SWE-agent
The open-source Princeton agent that introduced an agent-computer interface (ACI) for repositories; its 100-line successor scores 65% on SWE-bench Verified.
SWE-agent showed that the interface matters as much as the model: giving the model purpose-built commands for search, file viewing and editing raised SWE-bench scores from ~4% to 12.5% in March 2024. mini-SWE-agent, a 100-line Python re-implementation, later reached 65% on Verified with a stronger model — evidence that much of agent capability now lives in the model rather than the scaffold.