Tag
A detailed thread explaining the high-level architecture of the SWE AI coding agent, showing how a GitHub issue flows through ingestion, an orchestrator, model gateway, tools, code intelligence, sandbox environment, PR builder, guardrails, and observability to autonomously produce a pull request.
mini-swe-agent is a minimal, open-source SWE-agent implementation that tops DeepSWE benchmarks with just 100 lines of code and a single bash tool. The team also open-sourced mini-swe-code for interactive use and mini-swe-acp for evaluation harness across benchmarks.
AgentLens is a framework for process-level assessment of software engineering agent trajectories, revealing that over 10% of passing trajectories exhibit a 'Lucky Pass' behavior. It introduces AgentLens-Bench, a dataset annotated with quality scores, and shows that ranking by quality score can shift model rankings significantly.