Review Linux kernel patches before they land
Google’s Sashiko scans Linux kernel submissions and flags bugs that human reviewers missed.
- What they did
- Roman Gushchin’s team at Google built Sashiko over several months and tried it inside the company before making it public. The system now reviews every post sent to the Linux kernel mailing list and is open source. It was tuned for Gemini 3.1 Pro, but it can also run with Claude and likely other models.
- How it went
- In tests on 1,000 recent upstream issues marked with Fixes: tags, it caught about 53% of the bugs. Human reviewers had missed those problems entirely, though Gushchin said that hit rate is not especially high.
- Tips
- If you try something similar, budget for the model usage and hosting costs yourself. Google is covering those costs for the hosted setup while the Linux Foundation takes over the hosting side.
Try it yourself with Codex
Act as a senior kernel code reviewer. Examine [patch or diff] in [repo or path] for correctness, memory issues, race conditions, error handling gaps, and confusing names. Cite the exact lines or hunk numbers for each issue. Pause before any irreversible action such as sending, submitting, deleting, or applying changes: first summarize your findings and ask me to approve. Then give a ranked list of problems and a short verdict on whether the patch is ready to merge.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Build a staged agent pipeline for PRs
Create a multi-step system that assigns issues, tests changes, and leaves the final merge to a person.
Assemble a R coding agent from three basic tools
They wired a Claude agent in R to file and shell helpers, then hardened it with safer path checks and editing functions.
Set up a GitHub bot that triages and patches bugs
They built a GitHub bot that sorts issues, reviews pull requests, answers with citations, and opens patch PRs for humans to check.
Check upgrade risks before you ship
A ReAct agent reviewed dependency data for a planned cloud upgrade and blocked it when it found an incompatible SDK version.
Connect a coding agent to an Excalidraw board
They hooked an installed coding agent to a live Excalidraw canvas so it could read the scene and edit the drawing directly.
Use parallel agents to build a database engine
They split a SQLite implementation across several coding agents working in separate branches, then merged the pieces after repeated test-and-fix loops.