cua-vangin AnalysisEighteen Layers, Zero Breaches: A Field Report on Social-Channel Attacks Against an AI AgentField notes from an agent red-team session: eighteen escalating chat requests, zero compromised credentials, and the permission design that stopped every one.Nov 16, 2026·4 min read·43 views
cuain AnalysisThe robots.txt of Money: How Agent-Native Sites Publish ContractsSkill files and discovery manifests now replace pricing pages. What gets published, why it is checkable, and where prompt injection sits.Nov 16, 2026·7 min read·18 views