SWE-agent

SWE-agent is an open-source, research-oriented agent built and maintained by researchers from Princeton University and Stanford University. It takes a GitHub issue and lets a language model such as GPT-4o or Claude Sonnet 4 autonomously use tools to try to fix it. The project is now in maintenance-only mode, and its developers recommend mini-swe-agent as the successor.

What It Is

SWE-agent is a command-line software engineering agent. You pair it with the language model of your choice, and the model works through tools to resolve issues in real GitHub repositories, tackle competitive coding challenges, or perform custom tasks. A single YAML configuration file governs its behavior, and the docs cover models, templates, tools, environments, and history processors.

Workflow and Configuration

Runs can be started from the command line, either as a single run or in batch mode for benchmarking on SWE-bench. A trajectory inspector lets users review output files from each run. Users can add custom tools and run competitively. Multimodal support allows vision-capable models to process images from GitHub issues. The project also offers a browser-based trial through GitHub Codespaces.

Cybersecurity Mode (EnIGMA)

EnIGMA is a mode for solving offensive cybersecurity capture-the-flag challenges. The README says to use SWE-agent 0.7 while EnIGMA is updated for version 1.0.

Current Status

The documentation states that SWE-agent has been superseded by mini-swe-agent and is in maintenance-only mode. The project says mini-swe-agent matches its performance while being simpler. The latest listed release is v1.1.0, published May 2025. The code is released under the MIT license.