HCAG: Hierarchical Abstraction and Retrieval-Augmented Generation on Theoretical Repositories with LLMs

Collection of arXiv CS/SE papers (2026-03-24) covering LLM-agent and software-engineering advances: HCAG proposes hierarchical abstraction + retrieval-augmented generation for repository-level code and theory-aligned implementations; a paper reframes software design for AI-native, agent-invoked interfaces; kRAIG autoregenerates Kubeflow pipelines with intent-clarification and validation; a semantic vector-based MCP tool discovery method reduces token overhead while improving tool selection; an empirical study shows tool affordances substantially degrade LLM-agent safety despite text-only pass;