A coding agent built on LangGraph from scratch — a planner, a ReAct tool loop, and a network-disabled Docker sandbox — graded against 50 real Django and Sphinx issues from SWE-bench Verified Mini.
· Solo — design, build, benchmark
- langgraph
- agents
- evaluation
- swe-bench
- python