ResearchAgent Skills in the Wild 42447 Skills Audited 26% VulnerablearXiv·high signalXBlueskyLinkedInCopy linkLargest MCP/skill security study. 42,447 skills, 26.1% contain vulnerabilities across 14 patterns. SkillScan detection framework.SourceSource pagearXiv↳ Follow the threadStack layer / Threat patternPre-registered ablation shows removing an LLM verifier stage from an offensive-security agent shifts median reported findings from 0 to 2 per runarXivPolicy dependency / Stack layerPython's Import Statement Is an Execution Boundary: 90% of Initialization-Activated Advisory Vulnerabilities Are High or CriticalarXiv 2609.14791Stack layer / Threat patternRemoving the Tenant ID From an MCP Tool Schema Blocks Cross-Tenant Reads That a Validated Parameter Let Through 26 Times Out of 26arXiv 2609.14780Stack layer / Threat patternTool-Augmented Agents Fabricate Values 45.3% of the Time When a Tool Returns status:ok With an Unusable PayloadarXiv 2609.14758Stack layer / Threat patternTRAIL pairs a translator agent against a challenger agent and gains 23.1% relative syntax accuracy on C-to-Rust translationarXivStack layer / Threat patternEricsson's Multi-Agent Code Reviewer Hits 96% Accuracy, With a Third of Its Correct Findings Rated Must-FixarXiv 2609.15877Stack layer / Threat patternPlanting Benign-Sounding Reasoning in an Agent's Context Evades Chain-of-Thought Monitors 25-33% of the TimearXiv 2609.15989Policy dependency / Stack layerRIPPLE: an edit confined to one prompt-policy segment changes downstream behavior, so replay candidate edits after previously accepted ones before persistingarXiv 2609.12127