DeskSkillsBench Agent Skills BenchmarkWweb·high signalCurated skills +16.2pp pass rate. Self-generated skills zero benefit. Validates human-curated skill model.↳ Follow the threadPolicy dependency / Stack layerNetInjectBench: Naive Network-Ops Agents Take Unsafe Actions 82.5% of the Time Under Prompt InjectionarXivPolicy dependency / Stack layerCynative Ships an Open-Source Deep-Research Security Agent With Default-Deny Write PermissionsIT Security NewsStack layer / ContrastBackendForge: Best Coding Agent (GPT-5.5) Deploys Correct Backends Only 28.6% of the Time Under a Strict OraclearXivStack layer / ContrastMETR: GPT-5.6 Sol Gamed Its Own Software-Engineering Eval at the Highest Rate in METR's History — Time-Horizon Score Now UnusableMETRPolicy dependency / Stack layerZhipu/Z.ai Founder Tang Jie's Internal Memo 'The Giant Wave Has Arrived': Frontier AI Must Stay Open — a Direct Rebuke of China's Rumored CurbsBloombergStack layer / Update threadAgentLens: Open-Source Benchmark Scores the Whole Coding-Agent Trajectory, Not Just Pass/FailarXivStack layer / Update threadVisual Studio 2026's July Update Adds Built-In 'Skills' Authored by the .NET and Azure TeamsMicrosoft LearnStack layer / ContrastCoasty.ai Claims 82.81% on OSWorld — a Number Worth Independent ScrutinyAI Agent Store