ResearchOPENDEV: Open-Source Terminal Coding Agent BlueprintarXiv·high signalXBlueskyLinkedInCopy linkTwo-phase scaffolding/harness architecture solving context bloat, destructive execution, and capability extension. Directly applicable blueprint for CLI agent builders.SourceSource pagearXiv↳ Follow the threadPolicy dependency / Stack layerA Cheap Read-Only Verifier Captures Nearly All the False-Pass Benefit of a Full Planning StackarXiv 2609.20474Stack layer / Threat patternJames Mickens argues chain-of-thought monitoring can never be a sound security controlarXivStack layer / ContrastMAGS routes coding-agent output through Dafny and reports 100% success at producing verified programs on 220 tasksarXivStack layer / ContrastAn Auto-Optimized Agent Harness Can Cheat the Whole Benchmark, Not Just the TasksarXiv 2609.18366Stack layer / ContrastHarness-Layer Auto-Research Cut Agent Token Traffic 44.7-49.0% at Equal Task PerformancearXiv 2609.20519Stack layer / Update threadActObs: supervising environment observations during SFT changes how agents explore under RL, +3.4pp pass@16 on Terminal-Bench 2.0arXiv / HuggingFace Daily PapersStack layerMulti-Agent Collaboration Pays Only on Long-Horizon Tasks With Sparse DependenciesarXiv 2609.19759Stack layerAn Agent Meant to Run for Weeks Needs Levels, Ticks and Escalation, All in the HarnessarXiv 2609.19519