ResearchAgentDropoutV2 Test-Time Rectify-or-Reject Pruning for Multi-Agent SystemsarXiv·high signalXBlueskyLinkedInCopy linkTackles cascading errors in multi-agent systems. Test-time pruning framework as active firewall between agent handoffs without retraining.SourceSource pagearXiv↳ Follow the threadPolicy dependency / Stack layerCROSS-CATEGORY: Three Independent Agent-Action Gates Shipped in 48 Hours, All Judging the Command Against Stated IntentProduct Hunt, github.com/AGGIB/Stroq and rewarelabs.com (three independent sources; the 72% figure is Reware's own)Stack layer / ContrastA capability-scoped harness cut prompt-injection execution from 33-47/75 runs to 3/75 without asking the model to spot malicious textarXiv 2609.08371Stack layer / ContrastA spec-first agent framework taxonomy: persuasion, front-loaded structure, or controls the agent cannot editarXiv 2609.09671Policy dependency / Update threadNone of five agent-memory systems actually enforce revocation at retrieval timearXivStack layer / ContrastA knowledge graph for what-to-do: procedure triplets that self-evolve by contrasting failed trajectories with successful onesarXiv 2609.09153Stack layer / Threat pattern16% of 3,171 public agent-harness setups carry a confirmed security defect, and 3.8% ship a skill that pre-approves your shellarXivStack layer / Follow-up threadModels Spot Only 9.6% of Implementation Gaps in Research Specs but Fix 80.6% Once You Point Them OutarXiv 2609.10539Policy dependency / Stack layerMemSentry gates persistent memory writes on a signed security-state delta rather than on content classificationarXiv