism that selects token-budgeted context for the same CLI. For credit assignment, we propose Action Advantage Assignment (A^3), a native agentic RL method that preserves the algorithmic complexity of standard agentic RL. A^3 constructs turn-level advantages from episode-level relative feedback, abstract syntax tree (AST) based action sub-chain residuals, and tree-level trajectory margins. To further evaluate this problem setting, we construct ShellOps, a verifiable dataset suite covering CLI tasks in repository environments.

We present CoDEx, a set of knowledge graph completion datasets extracted from Wikidata and Wikipedia that improve upon existing knowledge graph completion benchmarks in scope and level of difficulty. In terms of scope, CoDEx comprises three knowledge graphs varying in size and structure, multilingual descriptions of entities and relations, and tens of thousands of hard negative triples that are plausible but verified to be false. To characterize CoDEx, we contribute thorough empirical analyses and benchmarking experiments. First, we analyze each CoDEx dataset in terms of logical relation patterns. Next, we report baseline link prediction and triple classification results on CoDEx for five extensively tuned embedding models. Finally, we differentiate CoDEx from the popular FB15K-237 knowledge graph completion dataset by showing that CoDEx covers more diverse and interpretable content, and is a more difficult link prediction benchmark. Data, code, and pretrained models are available at https://bit.ly/2EPbrJs.

Recent autonomous penetration testing papers report high benchmark scores while adding multi-component security harnesses around frontier LLMs. Because these systems often change both architecture and backbone model, it is difficult to tell how much performance comes from the harness rather than from the underlying model. This paper presents a controlled study on the 104-task XBOW benchmark using default coding CLI agents as plain-agent baselines. We first run Codex, OpenCode, and Pi with the same GPT-5 model, budget, target interface, and scoring rule. This phase identifies the strongest same-model baseline and tests whether security-specific prompt variants improve its observed score. We then compare the default Codex scaffold with published MAPTA and PentestGPT V2 results under the closest available model matches. Finally, we repeat the plain-agent experiment with GPT-5.2 and GPT-5.5 to measure model scaling inside the same scaffold. The results show a mixed but practical picture. Specialised harnesses can add measurable benchmark lift and may improve cost efficiency, but plain coding agents already solve a large share of the benchmark; repeated plain-agent runs can match or exceed some published architecture scores in union coverage, and newer models substantially improve the same scaffold. Future evaluations should report model-matched plain-agent baselines before attributing benchmark gains to architecture design alone.

Open-source libraries and tools are widely reused, but compatibility maintenance is expensive. Once maintainers leave, useful repositories can stop working as runtimes and dependencies evolve. We study whether LLM agents can adapt old repositories to modern environments, a task we call compatibility rescue. Unlike bug repair, compatibility rescue starts from a repository that worked in its original environment but fails after ecosystem drift. RepoRescue gives agents only the repository and its failing modern environment; the agent must diagnose the failure, locate affected code, and produce a source-code rescue that restores the historical test suite. We build RepoRescue from 193 Python and 122 Java repositories, each verified to pass historically and fail after modernization. We evaluate five deployed agent systems on Python and three on Java. Beyond full-patch pass rate, we rerun patches after removing test-file edits to measure source-only repair, add a runtime-enforced regime that blocks test edits, and validate practical use for repositories whose suites pass after rescue. We find that Claude Code systems sometimes edit failing tests even when prompted not to; with runtime blocking, Kimi still rescues 41.5% of repositories. Systems are complementary: their union reaches 62.7%, exceeding the best single system by 10.9 points. Difficulty concentrates in cross-file coordination: on 14 repositories requiring coordinated whole-codebase changes, GPT-5.2 through Codex passes all 14, while every Claude Code system passes at most two. Finally, a passing suite is only an initial signal: among 34 unmaintained Python candidates whose suites pass after rescue, 22 work in realistic scenarios and 12 pass bug-hunt with patches that address the compatibility failure. RepoRescue benchmarks compatibility rescue with source-only auditing, runtime enforcement, practical validation, and reasoning labels.

Large language models (LLMs), such as Codex and GPT-4, have recently showcased their remarkable code generation abilities, facilitating a significant boost in coding efficiency. This paper will delve into utilizing LLMs for code generation in private libraries, as they are widely employed in everyday programming. Despite their remarkable capabilities, generating such private APIs poses a formidable conundrum for LLMs, as they inherently lack exposure to these private libraries during pre-training. To address this challenge, we propose a novel framework that emulates the process of programmers writing private code. This framework comprises two modules: APIFinder first retrieves potentially useful APIs from API documentation; and APICoder then leverages these retrieved APIs to generate private code. Specifically, APIFinder employs vector retrieval techniques and allows user involvement in the retrieval process. For APICoder, it can directly utilize off-the-shelf code generation models. To further cultivate explicit proficiency in invoking APIs from prompts, we continuously pre-train a reinforced version of APICoder, named CodeGenAPI. Our goal is to train the above two modules on vast public libraries, enabling generalization to private ones. Meanwhile, we create four private library benchmarks, including TorchDataEval, TorchDataComplexEval, MonkeyEval, and BeatNumEval, and meticulously handcraft test cases for each benchmark to support comprehensive evaluations. Numerous experiments on the four benchmarks consistently affirm the effectiveness of our approach. Furthermore, deeper analysis is also conducted to glean additional insights.

Current approaches to program synthesis with Large Language Models (LLMs) exhibit a "near miss syndrome": they tend to generate programs that semantically resemble the correct answer (as measured by text similarity metrics or human evaluation), but achieve a low or even zero accuracy as measured by unit tests due to small imperfections, such as the wrong input or output format. This calls for an approach known as Synthesize, Execute, Debug (SED), whereby a draft of the solution is generated first, followed by a program repair phase addressing the failed tests. To effectively apply this approach to instruction-driven LLMs, one needs to determine which prompts perform best as instructions for LLMs, as well as strike a balance between repairing unsuccessful programs and replacing them with newly generated ones. We explore these trade-offs empirically, comparing replace-focused, repair-focused, and hybrid debug strategies, as well as different template-based and model-based prompt-generation techniques. We use OpenAI Codex as the LLM and Program Synthesis Benchmark 2 as a database of problem descriptions and tests for evaluation. The resulting framework outperforms both conventional usage of Codex without the repair phase and traditional genetic programming approaches.

We present DictaLM, a large-scale language model tailored for Modern Hebrew. Boasting 7B parameters, this model is predominantly trained on Hebrew-centric data. As a commitment to promoting research and development in the Hebrew language, we release both the foundation model and the instruct-tuned model under a Creative Commons license. Concurrently, we introduce DictaLM-Rab, another foundation model geared towards Rabbinic/Historical Hebrew. These foundation models serve as ideal starting points for fine-tuning various Hebrew-specific tasks, such as instruction, Q&A, sentiment analysis, and more. This release represents a preliminary step, offering an initial Hebrew LLM model for the Hebrew NLP community to experiment with.

Large language models (LLMs), such as OpenAI's Codex, have demonstrated their potential to generate code from natural language descriptions across a wide range of programming tasks. Several benchmarks have recently emerged to evaluate the ability of LLMs to generate functionally correct code from natural language intent with respect to a set of hidden test cases. This has enabled the research community to identify significant and reproducible advancements in LLM capabilities. However, there is currently a lack of benchmark datasets for assessing the ability of LLMs to generate functionally correct code edits based on natural language descriptions of intended changes. This paper aims to address this gap by motivating the problem NL2Fix of translating natural language descriptions of code changes (namely bug fixes described in Issue reports in repositories) into correct code fixes. To this end, we introduce Defects4J-NL2Fix, a dataset of 283 Java programs from the popular Defects4J dataset augmented with high-level descriptions of bug fixes, and empirically evaluate the performance of several state-of-the-art LLMs for the this task. Results show that these LLMS together are capable of generating plausible fixes for 64.6% of the bugs, and the best LLM-based technique can achieve up to 21.20% top-1 and 35.68% top-5 accuracy on this benchmark.

The waning of Moore's Law has shifted the focus of the tech industry towards alternative methods for continued performance gains. While optimizing compilers are a standard tool to help increase program efficiency, programmers continue to shoulder much responsibility in crafting and refactoring code with better performance characteristics. In this paper, we investigate the ability of large language models (LLMs) to suggest functionally correct, performance improving code edits. We hypothesize that language models can suggest such edits in ways that would be impractical for static analysis alone. We investigate these questions by curating a large-scale dataset of Performance-Improving Edits, PIE. PIE contains trajectories of programs, where a programmer begins with an initial, slower version and iteratively makes changes to improve the program's performance. We use PIE to evaluate and improve the capacity of large language models. Specifically, use examples from PIE to fine-tune multiple variants of CODEGEN, a billion-scale Transformer-decoder model. Additionally, we use examples from PIE to prompt OpenAI's CODEX using a few-shot prompting. By leveraging PIE, we find that both CODEX and CODEGEN can generate performance-improving edits, with speedups of more than 2.5x for over 25% of the programs, for C++ and Python, even after the C++ programs were compiled using the O3 optimization level. Crucially, we show that PIE allows CODEGEN, an open-sourced and 10x smaller model than CODEX, to match the performance of CODEX on this challenging task. Overall, this work opens new doors for creating systems and methods that can help programmers write efficient code.

We evaluate AI-assisted generative capabilities on fundamental numerical kernels in high-performance computing (HPC), including AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG. We test the generated kernel codes for a variety of language-supported programming models, including (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numba, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). We use the GitHub Copilot capabilities powered by OpenAI Codex available in Visual Studio Code as of April 2023 to generate a vast amount of implementations given simple <kernel> + <programming model> + <optional hints> prompt variants. To quantify and compare the results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. Results suggest that the OpenAI Codex outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general-purpose Python can benefit from adding code keywords, while Julia prompts perform acceptably well for its mature programming models (e.g., Threads and CUDA.jl). We expect for these benchmarks to provide a point of reference for each programming model's community. Overall, understanding the convergence of large language models, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

A code generation model generates code by taking a prompt from a code comment, existing code, or a combination of both. Although code generation models (e.g. GitHub Copilot) are increasingly being adopted in practice, it is unclear whether they can successfully be used for unit test generation without fine-tuning. We investigated how well three generative models (Codex, GPT-3.5-Turbo, and StarCoder) can generate test cases to fill this gap. We used two benchmarks (HumanEval and Evosuite SF110) to investigate the context generation's effect in the unit test generation process. We evaluated the models based on compilation rates, test correctness, coverage, and test smells. We found that the Codex model achieved above 80% coverage for the HumanEval dataset, but no model had more than 2% coverage for the EvoSuite SF110 benchmark. The generated tests also suffered from test smells, such as Duplicated Asserts and Empty Tests.

Learning effective representations of source code is critical for any Machine Learning for Software Engineering (ML4SE) system. Inspired by natural language processing, large language models (LLMs) like Codex and CodeGen treat code as generic sequences of text and are trained on huge corpora of code data, achieving state of the art performance on several software engineering (SE) tasks. However, valid source code, unlike natural language, follows a strict structure and pattern governed by the underlying grammar of the programming language. Current LLMs do not exploit this property of the source code as they treat code like a sequence of tokens and overlook key structural and semantic properties of code that can be extracted from code-views like the Control Flow Graph (CFG), Data Flow Graph (DFG), Abstract Syntax Tree (AST), etc. Unfortunately, the process of generating and integrating code-views for every programming language is cumbersome and time consuming. To overcome this barrier, we propose our tool COMEX - a framework that allows researchers and developers to create and combine multiple code-views which can be used by machine learning (ML) models for various SE tasks. Some salient features of our tool are: (i) it works directly on source code (which need not be compilable), (ii) it currently supports Java and C#, (iii) it can analyze both method-level snippets and program-level snippets by using both intra-procedural and inter-procedural analysis, and (iv) it is easily extendable to other languages as it is built on tree-sitter - a widely used incremental parser that supports over 40 languages. We believe this easy-to-use code-view generation and customization tool will give impetus to research in source code representation learning methods and ML4SE. Tool: https://pypi.org/project/comex - GitHub: https://github.com/IBM/tree-sitter-codeviews - Demo: https://youtu.be/GER6U87FVbU

Extracting textual information from scanned medical documents, such as external laboratory reports and manually filled forms, has been a major challenge in modern electronic health records (EHRs). Recent advancements in vision language models (VLMs) have shown great promise over traditional OCR tools. However, at this point, most clinical OCR studies were conducted on private, institutional data. To our knowledge, there are few publicly available datasets for evaluating OCR models in the clinical domain. Furthermore, common scanning artifacts that undermine OCR performance are not reflected in those datasets, leaving a systematic evaluation unfeasible. Therefore, we release a publicly available, realistic-looking OCR benchmark dataset, ClinOCR-Bench, with 384 scanned images across 6 subsets: Normal, Handwriting, Poor Quality, Rotation, Tables, and Mix-artifacts. ClinOCR-Bench features: 1) diverse document types and layouts, 2) full coverage of common EHR scan artifacts, 3) protected health information-free, 4) template-aware train/test split, and 5) adequate sample size for OCR benchmarking. Baseline OCR performance was evaluated using state-of-the-art open-weight and proprietary VLMs. The dataset and documentation are available on GitHub (https://github.com/ClinOCR-Bench/ClinOCR-Bench).

Large language models (LLMs), such as Codex, hold great promise in enhancing programming education by automatically generating feedback for students. We investigate using LLMs to generate feedback for fixing syntax errors in Python programs, a key scenario in introductory programming. More concretely, given a student's buggy program, our goal is to generate feedback comprising a fixed program along with a natural language explanation describing the errors/fixes, inspired by how a human tutor would give feedback. While using LLMs is promising, the critical challenge is to ensure high precision in the generated feedback, which is imperative before deploying such technology in classrooms. The main research question we study is: Can we develop LLMs-based feedback generation techniques with a tunable precision parameter, giving educators quality control over the feedback that students receive? To this end, we introduce PyFiXV, our technique to generate high-precision feedback powered by Codex. The key idea behind PyFiXV is to use a novel run-time validation mechanism to decide whether the generated feedback is suitable for sharing with the student; notably, this validation mechanism also provides a precision knob to educators. We perform an extensive evaluation using two real-world datasets of Python programs with syntax errors and show the efficacy of PyFiXV in generating high-precision feedback.

Legal general intelligence (GI) refers to artificial intelligence (AI) that encompasses legal understanding, reasoning, and decision-making, simulating the expertise of legal experts across domains. However, existing benchmarks are result-oriented and fail to systematically evaluate the legal intelligence of large language models (LLMs), hindering the development of legal GI. To address this, we propose LexGenius, an expert-level Chinese legal benchmark for evaluating legal GI in LLMs. It follows a Dimension-Task-Ability framework, covering seven dimensions, eleven tasks, and twenty abilities. We use the recent legal cases and exam questions to create multiple-choice questions with a combination of manual and LLM reviews to reduce data leakage risks, ensuring accuracy and reliability through multiple rounds of checks. We evaluate 12 state-of-the-art LLMs using LexGenius and conduct an in-depth analysis. We find significant disparities across legal intelligence abilities for LLMs, with even the best LLMs lagging behind human legal professionals. We believe LexGenius can assess the legal intelligence abilities of LLMs and enhance legal GI development. Our project is available at https://github.com/QwenQKing/LexGenius.

Large language models (LLMs) have advanced rapidly from conversational problem solving to addressing real-world tasks involving tool use, such as software engineering (SWE). Recent LLM-powered toolkits, such as OpenAI Codex and Cursor, have offered end-to-end automation of the software development process. However, building effective SWE agents remains challenging due to the lack of high-quality training data and effective test cases. To address this issue, we present SWE-Dev, an SWE agent built upon open-source LLMs. First, we develop a robust pipeline to synthesize test cases for patch evaluation. Second, we scale up agent trajectories to construct the training data for building SWE-Dev. Experiments on the SWE-bench-Verified benchmark show that the SWE-Dev models can achieve top performance among all open SWE agents. Specifically, the success rates of the SWE-Dev 7B and 32B parameter models reach 23.4% and 36.6%, respectively, outperforming state-of-the-art open-source models. All code, models, and datasets are publicly available at https://github.com/THUDM/SWE-Dev.

Modern LLM training pipelines increasingly rely on other models to generate data, filter corpora, judge outputs, and guide development decisions. These dependencies are recursive: a model may depend on an upstream artifact whose own dependencies are documented only in separate releases and artifacts. As a result, the full dependency structure is fragmented across heterogeneous public artifacts, with complexity and recursive depth far outpacing humans' ability to trace. We introduce ModSleuth, an agentic system that recursively reconstructs LLM dependency graphs from public artifacts with source-grounded evidence. We find that the primary challenge is no longer information extraction, but defining what constitutes a dependency and reconciling artifact references across inconsistent documentation. We address these challenges through a formalization that distinguishes direct and indirect dependencies, represents heterogeneous pipeline roles through operation-centered relationships, and resolves artifact identities across names, versions, and repositories. Applying ModSleuth to four public-artifact-rich LLM releases, we recover 1,060 source-verified dependencies and construct large-scale dependency graphs of modern LLM development. These graphs reveal multi-hop license obligations, train-evaluation coupling, discrepancies between released and training-time artifacts, and documentation inconsistencies that would otherwise be difficult to uncover. We release ModSleuth and the resulting dependency graphs to support transparent analysis of the increasingly complex ecosystems underlying modern LLMs.

Recent autonomous AI agents such as Codex, and Claude Code have made it increasingly practical for users to delegate complex tasks, including writing emails, executing code, issuing shell commands, and carrying out multi-step plans. However, despite these capabilities, human-agent interaction still largely happens through terminal interfaces or remote text-based channels such as Discord. These interaction modes are often inefficient and unfriendly: long text outputs are difficult to read and review, proposed actions lack clear structure and visual context, and users must express feedback by typing detailed corrections, which is cumbersome and often discourages effective collaboration. As a result, non-expert users in particular face a high barrier to working productively with agents. To address this gap, we present AgentClick, an interactive review layer for terminal-based agents. AgentClick is implemented as a localhost npm server paired with a skill-based plugin that connects the running agent to a browser interface, allowing users to supervise and collaborate with agents through a structured web UI rather than raw terminal text alone. The system supports a range of human-in-the-loop workflows, including email drafting and revision, plan review and modification, memory management, trajectory inspection and visualization, and error localization during agent execution. It also turns code generation and execution into a reviewable process, enabling users to inspect and intervene before consequential actions are taken. In addition, AgentClick supports persistent preference capture through editable memory and remote access over HTTP, allowing users to review agents running on servers from their personal devices. Our goal is to lower the barrier for non-expert users and improve the efficiency and quality of human-agent co-work.

While recent LLM-based terminal agents have demonstrated promising capabilities, the scarcity of high-quality, executable training data remains a critical bottleneck. Existing synthesis pipelines typically scale by retrofitting surface-level artifacts into tasks, frequently yielding ambiguous instructions, shallow execution paths, and brittle tests that provide weak learning signals. To overcome this, we introduce CLI-Universe, a principled synthesis engine that constructs terminal-agent tasks. CLI-Universe generates candidate tasks by sampling combinations across a multi-dimensional capability taxonomy (domain, skill type, capability, and engineering pillar), then grounds each candidate through evidence-guided deep research over real-world technical materials. To ensure rigorous supervision, validated blueprints are instantiated into Dockerized environments and subjected to a multi-stage executable verification pipeline featuring rubric-gated test construction, hint-conditional filtering, and strict fail-to-pass checking. Across the full pipeline, from candidate generation to verification, approximately two-thirds of candidates are discarded, retaining only those that are genuine, verifiable, and non-trivially challenging. To validate our framework, we instantiate a highly distilled dataset of 6,000 trajectories called CLI-Universe-6K. Remarkably, fine-tuning Qwen3-32B on CLI-Universe-6K achieves 33.4% on Terminal-Bench 2.0. This sets a new state-of-the-art for models trained on open-source data at or below 32B parameters, and outperforms several models an order of magnitude larger, demonstrating the profound data efficiency of structured, high-fidelity synthesis.

AI coding agents are reshaping software development through both autonomous and human-mediated pull requests (PRs). When developers use AI agents to generate code under their own accounts, code authorship attribution becomes critical for repository governance, research validity, and understanding modern development practices. We present the first study on fingerprinting AI coding agents, analyzing 33,580 PRs from five major agents (OpenAI Codex, GitHub Copilot, Devin, Cursor, Claude Code) to identify behavioral signatures. With 41 features spanning commit messages, PR structure, and code characteristics, we achieve 97.2% F1-score in multi-class agent identification. We uncover distinct fingerprints: Codex shows unique multiline commit patterns (67.5% feature importance), and Claude Code exhibits distinctive code structure (27.2% importance of conditional statements). These signatures reveal that AI coding tools produce detectable behavioral patterns, suggesting potential for identifying AI contributions in software repositories.

Recent breakthroughs in pre-trained code models, such as CodeBERT and Codex, have shown their superior performance in various downstream tasks. The correctness and unambiguity of API usage among these code models are crucial for achieving desirable program functionalities, requiring them to learn various API fully qualified names structurally and semantically. Recent studies reveal that even state-of-the-art pre-trained code models struggle with suggesting the correct APIs during code generation. However, the reasons for such poor API usage performance are barely investigated. To address this challenge, we propose using knowledge probing as a means of interpreting code models, which uses cloze-style tests to measure the knowledge stored in models. Our comprehensive study examines a code model's capability of understanding API fully qualified names from two different perspectives: API call and API import. Specifically, we reveal that current code models struggle with understanding API names, with pre-training strategies significantly affecting the quality of API name learning. We demonstrate that natural language context can assist code models in locating Python API names and generalize Python API name knowledge to unseen data. Our findings provide insights into the limitations and capabilities of current pre-trained code models, and suggest that incorporating API structure into the pre-training process can improve automated API usage and code representations. This work provides significance for advancing code intelligence practices and direction for future studies. All experiment results, data and source code used in this work are available at https://doi.org/10.5281/zenodo.7902072.

Advances in handwritten text recognition have enabled large-scale transcription of historical documents, but still provide limited access to interpretable visual measurements for paleography, the study of historical scripts. In this paper, our main insight is that morphological script analysis, in particular the capacity to learn character prototypes from line-level transcriptions, enables the definition of scalable, meaningful, and stable paleographic measurements. More precisely, we leverage a transformer-based detection architecture together with a prototype-based line reconstruction module to learn prototypical characters and their occurrence, deformation, and positioning. Our contributions are twofold. First, we introduce a deep architecture and learning methodology that enables efficient character modeling with only line-level transcription supervision, significantly improving over the Learnable Typewriter baseline and enabling accurate character bounding box prediction, unlocking its potential for paleographic measurements. Second, we introduce and demonstrate the paleographical relevance of automatic measurements enabled by our architecture for characters, bi-grams, and spaces between graphical units. For this demonstration, we extend the annotations of the codex Paris, BnF, fr. 2813, commissioned in the late fourteenth century by Charles V and copied by four hands, to 160 pages. We visualize our measurements over these pages, showing how they enable us not only to differentiate graphical profiles, but also to discover and analyze subtle variations. This case study outlines the scalability of our approach and its frugality in terms of required training data, since a single column of text is sufficient to compute our measurements on each of the 160 pages. Data and code are publicly available at: https://malamatenia.github.io/morphology4metrology-analysis.

Large Language Models (LLMs) are driving a shift towards intent-driven development, where agents build complete software from scratch. However, existing benchmarks fail to assess this 0-to-1 generation capability due to two limitations: reliance on predefined scaffolds that ignore repository structure planning, and rigid white-box unit testing that lacks end-to-end behavioral validation. To bridge this gap, we introduce CLI-Tool-Bench, a structure-agnostic benchmark for evaluating the ground-up generation of Command-Line Interface (CLI) tools. It features 100 diverse real-world repositories evaluated via a black-box differential testing framework. Agent-generated software is executed in sandboxes, comparing system side effects and terminal outputs against human-written oracles using multi-tiered equivalence metrics. Evaluating seven state-of-the-art LLMs, we reveal that top models achieve under 43% success, highlighting the ongoing challenge of 0-to-1 generation. Furthermore, higher token consumption does not guarantee better performance, and agents tend to generate monolithic code.

Recent developments show that Large Language Models (LLMs) produce state-of-the-art performance on natural language (NL) to code generation for resource-rich general-purpose languages like C++, Java, and Python. However, their practical usage for structured domain-specific languages (DSLs) such as YAML, JSON is limited due to domain-specific schema, grammar, and customizations generally unseen by LLMs during pre-training. Efforts have been made to mitigate this challenge via in-context learning through relevant examples or by fine-tuning. However, it suffers from problems, such as limited DSL samples and prompt sensitivity but enterprises maintain good documentation of the DSLs. Therefore, we propose DocCGen, a framework that can leverage such rich knowledge by breaking the NL-to-Code generation task for structured code languages into a two-step process. First, it detects the correct libraries using the library documentation that best matches the NL query. Then, it utilizes schema rules extracted from the documentation of these libraries to constrain the decoding. We evaluate our framework for two complex structured languages, Ansible YAML and Bash command, consisting of two settings: Out-of-domain (OOD) and In-domain (ID). Our extensive experiments show that DocCGen consistently improves different-sized language models across all six evaluation metrics, reducing syntactic and semantic errors in structured code. We plan to open-source the datasets and code to motivate research in constrained code generation.

AI agents powered by large language models exhibit strong reasoning and problem-solving capabilities, enabling them to assist scientific research tasks such as formula derivation and code generation. However, whether these agents can reliably perform end-to-end reproduction from real scientific papers remains an open question. We introduce PRBench, a benchmark of 30 expert-curated tasks spanning 11 subfields of physics. Each task requires an agent to comprehend the methodology of a published paper, implement the corresponding algorithms from scratch, and produce quantitative results matching the original publication. Agents are provided only with the task instruction and paper content, and operate in a sandboxed execution environment. All tasks are contributed by domain experts from over 20 research groups at the School of Physics, Peking University, each grounded in a real published paper and validated through end-to-end reproduction with verified ground-truth results and detailed scoring rubrics. Using an agentified assessment pipeline, we evaluate a set of coding agents on PRBench and analyze their capabilities across key dimensions of scientific reasoning and execution. The best-performing agent, OpenAI Codex powered by GPT-5.3-Codex, achieves a mean overall score of 34%. All agents exhibit a zero end-to-end callback success rate, with particularly poor performance in data accuracy and code correctness. We further identify systematic failure modes, including errors in formula implementation, inability to debug numerical simulations, and fabrication of output data. Overall, PRBench provides a rigorous benchmark for evaluating progress toward autonomous scientific research.

Unlocking medieval archives at scale is bottlenecked by the cost of line-level diplomatic transcriptions and manual abbreviation expansion. Focusing on the registers of Pope Honorius III (1216-1227) over 5,000 letters surviving in full only as manuscript images - we show that existing, line-break-free expert transcriptions (edition drafts) suffice to train end-to-end hand-written text recognition (HTR) that jointly transcribes and expands heavily abbreviated Latin. Using forced alignment on 860 unaligned edition entries across 394 pages of Reg. Vat. 9, we automatically bootstrap 13,122 training lines without manual alignment of segmented lines. While the generic CATMuS baseline struggles on this domain (18.9% CER) and a modular space-free HTR + word-segmentation pipeline (cf. Camps et al. 2021) plateaus at 15.6% WER, our from-scratch hybrid Conformer - combining CTC with bidirectional subword attention decoders, iterative self-realignment and n-best LM rescoring - achieves 4.7% CER and 8.7% WER (8.4% WER with ROVER system combination), nearly halving errors on clause-final words (22.7% to 12.2% WER). These figures are measured on the test lines that align with the transcription (86% of all test lines for the best hybrid). Furthermore, we address a tension between paleographic fidelity and editorial normalization: training on targets that preserve scribal name initials reduces the WER against the manuscript text on the 31 affected test lines from 13.3% to 9.2%, at a small cost against the expanded transcription (WER +0.3 points, CER +0.4). Finally, extensive ablations show that none of an auxiliary diplomatic CTC head, a copy mechanism, a permutation decoder or masked-image pretraining improves on direct end-to-end supervision, while strict padding-mask propagation and per-line beam termination make inference batch-invariant.

Software engineering activities such as package migration, fixing errors reports from static analysis or testing, and adding type annotations or other specifications to a codebase, involve pervasively editing the entire repository of code. We formulate these activities as repository-level coding tasks. Recent tools like GitHub Copilot, which are powered by Large Language Models (LLMs), have succeeded in offering high-quality solutions to localized coding problems. Repository-level coding tasks are more involved and cannot be solved directly using LLMs, since code within a repository is inter-dependent and the entire repository may be too large to fit into the prompt. We frame repository-level coding as a planning problem and present a task-agnostic framework, called CodePlan to solve it. CodePlan synthesizes a multi-step chain of edits (plan), where each step results in a call to an LLM on a code location with context derived from the entire repository, previous code changes and task-specific instructions. CodePlan is based on a novel combination of an incremental dependency analysis, a change may-impact analysis and an adaptive planning algorithm. We evaluate the effectiveness of CodePlan on two repository-level tasks: package migration (C#) and temporal code edits (Python). Each task is evaluated on multiple code repositories, each of which requires inter-dependent changes to many files (between 2-97 files). Coding tasks of this level of complexity have not been automated using LLMs before. Our results show that CodePlan has better match with the ground truth compared to baselines. CodePlan is able to get 5/6 repositories to pass the validity checks (e.g., to build without errors and make correct code edits) whereas the baselines (without planning but with the same type of contextual information as CodePlan) cannot get any of the repositories to pass them.

Translating a morphology-rich, low-resource language like Latin poses significant challenges. This paper introduces a reproducible draft-based refinement pipeline that elevates open-source Large Language Models (LLMs) to a performance level statistically comparable to top-tier proprietary systems. Our method first uses a fine-tuned NLLB-1.3B model to generate a high-quality, structurally faithful draft. A zero-shot LLM (Llama-3.3 or Qwen3) then polishes this draft, a process that can be further enhanced by augmenting the context with retrieved out-context examples (RAG). We demonstrate the robustness of this approach on two distinct benchmarks: a standard in-domain test set (Rosenthal, 2023) and a new, challenging out-of-domain (OOD) set of 12th-century Latin letters (2025). Our central finding is that this open-source RAG system achieves performance statistically comparable to the GPT-5 baseline, without any task-specific LLM fine-tuning. We release the pipeline, the Chartres OOD set, and evaluation scripts and models to facilitate replicability and further research.

Agentic coding requires agents to effectively interact with runtime environments, e.g., command line interfaces (CLI), so as to complete tasks like resolving dependency issues, fixing system problems, etc. But it remains underexplored how such environment-intensive tasks can be obtained at scale to enhance agents' capabilities. To address this, based on an analogy between the Dockerfile and the agentic task, we propose to employ agents to simulate and explore environment histories, guided by execution feedback. By tracing histories of a healthy environment, its state can be inverted to an earlier one with runtime failures, from which a task can be derived by packing the buggy state and the corresponding error messages. With our method, named CLI-Gym, a total of 1,655 environment-intensive tasks are derived, being the largest collection of its kind. Moreover, with curated successful trajectories, our fine-tuned model, named LiberCoder, achieves substantial absolute improvements of +21.1% (to 46.1%) on Terminal-Bench, outperforming various strong baselines. To our knowledge, this is the first public pipeline for scalable derivation of environment-intensive tasks.

The task of generating code solutions for a given programming problem can benefit from the use of pre-trained language models such as Codex, which can produce multiple diverse samples. However, a major challenge for this task is to select the most appropriate solution from the multiple samples generated by the pre-trained language models. A natural way to evaluate the quality and correctness of a code solution is to run it against a set of test cases, but the manual creation of such test cases is often costly and time-consuming. In this paper, we propose a novel method, CodeT, that leverages the same pre-trained language models to automatically generate test cases for the code samples, thus reducing the human effort and increasing the coverage of the test scenarios. CodeT then executes the code samples using the generated test cases, and performs a dual execution agreement, which considers both the consistency of the outputs against the generated test cases and the agreement of the outputs with other code samples. We conduct comprehensive experiments on four benchmarks, HumanEval, MBPP, APPS and CodeContests, using five different pre-trained language models with varying sizes and capabilities. Our results show that CodeT can significantly improve the performance of code solution selection over previous methods, achieving remarkable and consistent gains across different models and benchmarks. For instance, CodeT improves the pass@1 metric on HumanEval to 65.8%, which represents an absolute improvement of 18.8% over the code-davinci-002 model, and an absolute improvement of more than 20% over the previous state-of-the-art results.

Regesta are catalogs of summaries of other documents and, in some cases, are the only source of information about the content of such full-length documents. For this reason, they are of great interest to scholars in many social and humanities fields. In this work, we focus on Regesta Pontificum Romanum, a large collection of papal registers. Regesta are visually rich documents, where the layout is as important as the text content to convey the contained information through the structure, and are inherently multi-page documents. Among Digital Humanities techniques that can help scholars efficiently exploit regesta and other documental sources in the form of scanned documents, Document Parsing has emerged as a task to process document images and convert them into machine-readable structured representations, usually markup language. However, current models focus on scientific and business documents, and most of them consider only single-paged documents. To overcome this limitation, in this work, we propose {\mu}gat, an extension of the recently proposed Document parsing Nougat architecture, which can handle elements spanning over the single page limits. Specifically, we adapt Nougat to process a larger, multi-page context, consisting of the previous and the following page, while parsing the current page. Experimental results, both qualitative and quantitative, demonstrate the effectiveness of our proposed approach also in the case of the challenging Regesta Pontificum Romanorum.

Image segmentation is usually addressed by training a model for a fixed set of object classes. Incorporating additional classes or more complex queries later is expensive as it requires re-training the model on a dataset that encompasses these expressions. Here we propose a system that can generate image segmentations based on arbitrary prompts at test time. A prompt can be either a text or an image. This approach enables us to create a unified model (trained once) for three common segmentation tasks, which come with distinct challenges: referring expression segmentation, zero-shot segmentation and one-shot segmentation. We build upon the CLIP model as a backbone which we extend with a transformer-based decoder that enables dense prediction. After training on an extended version of the PhraseCut dataset, our system generates a binary segmentation map for an image based on a free-text prompt or on an additional image expressing the query. We analyze different variants of the latter image-based prompts in detail. This novel hybrid input allows for dynamic adaptation not only to the three segmentation tasks mentioned above, but to any binary segmentation task where a text or image query can be formulated. Finally, we find our system to adapt well to generalized queries involving affordances or properties. Code is available at https://eckerlab.org/code/clipseg.

LLMs are increasingly applied to cybersecurity workflows, where they are expected to translate analysts' intent into tool invocations. However, existing evaluations focus on knowledge-based assessments or end-to-end agentic tasks, and do not directly measure LLMs' ability to generate executable commands for real-world cybersecurity tools. This gap is critical because cybersecurity operations rely on strict command-line interfaces (CLIs), where minor syntax errors, incorrect flag--value bindings, or argument misordering can invalidate execution. We introduce KaliBench, a fine-grained benchmark and dataset for natural-language--to--CLI translation on Kali Linux, comprising 8,504 query--command pairs spanning 1,642 tools across 23 capability dimensions and 5 security phases. KaliBench is constructed via a manuscript-grounded pipeline with deterministic canonicalization and alias-aware evaluation, enabling precise and reproducible assessment of tool selection and argument construction. To ensure both semantic correctness and practical executability, we develop a multi-stage verification pipeline that combines LLM-based validation, sandboxed terminal execution, and human-in-the-loop refinement. Building on these fine-grained, deterministic signals, KaliBench further enables runtime-free verifiable rewards for training. Across three evaluation modes and 24 configurations of general-purpose and security-focused open-weight models, no open-weight model exceeds 42% exact-command accuracy in the unrestricted setting, highlighting the difficulty of accurate CLI-based cybersecurity tool use without explicit tool hints. We further show that supervised fine-tuning and reinforcement learning with verifiable rewards derived from KaliBench significantly improve an 8B model and achieve performance comparable to a 685B MoE model.

Command-line interface (CLI) agents increasingly execute longer-horizon tasks, yet existing benchmarks either assume that user instructions remain fixed throughout execution or evaluate task outcomes without capturing runtime collaboration quality. In this paper, we present REALCLI, a paired benchmark that unifies the evaluation of task execution and human-agent collaboration quality on realistic CLI tasks. To comprehensively measure collaboration, REALCLI builds a five-dimensional framework, operationalized into 22 sub-dimensions and 51 interaction scenario templates. REALCLI provides two modes for paired comparison: REALCLI-EXEC supplies complete specifications upfront, whereas REALCLI-COLLAB introduces a scripted user that delivers progressive information, corrections, constraint revisions, and authorization decisions. Evaluations reveal profound capability gaps in frontier agents: the best configuration attains an average score of only 49.2%, and when evaluated by the joint collaboration score (J_{collab}), the peak performance reaches merely 37.6%. These results show that task effectiveness diverges sharply from collaboration quality, with models consistently struggling in steerability, trust calibration, and safety boundary management. REALCLI provides a rigorous, reproducible foundation for measuring and advancing collaborative CLI agents.

We introduce the workbook time machine, a pipeline that automatically creates benchmarks evaluating the ability of language models to create derived objects in spreadsheets (formulas, charts, pivot tables, and conditional formatting). Applied to public workbook corpora, it produces wtmcorpus--a collection of (input workbook, output workbook, query) triples spanning four artifact types and varying complexity. From this corpus we curate wtmbench, a 150-task evaluation benchmark with queries at three levels of specificity. We evaluate existing spreadsheet manipulation agents and baselines on wtmbench across artifact types, step complexity, and instruction granularity. Our evaluations show that query specificity, agent orchestration, and interface API used to control spreadsheets play a big role in LLM performance on Excel tasks.

The ultimate goal of code agents is to solve complex tasks autonomously. Although large language models (LLMs) have made substantial progress in code generation, real-world tasks typically demand full-fledged code repositories rather than simple scripts. Building such repositories from scratch remains a major challenge. Fortunately, GitHub hosts a vast, evolving collection of open-source repositories, which developers frequently reuse as modular components for complex tasks. Yet, existing frameworks like OpenHands and SWE-Agent still struggle to effectively leverage these valuable resources. Relying solely on README files provides insufficient guidance, and deeper exploration reveals two core obstacles: overwhelming information and tangled dependencies of repositories, both constrained by the limited context windows of current LLMs. To tackle these issues, we propose RepoMaster, an autonomous agent framework designed to explore and reuse GitHub repositories for solving complex tasks. For efficient understanding, RepoMaster constructs function-call graphs, module-dependency graphs, and hierarchical code trees to identify essential components, providing only identified core elements to the LLMs rather than the entire repository. During autonomous execution, it progressively explores related components using our exploration tools and prunes information to optimize context usage. Evaluated on the adjusted MLE-bench, RepoMaster achieves a 110% relative boost in valid submissions over the strongest baseline OpenHands. On our newly released GitTaskBench, RepoMaster lifts the task-pass rate from 40.7% to 62.9% while reducing token usage by 95%. Our code and demonstration materials are publicly available at https://github.com/QuantaAlpha/RepoMaster.

Large Language Model (LLM) agents, capable of performing a broad range of actions, such as invoking tools and controlling robots, show great potential in tackling real-world challenges. LLM agents are typically prompted to produce actions by generating JSON or text in a pre-defined format, which is usually limited by constrained action space (e.g., the scope of pre-defined tools) and restricted flexibility (e.g., inability to compose multiple tools). This work proposes to use executable Python code to consolidate LLM agents' actions into a unified action space (CodeAct). Integrated with a Python interpreter, CodeAct can execute code actions and dynamically revise prior actions or emit new actions upon new observations through multi-turn interactions. Our extensive analysis of 17 LLMs on API-Bank and a newly curated benchmark shows that CodeAct outperforms widely used alternatives (up to 20% higher success rate). The encouraging performance of CodeAct motivates us to build an open-source LLM agent that interacts with environments by executing interpretable code and collaborates with users using natural language. To this end, we collect an instruction-tuning dataset CodeActInstruct that consists of 7k multi-turn interactions using CodeAct. We show that it can be used with existing data to improve models in agent-oriented tasks without compromising their general capability. CodeActAgent, finetuned from Llama2 and Mistral, is integrated with Python interpreter and uniquely tailored to perform sophisticated tasks (e.g., model training) using existing libraries and autonomously self-debug.

Large language model (LLM) agents are evolving from request-response assistants into long-running software actors: they maintain state across model calls, fork subtasks, wait for external events, request human authority, generate tools, and perform side effects that must be resumed and audited. This paper presents Agent libOS, a library-OS-inspired runtime substrate for LLM agents. Agent libOS runs above a conventional host operating system; it does not implement hardware drivers, kernel-mode isolation, or a POSIX-compatible operating system. Instead, it treats an agent as an AgentProcess: a schedulable execution subject with process identity, parent-child lineage, lifecycle state, a tool table derived from an AgentImage, typed Object Memory, explicit capabilities, human queues, checkpoints, events, and audit records. Its central design rule is tools are libc-like wrappers; runtime primitives are the authority boundary. Filesystem access, object access, sleeps, human approval, JIT tool registration, and external side effects are checked at primitive boundaries under explicit capabilities and policy. We describe the design, threat model, Python prototype, and safety-oriented evaluation. The current prototype implements async scheduling, namespace-local Object Memory, runtime-integrated human approval, one-shot permission grants, per-process working directories, shell and image-registration primitives, Deno/TypeScript JIT tools over a libOS syscall broker, filesystem/object bridge tools, an injectable Resource Provider Substrate, deterministic demos, real-model smoke scripts, and 123 regression tests at the time of writing. Rather than improving planner accuracy, Agent libOS demonstrates a runtime substrate in which long-running LLM agents can be scheduled, authorized, resumed, and audited without treating tool dispatch as the trust boundary.

Code completion, which aims to predict the following code token(s) according to the code context, can improve the productivity of software development. Recent work has proved that statistical language modeling with transformers can greatly improve the performance in the code completion task via learning from large-scale source code datasets. However, current approaches focus only on code context within the file or project, i.e. internal context. Our distinction is utilizing "external" context, inspired by human behaviors of copying from the related code snippets when writing code. Specifically, we propose a retrieval-augmented code completion framework, leveraging both lexical copying and referring to code with similar semantics by retrieval. We adopt a stage-wise training approach that combines a source code retriever and an auto-regressive language model for programming language. We evaluate our approach in the code completion task in Python and Java programming languages, achieving a state-of-the-art performance on CodeXGLUE benchmark.

The landscape of AI coding assistance is undergoing a fundamental shift from complex IDE plugins to versatile, terminal-native agents. Operating directly where developers manage source control, execute builds, and deploy environments, CLI-based agents offer unprecedented autonomy for long-horizon development tasks. In this paper, we present OPENDEV, an open-source, command-line coding agent engineered specifically for this new paradigm. Effective autonomous assistance requires strict safety controls and highly efficient context management to prevent context bloat and reasoning degradation. OPENDEV overcomes these challenges through a compound AI system architecture with workload-specialized model routing, a dual-agent architecture separating planning from execution, lazy tool discovery, and adaptive context compaction that progressively reduces older observations. Furthermore, it employs an automated memory system to accumulate project-specific knowledge across sessions and counteracts instruction fade-out through event-driven system reminders. By enforcing explicit reasoning phases and prioritizing context efficiency, OPENDEV provides a secure, extensible foundation for terminal-first AI assistance, offering a blueprint for robust autonomous software engineering.

Recent advances in large language models (LLMs) like GPT-3.5 and GPT-4 promise automation with better results and less programming, opening up new opportunities for text analysis in political science. In this study, we evaluate LLMs on three original coding tasks involving typical complexities encountered in political science settings: a non-English language, legal and political jargon, and complex labels based on abstract constructs. Along the paper, we propose a practical workflow to optimize the choice of the model and the prompt. We find that the best prompting strategy consists of providing the LLMs with a detailed codebook, as the one provided to human coders. In this setting, an LLM can be as good as or possibly better than a human annotator while being much faster, considerably cheaper, and much easier to scale to large amounts of text. We also provide a comparison of GPT and popular open-source LLMs, discussing the trade-offs in the model's choice. Our software allows LLMs to be easily used as annotators and is publicly available: https://github.com/lorelupo/pappa.

We introduce Context Language Models (CLMs), language models that natively manage their own context. We implement this by treating the context as a file and allowing the model to make unrestricted updates to this file. This allows the model to learn what is most important to maintain in context, and naturally extends to multi-agent systems where multiple agent contexts coexist as files. Building CLMs zero-shot with existing models outperforms SOTA context management strategies across a variety of tasks: 11.4% higher accuracy with 21.5% fewer FLOPs on BrowseComp-Plus, 5% higher scores with 59% fewer FLOPs on 12-hour EdgeBench, and 65% greater improvement with the same compute on a 24-hour multi-repository agent-swarm task. Moreover, by shifting context management from external harness control to intrinsic model behavior, CLMs naturally enable both in-context and parametric learning of context-management strategies. We show that CLMs can be steered with natural-language instructions evolved through a standard skill-optimization loop, improving held-out accuracy by up to 35.9 points on a context-management task while reducing compute. We also introduce an online reinforcement learning method for CLMs, improving Qwen3.5-9B performance on BrowseComp-Plus by 47.6% while using 12% fewer FLOPs. Finally, we co-design Suffix Cache Reuse for CLM serving, further reducing server-side compute by 35% relative to standard SGLang at matched performance.

Many European languages possess rich biblical translation histories, yet existing corpora - in prioritizing linguistic breadth - often fail to capture this depth. To address this gap, we introduce a multilingual corpus of 657 New Testament translations, of which 352 are unique, with unprecedented depth in five languages: English (208 unique versions from 396 total), French (41 from 78), Italian (18 from 33), Polish (30 from 48), and Spanish (55 from 102). Aggregated from 12 online biblical libraries and one preexisting corpus, each translation is manually annotated with metadata that maps the text to a standardized identifier for the work, its specific edition, and its year of revision. This canonicalization empowers researchers to define "uniqueness" for their own needs: they can perform micro-level analyses on translation families, such as the KJV lineage, or conduct macro-level studies by deduplicating closely related texts. By providing the first resource designed for such flexible, multilevel analysis, our corpus establishes a new benchmark for the quantitative study of translation history.

Large Language Models (LLMs) have shown significant capability across various tasks, with their real-world effectiveness often driven by prompt design. While recent research has focused on optimizing prompt content, the role of prompt formatting, a critical but often overlooked dimension, has received limited systematic investigation. In this paper, we introduce Content-Format Integrated Prompt Optimization (CFPO), an innovative methodology that jointly optimizes both prompt content and formatting through an iterative refinement process. CFPO leverages natural language mutations to explore content variations and employs a dynamic format exploration strategy that systematically evaluates diverse format options. Our extensive evaluations across multiple tasks and open-source LLMs demonstrate that CFPO demonstrates measurable performance improvements compared to content-only optimization methods. This highlights the importance of integrated content-format optimization and offers a practical, model-agnostic approach to enhancing LLM performance. Code will be available at https://github.com/HenryLau7/CFPO.

This paper presents a novel task of extracting low-resourced and noisy Latin fragments from mixed-language historical documents with varied layouts. We benchmark and evaluate the performance of large foundation models against a multimodal dataset of 724 annotated pages. The results demonstrate that reliable Latin detection with contemporary zero-shot models is achievable, yet these models lack a functional comprehension of Latin. This study establishes a comprehensive baseline for processing Latin within mixed-language corpora, supporting quantitative analysis in intellectual history and historical linguistics. Both the dataset and code are available at https://github.com/COMHIS/EACL26-detect-latin.

Coding agents, autonomous systems that use large language models (LLMs) to resolve software engineering tasks, rely on agentic scaffolding: a middleware layer in between a developer and a large language model that orchestrates system prompts, tool execution, context management, and iterative reasoning loops. While these scaffoldings evolve at extreme velocities, no study has examined how this evolution affects agent quality (i.e., effectiveness and efficiency) over time. Practitioners regularly report quality regressions after scaffolding updates, yet consistently attribute them to the underlying model rather than the scaffolding itself. In this paper, we address this gap by conducting the first controlled longitudinal study that isolates the scaffolding's contribution. Unlike prior work that fixes the scaffolding and varies the model, we fix the model and vary only the scaffolding, evaluating 35 sequential releases to measure their impact on agent effectiveness and efficiency. We first empirically study the development and release evolution of five major open-source scaffoldings (i.e., Codex, Qwen Code, Gemini, OpenCode, and OpenHands), revealing extreme release velocities exceeding two releases per day and thousands of issues within months. We then perform a controlled deep dive into 35 sequential releases of the Qwen Code CLI, evaluating each against 50 stratified SWE-bench Verified tasks while holding the underlying LLM constant. We trace the resulting quality fluctuations to specific development patterns and architectural components, and illustrate our findings with concrete qualitative evidence linking individual pull requests to measured quality shifts.

Benchmarks are widely used to evaluate task completion by Large Language Models (LLMs), but this approach has accumulated construction-validity problems, and a passing score may not show whether the requested task was delivered. We study both problems. In a controlled code-as-spec setup, two production Copilot CLI agents (claude-opus-4.7, gpt-5.5) re-implement a React Fluent-UI data table in Angular as a reusable library under a hidden 222-test Playwright oracle across 18 runs and three oracle-availability conditions. Alongside the score, we run a mechanical library audit and check each verdict with a no-op ablation. Without the oracle, the library is present but unfinished, revealed by scores. With the oracle in the loop, the score reaches near-perfect, but from a demo holding the tested behavior directly, the library left dead or absent. We call this building to the test; the broader disposition behind both we call validation self-awareness. The agent does not, on its own, validate what it ships as a user would. Prevalence remains an open question across other agents, signals, and model families. Beyond benchmark scores, dispositions like validation self-awareness merit research attention.

Computer-use agents have advanced on benchmarks like OSWorld and AndroidWorld, but still act mostly through the GUI, often producing inefficient trajectories. Real-world computer work is hybrid, combining visual-state inspection with precise, high-throughput command-line operations, so capable agents must coordinate both modalities over shared application state. Yet scalable hybrid environments remain scarce because supporting both GUI and CLI over real applications typically requires substantial manual engineering for each application. Existing agents also struggle to use the two interfaces complementarily: CLI-native agents lack visual perception for tasks involving interface state or layout, while GUI-native agents are inefficient for operations better executed through commands. We introduce CUA-Universe, a scalable environment-to-data pipeline that turns real desktop software into hybrid GUI+CLI environments. App-Forge adapts applications into reproducible VMs and command-line surfaces it discovers, wraps, or generates, scaling to 16 applications; Task-Weave synthesizes diverse hybrid tasks of controllable difficulty from reusable operations over seed files; and Path-Steer steers rollouts along efficient hybrid paths and harvests verified trajectories for post-training. Training on this data shifts behavior from inefficient GUI interaction and brittle CLI scripting toward effective GUI+CLI orchestration. Our 9B model improves both success and efficiency on CUA-Verse (Score +39.3 pts; -37% steps, -60% tokens), OSWorld (SR +16.8 pts; -57% steps, -44% tokens), and OSWorld-MCP (Score +7.84 pts; -27% steps, -30% tokens). CUA-Universe provides a scalable path toward more capable and efficient computer-use agents.

Large language models are increasingly used within larger systems ("LLM agents"). These make a sequence of LLM calls, each call providing the LLM with a combination of instructions, observations, and interaction history. The design of the encoded information and its structure play a central role in the quality of the resulting system, leading to efforts spent on context engineering. It is therefore critical to communicate the composition of the LLM context in a system, and how it evolves over time. Yet, no standard exists for doing so: context construction is typically conveyed through informal prose, ad hoc diagrams, or direct inspection of code, none of which precisely capture how a prompt evolves across interaction steps or how two context representation strategies differ. To remedy this, we introduce the Agentic Context Description Language (ACDL), a language for specifying the structure and dynamics of LLM input contexts in a precise, readable, and standard manner, along with visualizations. ACDL provides constructs for specifying context aspects such as role message sequences, dynamic content, time-indexed references, and conditional or iterative structure, capturing the full architecture of a prompt independently of any particular implementation. ACDL diagrams can be hand drawn on a whiteboard, or written in formal language which can then be rendered. We describe the language, demonstrate it by documenting several existing systems and their variants, and encourage the community to adopt it for describing LLM systems context, both in day-to-day communication and in papers. Tooling, examples and documentation are available at www.acdlang.org.

Although large language models (LLMs) have demonstrated impressive coding capabilities, their ability to autonomously build production-scale software from explicit specifications remains an open question. We introduce SWE-AGI, an open-source benchmark for evaluating end-to-end, specification-driven construction of software systems written in MoonBit. SWE-AGI tasks require LLM-based agents to implement parsers, interpreters, binary decoders, and SAT solvers strictly from authoritative standards and RFCs under a fixed API scaffold. Each task involves implementing 1,000-10,000 lines of core logic, corresponding to weeks or months of engineering effort for an experienced human developer. By leveraging the nascent MoonBit ecosystem, SWE-AGI minimizes data leakage, forcing agents to rely on long-horizon architectural reasoning rather than code retrieval. Across frontier models, gpt-5.3-codex achieves the best overall performance (solving 19/22 tasks, 86.4%), outperforming claude-opus-4.6 (15/22, 68.2%), and kimi-2.5 exhibits the strongest performance among open-source models. Performance degrades sharply with increasing task difficulty, particularly on hard, specification-intensive systems. Behavioral analysis further reveals that as codebases scale, code reading, rather than writing, becomes the dominant bottleneck in AI-assisted development. Overall, while specification-driven autonomous software engineering is increasingly viable, substantial challenges remain before it can reliably support production-scale development.

The rise of large language models (LLMs) has introduced a new type of programming: natural language programming. By writing prompts that direct LLMs to perform natural language processing, code generation, reasoning, etc., users are writing code in natural language -- natural language code -- for the LLM to execute. An emerging area of research enables interoperability between natural language code and formal languages such as Python. We present a novel programming abstraction, shared program state, that removes the manual work required to enable interoperability between natural language code and program state. With shared program state, programmers can write natural code that directly writes program variables, computes with program objects, and implements control flow in the program. We present a schema for specifying natural function interfaces that extend programming systems to support natural code and leverage this schema to specify shared program state as a natural function interface. We implement shared program state in the Nightjar programming system. Nightjar enables programmers to write Python programs that contain natural code that shares the Python program state. We show that Nightjar programs achieve comparable or higher task accuracy than manually written implementations (+4-19%), while decreasing the lines of code by 39.6% on average. The tradeoff to using Nightjar is that it may incur runtime overhead (0.4-4.3x runtime of manual implementations).

This document consolidates publicly reported technical details about Metas Llama 4 model family. It summarizes (i) released variants (Scout and Maverick) and the broader herd context including the previewed Behemoth teacher model, (ii) architectural characteristics beyond a high-level MoE description covering routed/shared-expert structure, early-fusion multimodality, and long-context design elements reported for Scout (iRoPE and length generalization strategies), (iii) training disclosures spanning pre-training, mid-training for long-context extension, and post-training methodology (lightweight SFT, online RL, and lightweight DPO) as described in release materials, (iv) developer-reported benchmark results for both base and instruction-tuned checkpoints, and (v) practical deployment constraints observed across major serving environments, including provider-specific context limits and quantization packaging. The manuscript also summarizes licensing obligations relevant to redistribution and derivative naming, and reviews publicly described safeguards and evaluation practices. The goal is to provide a compact technical reference for researchers and practitioners who need precise, source-backed facts about Llama 4.

We present CoTexT, a pre-trained, transformer-based encoder-decoder model that learns the representative context between natural language (NL) and programming language (PL). Using self-supervision, CoTexT is pre-trained on large programming language corpora to learn a general understanding of language and code. CoTexT supports downstream NL-PL tasks such as code summarizing/documentation, code generation, defect detection, and code debugging. We train CoTexT on different combinations of available PL corpus including both "bimodal" and "unimodal" data. Here, bimodal data is the combination of text and corresponding code snippets, whereas unimodal data is merely code snippets. We first evaluate CoTexT with multi-task learning: we perform Code Summarization on 6 different programming languages and Code Refinement on both small and medium size featured in the CodeXGLUE dataset. We further conduct extensive experiments to investigate CoTexT on other tasks within the CodeXGlue dataset, including Code Generation and Defect Detection. We consistently achieve SOTA results in these tasks, demonstrating the versatility of our models.

A coding agent works on repositories that can be far larger than its context window can hold. Prompt compressors and retrievers offer solutions to this problem, but patching an issue needs two things at once: the files that must change, and their code exactly as written, since the model must reproduce the lines it edits. Retrievers rank code by its similarity to the issue rather than by whether it must change, and compressors delete text to fit the budget. We show that at an equal token budget, an agent resolves more issues when the budget goes to choosing which files to show verbatim, as our simple selector CQ2-CARE does, than when it goes to pruning tokens or retrieving chunks. We fit every context to the same per-issue budget on 300 recent SWE-rebench issues and grade each patch with the benchmark's tests. With GPT-5.6 and Claude Opus 5, CQ2-CARE resolves 43.7% and 46.7% of the issues, 6.3 points more than the best of three compressors with either model, and 9.3 and 12.7 points more than the best of three retrievers. CQ2-CARE even resolves more issues than two methods trained for issue resolution, SweRank (by 0.7 and 1.7 points) and SWE-Pruner (by 2.0 and 7.3). Our analysis shows the defects of the baseline methods: retrievers often miss the files to edit, and compressors keep them but alter their text, so fewer patches apply. Exact-file selection needs no compression model and shows every line as the repository has it, so the model can quote the code it edits.

Agentic coding tools receive goals written in natural language as input, break them down into specific tasks, and write or execute the actual code with minimal human intervention. Central to this process are agent context files ("READMEs for agents") that provide persistent, project-level instructions. In this paper, we conduct the first large-scale empirical study of 2,303 agent context files from 1,925 repositories to characterize their structure, maintenance, and content. We find that these files are not static documentation but complex, difficult-to-read artifacts that evolve like configuration code, maintained through frequent, small additions. Our content analysis of 16 instruction types shows that developers prioritize functional context, such as build and run commands (62.3%), implementation details (69.9%), and architecture (67.7%). We also identify a significant gap: non-functional requirements like security (14.5%) and performance (14.5%) are rarely specified. These findings indicate that while developers use context files to make agents functional, they provide few guardrails to ensure that agent-written code is secure or performant, highlighting the need for improved tooling and practices.

Large Language Models (LLMs) have demonstrated their remarkable capabilities in numerous fields. This survey focuses on how LLMs empower users, regardless of their technical background, to use human languages to automatically generate executable code. We begin with understanding LLMs' limitations and challenges in automated code generation. Subsequently, we review various fine-tuning techniques designed to enhance both the performance and adaptability of LLMs in code generation tasks. We then review the existing metrics and benchmarks for evaluations to assess model performance based on fine-tuning techniques. Finally, we explore the applications of LLMs (e.g. CodeLlama, GitHub Copilot, ToolGen) in code generation tasks to illustrate their roles and functionalities. This survey provides a comprehensive overview of LLMs for code generation, helps researchers in diverse fields better understand the current state-of-the-art technologies, and offers the potential of effectively leveraging LLMs for code generation tasks.

Computer architecture has long used benchmarks to make progress measurable. LLM agents create a different measurement problem: success is not merely writing code or tuning parameters. The agent must interpret workloads, choose mechanisms, use simulators, predict performance, satisfy hard constraints, and decide which feasible design is worth evaluating. This paper introduces ArchEval, a benchmark and platform for evaluating LLM agents on computer architecture design and optimization. It contains 20 challenges across CPU core mechanisms, system architecture, memory systems, accelerators, and compute-in-memory, backed by eight simulators. Each challenge is posed under three settings: L1 full harness, with repeated simulator feedback; L2 simulator-code container, where simulator source is available but the agent must assemble its own workflow; and L3 agent-only, with no runnable feedback before submission. Each run reports baseline-normalized verifier performance and records the full trajectory, connecting results to workload analysis, simulator-tool use, prediction, constraint handling, and artifact integrity. Initial results show a sharp boundary in current agents. With L1 support, all four evaluated agents reach or exceed baseline and improve real designs across diverse simulators. Removing support exposes weaknesses: many agents fail to turn simulator source into useful experiments, and L3 predictions often disagree with verifier results. In L3, only GPT-5.5 + Codex remains above baseline, reaching 1.21x geomean performance and a 65% win rate; the other three fall below baseline. Even GPT-5.5 + Codex has only a 15% performance-modeling pass rate. ArchEval frames today's agents as useful optimization assistants rather than autonomous architects, and identifies capabilities needed next: simulator-tool use, calibrated prediction, pre-feedback judgment, and useful mechanism discovery.

Natural Language to Code Generation has made significant progress in recent years with the advent of Large Language Models(LLMs). While generation for general-purpose languages like C, C++, and Python has improved significantly, LLMs struggle with custom function names in Domain Specific Languages or DSLs. This leads to higher hallucination rates and syntax errors, specially for DSLs having a high number of custom function names. Additionally, constant updates to function names add to the challenge as LLMs need to stay up-to-date. In this paper, we present optimizations for using Retrieval Augmented Generation (or RAG) with LLMs for DSL generation along with an ablation study comparing these strategies. We generated a train as well as test dataset with a DSL to represent automation tasks across roughly 700 APIs in public domain. We used the training dataset to fine-tune a Codex model for this DSL. Our results showed that the fine-tuned model scored the best on code similarity metric. With our RAG optimizations, we achieved parity for similarity metric. The compilation rate, however, showed that both the models still got the syntax wrong many times, with RAG-based method being 2 pts better. Conversely, hallucination rate for RAG model lagged by 1 pt for API names and by 2 pts for API parameter keys. We conclude that an optimized RAG model can match the quality of fine-tuned models and offer advantages for new, unseen APIs.

Generative AI coding agents are entering the open-source supply chain, yet their diverse and often invisible traces leave their prevalence poorly understood. We introduce a multi-layered detection framework that integrates configuration-file scanning, commit-message analysis, author-identity matching, and bot-signature lookup across World of Code (180M+ Git repositories), classifying agent traces into four behavioral types. No single method captures more than a fraction of activity: multi-method detection identifies 850,157 Claude Code commits in one snapshot, of which bot-account lookup_the signal most adoption studies rely on_recovers only 28,154 (3.3%), a 30x relative-recall gap, so single-signal prevalence estimates are biased low by at least this factor. Every detection pattern is hand-validated (495 labels) with per-cell precision and Wilson confidence intervals. Across snapshots from December 2024 to April 2026, commit-attributed agents generate over 320,000 commits per month; Claude Code leads (886,122 commits across 17,295 projects) and dominates silent, configuration-file-only adoption (21,078 projects). Compared against an independent pull-request census (AIDev), the two channels capture nearly disjoint agent populations_a PR census misses 79% of commit-detected Claude Code adopters and essentially all Codex adopters_and different kinds of work: PR-deployed cloud agents (Codex, Cursor) surface as feature work, while commit-deployed in-editor agents (Claude Code, OpenHands, Aider) surface as maintenance. The observed work profile follows deployment and detection mode rather than the tool itself, so no single channel is representative.

Large Language Model (LLM) coding agents have achieved strong results on software engineering tasks, yet repository exploration remains a major bottleneck: locating relevant code consumes substantial token budget and pollutes the agent's context with irrelevant snippets. In most agents, the same model explores the repository and solves the task, leaving exploratory reads and searches in the solver's history. We present FastContext, a dedicated exploration subagent that separates repository exploration from solving. Invoked on demand, FastContext issues parallel tool calls and returns concise file paths and line ranges as focused context. FastContext is powered by specialized exploration models spanning 4B--30B parameters. We bootstrap them from strong reference-model trajectories and refine them with task-grounded rewards for broad first-turn search, multi-turn evidence gathering, and precise citation generation. Across SWE-bench Multilingual, SWE-bench Pro, and SWE-QA, integrating FastContext into Mini-SWE-Agent improves end-to-end resolution rates up to 5.5\% while reducing coding-agent token consumption up to 60\%, with marginal overhead. These results show that repository exploration can be separated from solving and handled effectively by specialized models. Code and data: https://github.com/microsoft/fastcontext

Coding agents re-send large file reads and tool outputs to a frontier LLM every turn, and this context dominates their token bill. General-purpose prompt compressors are trained on prose and suit code poorly: they paraphrase identifiers and drop the exact spans an agent needs to edit. We present Paritok-4B, a 4B LoRA compressor for coding-agent trajectories built on two commitments. It is extractive: it selects spans rather than rewriting them, and 96.0% of the identifiers, paths, and numbers it emits already appear in its input, holding at 96.2% on held-out SWE-bench Lite output. It is intent-conditioned: told the agent's current task, it acts chiefly inside a retained segment, selecting which lines survive (retained lines are +0.067 more intent-relevant than removed ones, paired 95% CI [+0.056, +0.078]) rather than changing how much is retained. We distil a gpt-4.1-mini teacher over 67,074 real OpenHands trajectories into 40,606 validated examples and fine-tune Qwen3-4B. On all 300 SWE-bench Lite instances, Paritok-4B compresses agent context to 25.7% of its size, 2.0x harder than a gpt-4.1-mini compressor (50.2%) and 2.4x harder than gpt-5 (61.9%), while retaining 86.5% of uncompressed single-shot solve quality. Fed the cat -n line-numbered input real agents produce, it compresses slightly less (27.8%) and retains more (89.3%); there the paired test is informative, with 30 instances solved only uncompressed and 17 only compressed, an exact McNemar p=0.079, so at this sample size compressing context to roughly a quarter of its size does not significantly reduce the solve rate. The model is a 264 MB adapter that self-hosts on one 24 GB GPU with no per-token compressor fee, which at list prices decides the economics: gpt-5 as a compressor is net-negative, costing more than the downstream tokens it saves. Weights, data, and evaluation scripts are open (Apache 2.0).

Large Language Models (LLMs) have shown promise in automated code generation but typically excel only in simpler tasks such as generating standalone code units. Real-world software development, however, often involves complex code repositories (named repo) with complex dependencies and extensive documentation. To fill this gap, our research pivots towards evaluating LLMs in a more realistic setting -- real-world repo-level code generation. We introduce CodeAgentBench, a manually curated benchmark for repo-level code generation. This benchmark comprises five high-quality Python projects, encompassing a total of 101 samples. We assess nine leading LLMs on repo-level tasks and observe a decline in their performance. To tackle this, we present CodeAgent, a novel LLM-based agent framework that employs external tools for effective repo-level code generation. CodeAgent integrates five programming tools, enabling interaction with software artifacts for information retrieval, code symbol navigation, and code testing. We implement four agent strategies to optimize these tools' usage. Our experiments on CodeAgentBench show that CodeAgent enhances LLM performance significantly, with improvements ranging from 18.1\% to 250\%. Further tests on the HumanEval benchmark confirm CodeAgent's adaptability and efficacy across various code generation tasks. Notably, CodeAgent outperforms commercial products like Github Copilot, showcasing superior accuracy and efficiency. These results demonstrate CodeAgent's robust capabilities in code generation, highlighting its potential for real-world repo-level coding challenges.

Large Language Models (LLMs) have demonstrated remarkable proficiency across a variety of complex tasks. One significant application of LLMs is in tackling software engineering challenges, particularly in resolving real-world tasks on GitHub by fixing code based on the issues reported by the users. However, many current approaches rely on proprietary LLMs, which limits reproducibility, accessibility, and transparency. The critical components of LLMs for addressing software engineering issues and how their capabilities can be effectively enhanced remain unclear. To address these challenges, we introduce SWE-Fixer, a novel open-source LLM designed to effectively and efficiently resolve GitHub issues. SWE-Fixer comprises two essential modules: a code file retrieval module and a code editing module. The retrieval module employs BM25 along with a lightweight LLM model to achieve coarse-to-fine file retrieval. Subsequently, the code editing module utilizes the other LLM model to generate patches for the identified files. Then, to mitigate the lack of publicly available datasets, we compile an extensive dataset that includes 110K GitHub issues along with their corresponding patches, and train the two modules of SWE-Fixer separately. We assess our approach on the SWE-Bench Lite and Verified benchmarks, achieving state-of-the-art performance among open-source models with scores of 23.3% and 30.2%, respectively. These outcomes highlight the efficacy of our approach. We will make our model, dataset, and code publicly available at https://github.com/InternLM/SWE-Fixer.

Coding agents can be guided with instruction files in a repository. We examine this emerging practice through a census of popular, active GitHub repositories in thirteen languages, treating each repository's root AGENTS.md and CLAUDE.md as one instruction set. Agent instructions appear in of repositories and of the most-starred; where both files exist, CLAUDE.md usually points to AGENTS.md. We analyse statements using model-assisted annotation and human review. The annotations reveal a practical mix of project knowledge and rules about code, building and workflow. Collaboration rules name concrete requirements, from test plans to commit formats, yet of directives give no reason. Projects disagree on AI attribution: commit-message rules more often prohibit it than require it, while pull-request rules more often require disclosure. A third of instruction sets have a generator-associated signature, a tool-inserted block, shared text or an AI-marked creation commit; four in five are edited after creation. Prespecified tests associate repository popularity with rules on pull requests and AI contributions. Exploratory analyses link activity to process-oriented content and show that LLM and agent projects are over-represented among adopters. The models agree much more closely with each other than with the human reviewer's topic assignments, a caution for automated annotation. We release the annotations, taxonomies and pointers to the analysed file versions.

Recent advancements in Large Language Models (LLMs) and their utilization in code generation tasks have significantly reshaped the field of software development. Despite the remarkable efficacy of code completion solutions in mainstream programming languages, their performance lags when applied to less ubiquitous formats such as OpenAPI definitions. This study evaluates the OpenAPI completion performance of GitHub Copilot, a prevalent commercial code completion tool, and proposes a set of task-specific optimizations leveraging Meta's open-source model Code Llama. A semantics-aware OpenAPI completion benchmark proposed in this research is used to perform a series of experiments through which the impact of various prompt-engineering and fine-tuning techniques on the Code Llama model's performance is analyzed. The fine-tuned Code Llama model reaches a peak correctness improvement of 55.2% over GitHub Copilot despite utilizing 25 times fewer parameters than the commercial solution's underlying Codex model. Additionally, this research proposes an enhancement to a widely used code infilling training technique, addressing the issue of underperformance when the model is prompted with context sizes smaller than those used during training. The dataset, the benchmark, and the model fine-tuning code are made publicly available.

The Bourne-Again Shell (Bash) command-line interface for Linux systems has complex syntax and requires extensive specialized knowledge. Using the natural language to Bash command (NL2SH) translation capabilities of large language models (LLMs) for command composition circumvents these issues. However, the NL2SH performance of LLMs is difficult to assess due to inaccurate test data and unreliable heuristics for determining the functional equivalence of Bash commands. We present a manually verified test dataset of 600 instruction-command pairs and a training dataset of 40,939 pairs, increasing the size of previous datasets by 441% and 135%, respectively. Further, we present a novel functional equivalence heuristic that combines command execution with LLM evaluation of command outputs. Our heuristic can determine the functional equivalence of two Bash commands with 95% confidence, a 16% increase over previous heuristics. Evaluation of popular LLMs using our test dataset and heuristic demonstrates that parsing, in-context learning, in-weight learning, and constrained decoding can improve NL2SH accuracy by up to 32%. Our findings emphasize the importance of dataset quality, execution-based evaluation and translation method for advancing NL2SH translation. Our code is available at https://github.com/westenfelder/NL2SH

Information localization within massive repositories is a cornerstone of agentic LLM systems. While synthetic data-driven optimization has proven successful in training LLMs, little attention has been paid to optimizing the agent's working environment (the repository itself) in a data-driven manner. To bridge this gap, we present Libra, a self-evolving framework that introduces mutable "catalogs" (hierarchical Markdown files serving as navigable indices) into the repository. Libra runs an LLM-driven optimization loop where a Prompter generates synthetic queries, a frozen Solver attempts to resolve them by navigating the catalogs, and a Healer rewrites the catalogs in response to the Solver's localization failures. Evaluations across 12 SWE-bench Lite repositories demonstrate that this environmental healing yields continual, logarithmic improvements in code localization accuracy. Furthermore, these environmental improvements transfer zero-shot across different LLMs and problem sets. Although the focus of this paper is to study the general behavior of such a system, we also demonstrate that a minimalist coding agent equipped with Libra-optimized catalogs outperforms state-of-the-art baselines. Code is available at https://github.com/salesforce-misc/Libra and data at https://huggingface.co/datasets/Salesforce/Libra.

Vision language models are serving as general-purpose interfaces for complex multimodal tasks. However, deployment still faces three gaps: VLMs typically incur high latency and cost when processing dense video frames and long prompts, the agent scaffold remains static after deployment, and standard video-QA benchmarks do not test whether agents can use visual evidence inside tool-using workspaces. We present VisualClaw, a self-evolving multimodal agent built around two principles. First, hybrid encoding reduces deployment cost by filtering less informative streaming frames with a cascaded gate and compressing the text skill bank through hot/cold top-k injection. Second, skill evolution lets the agent learn from failures: retrieved memories condition an evolver as direct concatenated context or as guided evidence, producing skill-bank updates that help future questions. Across 4 video-QA benchmarks with 2 VLMs, VisualClaw cuts per-question API cost by an average -98% versus full-frame upload and by -25.9% over the offline uniform 8 frame baseline, while boosting accuracy in most settings, e.g., an average +3.85% and a peak +15.80% on EgoSchema with Gemini 3 Flash. To address the gap, we curate VisualClawArena, a 200-scenario multimodal agentic benchmark built through a strict five-stage pipeline; models must use video evidence, documents, dynamic updates, and executable checks inside a workspace. On VisualClawArena, the same framework with computer-use agent backends improves macro accuracy by +2.9% for Codex (GPT-5.5) and +3.2% for Claude Code (Sonnet 4.6) over no-evolution baselines, with a -9.5% cost reduction compared to the uniform-sampled baseline. These properties make VisualClaw a natural fit for edge applications, where the cascade reduces a 1-hour streaming session from ~3,600 API uploads down to only 5-20 calls and the self-evolution makes it a perfect personalized assistant.

We present CodeNav, an LLM agent that navigates and leverages previously unseen code repositories to solve user queries. In contrast to tool-use LLM agents that require ``registration'' of all relevant tools via manual descriptions within the LLM context, CodeNav automatically indexes and searches over code blocks in the target codebase, finds relevant code snippets, imports them, and uses them to iteratively generate a solution with execution feedback. To highlight the core-capabilities of CodeNav, we first showcase three case studies where we use CodeNav for solving complex user queries using three diverse codebases. Next, on three benchmarks, we quantitatively compare the effectiveness of code-use (which only has access to the target codebase) to tool-use (which has privileged access to all tool names and descriptions). Finally, we study the effect of varying kinds of tool and library descriptions on code-use performance, as well as investigate the advantage of the agent seeing source code as opposed to natural descriptions of code. All code will be made open source under a permissive license.

Large language model (LLM) agents increasingly operate over evolving software repositories, where success depends on repository-specific procedural knowledge: which APIs to call, which scripts to run, and which conventions the current release expects. Agent skills externalize this knowledge into reusable units, and prior work shows that they can improve agent performance. What remains unclear is whether that improvement is durable. The same version specificity that makes a skill useful also makes it fragile: after a release, it may become stale without raising any explicit signal, while continuing to provide obsolete guidance. Externalizing knowledge into a skill can therefore make its decay invisible. We study whether agents can keep this externalized knowledge current. Repo2Skill-Evo casts each release transition as a skill-maintenance task: given a V1 skill set and the official V1-to-V2 patch, an agent must update obsolete skill content while preserving guidance that remains valid. Across 57 real-world repositories and 105 selected release transitions, every evaluated transition invalidates part of the V1 skill set. Yet six frontier agents reach only 29.9%-69.7% avg@3 macro F1 under a patch-grounded removal metric that balances stale-content recall against over-editing precision. Across runs, two opposing errors dominate: incomplete coverage of affected files in the skill set leaves stale content untouched, while overbroad editing is associated with higher recall but lower precision. Repository skills go stale in silence, and even frontier agents cannot reliably maintain them.