Learning When and How to Intervene: A Hindsight-Distilled Sentinel for Coding Agents
cs.SE, cs.AI, cs.LG
Submitted: 2026-09-30
Updated: 2026-09-30
Code: https://github.com/mariushobbhahn/SWEBench-verified-mini
Sources
- Open-SWE-Traces: Advancing Dual-Mode Multilingual Distillation for Software Engineering Agents
- AgentRx: Diagnosing AI Agent Failures from Execution Trajectories
- SWE-chat: Coding Agent Interactions From Real Users in the Wild
- DualOPSD: Closing the Feedback Loop in On-Policy Self-Distillation
- Beyond Final Code: A Process-Oriented Error Analysis of Software Development Agents in Real-World GitHub Scenarios
- dOPSD: On-Policy Self-Distillation for Diffusion Language Models
- TRAIL: Trace Reasoning and Agentic Issue Localization
- Ask or Assume? Uncertainty-Aware Clarification-Seeking in Coding Agents
- Privileged Foresight Distillation: Zero-Cost Future Correction for World Action Models
- When Agents go Astray: Course-Correcting SWE Agents with PRMs
- Steer, Don't Solve: Training Small Critic Models for Large Code Agents
- GenCLS++: Pushing the Boundaries of Generative Classification in LLMs Through Comprehensive SFT and RL Studies Across Diverse Datasets
- Rethinking On-Policy Self-Distillation for Thinking Models
- From SWE-ZERO to SWE-HERO: Execution-free to Execution-based Fine-tuning for Software Engineering Agents
- Understanding Code Agent Behaviour: An Empirical Study of Success and Failure Trajectories
- Privileged Information Distillation for Language Models
- TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories
- FALAT: Tracing Failures in LLM Agent Trajectories via Dependency-Guided Search
- Devstral: Fine-tuning Language Models for Coding Agent Applications
Related papers
- Falsification-Based Verification of LLM-Generated Optimization Models: Sound Test Batteries and Their Detection Limits
- GitSkills: A Dataset of Agent Skills on GitHub
- SABER: Benchmarking Operational Safety of LLM Coding Agents in Stateful Project Workspaces
- PackMonitor: Enabling Zero Package Hallucinations Through Decoding-Time Monitoring
- IntentCoding: Amplifying User Intent in Code Generation
- Incentives and Outcomes in Bug Bounties