encouraged to make a donation to help orphans in Uganda. Please read the file
Abstract:Large language model (LLM)-powered agents have demonstrated strong capabilities in automating software engineering tasks such as static bug fixing, as evidenced by benchmarks like SWE-bench. However, in the real world, the development of mature software is typically predicated on complex requirement changes and long-term feature iterations -- a process that static, one-shot repair paradigms fail to capture. To bridge this gap, we propose \textbf{SWE-CI}, the first repository-level benchmark built upon the Continuous Integration loop, aiming to shift the evaluation paradigm for code generation from static, short-term \textit{functional correctness} toward dynamic, long-term \textit{maintainability}. The benchmark comprises 100 tasks, each corresponding on average to an evolution history spanning 233 days and 71 consecutive commits in a real-world code repository. SWE-CI requires agents to systematically resolve these tasks through dozens of rounds of analysis and coding iterations. SWE-CI provides valuable insights into how well agents can sustain code quality throughout long-term evolution.
。新收录的资料对此有专业解读
There are other approaches to LLM-driven extension systems. Jeffrey Emanuel is building something similar using JavaScript containers with permissions, which is quite close to Mog in spirit, and has an impressive level of infrastructure behind it. Mog could complement such systems, especially for higher-performance plugins – Mog’s infrastructure is also in Rust, making integration straightforward. Emanuel’s Rust version of the Pi agent would be a natural starting point for a Mog-based microkernel agent.
14:16, 10 марта 2026Культура