Princeton's autonomous software-engineering agent — takes a GitHub issue and produces a patch. Its SWE-ReX backend runs the agent's commands in isolated sandboxes (Docker, Modal, or remote).
SWE-agent (Princeton) is the research harness that popularised the "give an LLM a repo + an issue, get back a patch" loop and set early SWE-bench records. You point it at a GitHub issue or a local task; it explores the repo, edits files, runs commands, and emits a diff.