ResearchCodified Context Three-Component Infrastructure for Agent CodingarXiv·high signalXBlueskyLinkedInCopy linkHot memory + 19 agents + cold knowledge base. Evaluated across 283 sessions on 108K-line codebase. Open-source.SourceSource pagearXiv↳ Follow the threadStack layer / Threat patternUnlearning Methods That Pass TOFU and MUSE Still Leak the Secret on 22-86% of Queries Once the Model Is an AgentarXiv 2609.12808Stack layer / Threat pattern787,562 Function Pairs Show AI Code Is Half the Size of Human Code With Different Defect Classes, Not FewerarXiv 2609.12708Policy dependency / Stack layerCodeBLEU Scored 91% for Both RAG Strategies While One of Them Hallucinated APIs 56.4% of the TimearXiv 2609.12464Policy dependency / Stack layerA Fine-Tuned RoBERTa-Large Permission Gate Matches Claude Haiku 4.5 at Deciding What an Agent May ToucharXiv 2609.15422Stack layer / ContrastReflexion-Style Verbal Memory Sometimes Lowers Success Versus Plain Retry, and Replay Experiments Show WhyarXiv 2609.12404Stack layer / Threat patternMemRiskBench Scores Long-Horizon Agent Memory Risks Deterministically, With No LLM Judge on the Pass/Fail PatharXiv 2609.14976Stack layer / ContrastActGuard audits the planned action instead of filtering tool output, comparing each step against a local tool priorarXivPolicy dependency / Stack layerDream-RSI replays an agent's own discovery tree as a simulator so it can tune exploration policies off-policyarXiv / HuggingFace Daily Papers