PurCL

A continuously updated collection of papers on agentic SE

🥇 Amazon Nova AI Challenge Winner - ASTRA emerged victorious as the top attacking team in Amazon's global AI safety competition, defeating elite defending teams from universities worldwide in live adversarial evaluation.

An autonomous LLM-agent for large-scale, repository-level code auditing

Official repo for "ProSec: Fortifying Code LLMs with Proactive Security Alignment"

An Educational Framework of LLM-driven Repository-level Program Slicing

[COLM 2025] Official implementation of μKE - edit LLM knowledge while preserving memory dependencies via Matryoshka-style objectives.