Sashiko AI Introducing Linux Performance Specialist Review Stage
The Google-developed Sashiko for agentic AI code review initially focused on the Linux kernel continues helping to uncover new bugs and issues as new patches hit the kernel mailing list as well as those manually running this open-source AI software. In addition to recently introducing an experimental patch series design assessment, they are introducing another new skill for helping with…
Sashiko AI has announced an optional performance specialist review stage for Linux performance reviews. This new stage will be available in both local and daemon review modes, as well as when using the --stages performance and --cheap flags.
The performance specialist stage will include second-order side-effect tracing and anti-noise guardrails. It will also support neutral benchmarking with a suite of 40 benchmark_performance.json commits, 20 of which are known upstream regressions and 20 are verified clean commits.
The implementation of this new stage will be integrated into the ANALYSIS_STAGES in linux_patch_review.rs file for both local and daemon review modes. This is part of a larger effort to address kernel performance regressions, which are often unexpected side effects of bug fixes, cleanups, and struct layout changes.
To ensure accurate results, the performance specialist stage will trace callers and outer embedding structs up to 1-2 hops outward using repository tools. Pure docs, comments, or ID tables will be skipped, as kernel performance regressions are typically unexpected side effects of ordinary changes.
The performance specialist stage will use controlled A/B benchmarks on the benchmark_performance.json suite, with the gemini-4-argon model at high effort. The positive detection rate is expected to be 15.0% (3/20), with 70.0% of commits (14/20) showing improvements (55.0% net-new wins originated solely from performance). The neutral clean rate is also expected to be 60.0% (12/20), with no new false-positive neutral commits.
The resource overhead for this new stage is expected to be a 11.0% increase in median input tokens (91.4% cached) and a 3.1% decrease in median per-commit duration. Michael Larabel, the principal author of Phoronix.com and founder of the site, will be leading the development of this new performance specialist stage.
Rewritten by urgent.news; may contain errors — check the original.