SWE-agent: Open-source tool uses LLMs to fix issues in GitHub repositories
By connecting powerful language models like GPT-4o and Claude Sonnet 3.5 to real-world tools, the open-source tool SWE-agent allows them to autonomously perform complex tasks: from fixing bugs in live GitHub repositories and solving cybersecurity challenges, to browsing the web or executing custom workflows.
“SWE-agent was the first open source agent to ever demonstrate significant numbers on the SWE-bench benchmark, i.e., it was the first software engineering agent that showed promising results in solving GitHub issues. Its unique in our aim for simplicity. The agent flow is almost entirely unconstrained: After an initial prompt, the LM can issue any actions, until it finally calls the ‘submit’ command or a cost limit is reached. This allows to adapt the agent very easily,” Kilian Lieret, Research Software Engineer at Princeton University and co-creator of SWE-agent, told Help Net Security.
SWE-agent has a mode called EnIGMA. It’s built to solve offensive cybersecurity tasks, like capture-the-flag challenges. On several benchmarks, EnIGMA ranks among the top performers.
Many of EnIGMA’s features are now part of the full SWE-agent tool. This includes a debugger, tools to connect to remote servers, and a summarizer that helps the AI deal with long or messy outputs. These tools make it easier for language models to solve complex problems on their own.
Future plans and download
“We want to continue with our current approach: Having a simple, configurable and hackable agent that allows researchers (and anyone else) to adapt it to any task, especially if it requires sandboxed execution environments (one of the engineering challenges),” Lieret said.
SWE-agent is available for free on GitHub.
Must read:
- GitHub CISO on security strategy and collaborating with the open-source community
- Don’t let these open-source cybersecurity tools slip under your radar
- 33 open-source cybersecurity solutions you didn’t know you needed
Subscribe to the Help Net Security ad-free monthly newsletter to stay informed on the essential open-source cybersecurity tools. Subscribe here!