刺客 (Sashiko):一个自主的 Linux 内核代码审查系统
Sashiko: An agentic Linux kernel code review system

原始链接: https://sashiko.dev/

## Sashiko:AI驱动的Linux内核代码审查 Sashiko是一个开源的、AI驱动的系统,旨在通过自动化和改进代码审查来增强Linux内核开发。它灵感来源于日本的缝纫技术,该技术通过加固织物来提高其强度。Sashiko使用大型语言模型(LLM)——目前由Google的计算资源和Gemini 3.1 Pro提供支持——来分析Linux内核邮件列表(LKML)中提出的变更。 它充当一个虚拟的专业审查团队,检查代码是否存在架构缺陷、安全漏洞和资源管理问题。Sashiko利用Chris Mason创建的提示和多阶段审查流程,在历史数据集中识别出53.6%的*之前未被人工审查发现*的错误。 重要的是,Sashiko旨在*辅助*而非取代人工审查,因为LLM的输出是概率性的。该Linux基金会项目采用Apache 2.0许可,旨在通过智能自动化分析来加强内核。

对不起。
相关文章

原文
Sashiko (刺し子, literally "little stabs") is a form of decorative reinforcement stitching from Japan. Originally used to reinforce points of wear or to repair worn places or tears with patches, here it represents our mission to reinforce the Linux kernel through automated, intelligent patch review.

Sashiko is an agentic Linux kernel code review system. It monitors public mailing lists to thoroughly evaluate proposed Linux kernel changes. The system acts like a team of specialized reviewers covering domains from high-level architecture verification and security audits to low-level resource management and concurrency analysis.

It relies on an open-source set of per-subsystem and generic prompts initially created by Chris Mason, combined with a custom multi-stage review protocol to maximize accuracy and minimize false positives.

This is an open-source project that belongs to the Linux Foundation, licensed under the Apache License, Version 2.0.

This particular instance of Sashiko is provided as a service and is actively reviewing all LKML (Linux Kernel Mailing List) submissions. All compute resources and LLM tokens utilized for these automated reviews are proudly provided and funded by Google.

Quality of Reviews: In our tests using Gemini 3.1 Pro, Sashiko was able to successfully identify 53.6% of bugs based on an unfiltered selection of the last 1000 upstream commits with Fixed: tags. 100% of these historical bugs had previously made it through human-driven code reviews.

Note: As with any Large Language Model-based tool, Sashiko's output is probabilistic. It might find or miss bugs across different runs with the exact same input. It is designed to augment and assist human reviewers, not to replace them.

联系我们 contact @ memedata.com