ResearchAgentic Code Reasoning Semi-Formal Reasoning 88 Percent AccuracyarXiv·high signalStructured prompting for code reasoning without execution. 88% accuracy on patch verification, 93% on real-world patches. Verifiable certificates.SourceSource pagearXiv↳ Follow the threadPolicy dependency / Stack layerNetInjectBench: Naive Network-Ops Agents Take Unsafe Actions 82.5% of the Time Under Prompt InjectionarXivStack layer / Threat patternMako: A 'Self-Evolving Agentic OS' That Writes New Exploits Against Live Targets and Hot-Loads Them Into ItselfarXivStack layer / ContrastBackendForge: Best Coding Agent (GPT-5.5) Deploys Correct Backends Only 28.6% of the Time Under a Strict OraclearXivPolicy dependency / Stack layerCynative Ships an Open-Source Deep-Research Security Agent With Default-Deny Write PermissionsIT Security NewsStack layer / Threat patternStudy of 44 Developers: AI Code Assistants Barely Improve Secure API UsagearXivStack layer / Update threadAgentLens: Open-Source Benchmark Scores the Whole Coding-Agent Trajectory, Not Just Pass/FailarXivStack layer / ContrastEmpirical Study: Microsoft's Early-2026 Rollout of Claude Code and GitHub Copilot CLIarXiv / Hacker NewsStack layer / Follow-up threadSurvey: Metacognition in LLMs — foundations, progress, and open problems for self-monitoring agentsarXiv