✨SWE-RM: Execution-free Feedback For Software Engineering Agents

📝 Summary:
This paper introduces SWE-RM, a robust, execution-free reward model for software engineering agents. It overcomes limitations of execution-based feedback, improving coding agent performance in both test-time scaling and reinforcement learning. SWE-RM achieves new state-of-the-art results for open...

🔹 Publication Date: Published on Dec 26

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2512.21919
• PDF: https://arxiv.org/pdf/2512.21919

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#SoftwareEngineering #AI #ReinforcementLearning #CodingAgents #RewardModels

✨ISO-Bench: Can Coding Agents Optimize Real-World Inference Workloads?

📝 Summary:
ISO-Bench evaluates coding agents on real-world LLM inference optimization tasks using combined execution and LLM metrics. Agents often identify bottlenecks but fail to execute working solutions, highlighting that scaffolding is as important as the model itself.

🔹 Publication Date: Published on Feb 23

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2602.19594
• PDF: https://arxiv.org/pdf/2602.19594

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#CodingAgents #LLMOptimization #AIResearch #Benchmarking #LargeLanguageModels

✨Squeez: Task-Conditioned Tool-Output Pruning for Coding Agents

📝 Summary:
A task-conditioned tool-output pruning model effectively reduces input tokens for coding agents. It achieves 0.86 recall and 0.80 F1, removing 92% of tokens, outperforming larger zero-shot models and heuristic baselines.

🔹 Publication Date: Published on Apr 4

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.04979
• PDF: https://arxiv.org/pdf/2604.04979
• Github: https://github.com/KRLabsOrg/squeez

🔹 Models citing this paper:
• https://huggingface.co/KRLabsOrg/squeez-2b

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#CodingAgents #LLM #TokenPruning #AI #MachineLearning

✨Scaling Test-Time Compute for Agentic Coding

📝 Summary:
This framework improves long-horizon agentic coding by using compact trajectory representations for test-time scaling. It employs Recursive Tournament Voting and adapted Parallel-Distill-Refine to significantly boost coding agent performance on benchmarks.

🔹 Publication Date: Published on Apr 16

🔹 Paper Links:
• arXiv Page: https://arxiv.org/abs/2604.16529
• PDF: https://arxiv.org/pdf/2604.16529

==================================

For more data science resources:
✓ https://xn--r1a.website/DataScienceT

#AgenticAI #CodingAgents #MachineLearning #AIResearch #DeepLearning

🔥 FastContext: Training Efficient Repository Explorer for Coding Agents

💡 The paper introduces FastContext, a dedicated exploration subagent designed to improve the efficiency of repository exploration in large language model coding agents. The problem addressed is that repository exploration is a major bottleneck in coding agents, consuming a substantial token budget and polluting the agent's context with irrelevant code snippets.

The method involves separating repository exploration from code solving using specialized exploration models. FastContext is invoked on demand and issues parallel tool calls to return concise file paths and line ranges as focused context. The exploration models used in FastContext are powered by 4B-30B parameters and are bootstrapped from strong reference-model trajectories. They are then refined with task-grounded rewards for broad first-turn search, multi-turn evidence gathering, and precise citation generation.

The results show that integrating FastContext into a coding agent improves end-to-end resolution rates by up to 5.5 percent while reducing coding-agent token consumption by up to 60 percent, with minimal overhead. The paper demonstrates that repository exploration can be effectively handled by specialized models, separate from the code solving process. The code and data for FastContext are made available, allowing for further research and development in this area. Overall, the paper presents a significant contribution to the field of coding agents and software engineering, providing a more efficient and effective approach to repository exploration.

📅 Published on Jun 12

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2606.14066
• PDF: https://arxiv.org/pdf/2606.14066
• Project Page: https://huggingface.co/microsoft/FastContext-1.0-4B-SFT

🤖 Models citing this paper:
• https://huggingface.co/microsoft/FastContext-1.0-4B-SFT
• https://huggingface.co/microsoft/FastContext-1.0-4B-RL

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#EfficientRepositoryExploration #CodingAgents #LargeLanguageModels #RepositoryExplorationSubagents #SpecializedExplorationModels

🔥 CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents

💡 The paper introduces CodeNib, a multi-view data system designed to provide repository context to coding agents. The problem addressed is that coding agents repeatedly search, navigate, and retain context from evolving repositories, but disconnected indexes, language servers, and task-local histories force repeated discovery and obscure life cycle costs. CodeNib builds reusable lexical, dense, and structural views per repository commit, maps outputs to repository-relative source ranges, maintains selected views across edits, and serves ranked search, symbol navigation, and bounded context through one runtime.

The method involves creating a system that can efficiently serve context to coding agents by building and maintaining multiple views of the repository. The system is evaluated across 100 snapshots, mapping quality-cost frontiers across the repository-context life cycle. The results show that when outputs match an independent rebuild, graph and vector updates are 8.7 times and 25.4 times faster at the median.

The system is also evaluated on a static-navigation subset matching normalized live-server locations, where the median per-request live/static latency ratio is 4.7 times. Additionally, the results show that selected context policies preserve localization with 50-87 percent fewer trajectory tokens than paired grep/read. Overall, the results support multi-view repository-context serving with explicit, operation-specific validity boundaries, providing a solution to the problem of efficiently serving context to coding agents.

📅 Published on Jul 28

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2607.25431
• PDF: https://arxiv.org/pdf/2607.25431
• Project Page: https://codenib.ai

📊 Datasets citing this paper:
• https://huggingface.co/datasets/sysevol-ai/codenib-synthesis

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#CodeAnalysisTools #MultiViewDataSystems #CodingAgents #RepositoryContext #SoftwareDevelopmentTools

🔥 Ouroboros: A Self-Developing Frontier Coding Agent with Reviewed Core Evolution

💡 The paper presents Ouroboros, a self-developing coding agent that improves its tools, prompts, context assembly, and core implementation through reviewed commits that become the runtime for later work. The core evolution proceeds in two modes: recursive free evolution, where improvement is itself a task, and experience-driven core evolution, where ordinary work and social interaction expose bugs, rough edges, and inefficient context construction that lead to reviewed structural changes.

The agent's self-developing nature allows it to rewrite its own code and select new model APIs, making operational safety a primary design problem. To address this, the authors implement guardrails that remain authoritative under evolutionary and public social pressure.

The paper evaluates Ouroboros on several benchmarks, including Terminal-Bench 2.1, OS World-Verified, and CL-Bench. The results show that Ouroboros achieves state-of-the-art performance, with an Opus 5 run scoring 86.74% on Terminal-Bench 2.1, exceeding the best previously reported score, and reaching 90.69% on OS World-Verified. A five-rollout CL-Bench campaign achieves a normalized reward of 0.2301, setting a new state of the art.

The paper also presents Hope, a 161-day living agent experiment in free evolution under governed human communication across seven surfaces. Human interaction surfaces fault and generate proposals, but the agent decides which changes to pursue. The results demonstrate the effectiveness of Ouroboros in self-developing and improving its performance over time.

Overall, the paper contributes to the development of self-developing agents that can improve their performance through reviewed commits and evolutionary changes, while addressing the design problem of operational safety. The results demonstrate the potential of Ouroboros to achieve state-of-the-art performance on various benchmarks and to continue improving over time through live evolution.

📅 Published on Aug 8

🔗 Links:
• GitHub: https://github.com/huggingface
• arXiv: https://arxiv.org/abs/2608.08311
• PDF: https://arxiv.org/pdf/2608.08311
• Project Page: https://ouroboros-agent.ai

📊 Datasets citing this paper:
• https://huggingface.co/datasets/razzant/swepro-luna-matched-pair
• https://huggingface.co/datasets/razzant/ouroboros-osworld-verified-sonnet46
• https://huggingface.co/datasets/razzant/ouroboros-osworld-verified-opus5

🚀 Spaces citing this paper:
• https://huggingface.co/spaces/razzant/ouroboros-benchmark-explorer

━━━━━━━━━━━━━━━━━━━━━━━━
📢 By: https://xn--r1a.website/PaperNexus

#ArtificialIntelligence #SelfModifyingCode #AutonomousAgents #EvolutionaryComputation #CodingAgents