Sashiko, the Google-developed open-source agentic AI code reviewer that started on the Linux kernel, added an optional “Linux Performance Specialist” review stage in Friday’s 0.9.5 release, Phoronix reports. It works in both daemon and local sashiko review modes (including --stages performance and --cheap) and traces second-order side effects, with “anti-noise guardrails” to keep false alarms down. The release also adds a balanced 40-commit benchmark suite: 20 known upstream performance regressions and 20 clean commits, plus per-stage false-positive tracking.
My take
Kernel performance regressions sneak in all the time, so an AI reviewer that’s specifically looking for them, and that’s measured against known-good commits so it doesn’t cry wolf, is the right way to do this. Benchmarking your AI reviewer against real regressions is a habit more projects should copy.
God Bless, and May the Source Be with You!