Should we hook up sashiko.dev?

Should we hook up sashiko.dev?

[Eric Sandeen]

Hi all - 

"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."

This is something built by Chris Mason and others, with Google providing the LLM and tokens.

The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.

We could send a PR to request adding linux-xfs, and I think we should try it.

However, if you look at the config, for example for linux-ext4:
https://github.com/sashiko-dev/sashiko/commit/d921c59

there are options to:

1) reply back to the patch other (or not), and
2) automatically cc: one more more people on every review if desired

so before single-handedly making this decision for every future patch author
and/or possible cc: individuals, what do we think?

(One can drive Sashiko from the command line, too, with or without the list
subscription mentioned above, see the github repo for details.)

Oh, and I guess we could add xfs-specific review prompts, too, see
https://github.com/masoncl/review-prompts/blob/main/kernel/subsystem/btrfs.md
for example. EXT4, OTOH, does not have this. (and the btrfs prompt looks quite
limited right now.) Maybe we could distill the xfs documentation into prompts...
but I'm getting ahead of myself now.

Thanks,
-Eric

Re: Should we hook up sashiko.dev?

[Carlos Maiolino]

On Wed, Apr 15, 2026 at 11:49:50AM -0500, Eric Sandeen wrote:
> Hi all - 
> 
> "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."
> 
> This is something built by Chris Mason and others, with Google providing the LLM and tokens.
> 
> The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.
> 
> We could send a PR to request adding linux-xfs, and I think we should try it.
> 
> However, if you look at the config, for example for linux-ext4:
> https://github.com/sashiko-dev/sashiko/commit/d921c59
> 
> there are options to:
> 
> 1) reply back to the patch other (or not), and
> 2) automatically cc: one more more people on every review if desired
> 
> so before single-handedly making this decision for every future patch author
> and/or possible cc: individuals, what do we think?
> 
> (One can drive Sashiko from the command line, too, with or without the list
> subscription mentioned above, see the github repo for details.)
> 
> Oh, and I guess we could add xfs-specific review prompts, too, see
> https://github.com/masoncl/review-prompts/blob/main/kernel/subsystem/btrfs.md
> for example. EXT4, OTOH, does not have this. (and the btrfs prompt looks quite
> limited right now.) Maybe we could distill the xfs documentation into prompts...
> but I'm getting ahead of myself now.
> 

I think it's a fair idea. I've been talking with Mason about it and it's
on my todo list pile to look into it for a while, but I didn't come to
that yet. I have an initial concern on who will be reviewing its output,
but I didn't really put any thoughts about it yet.

> Thanks,
> -Eric
> 

Re: Should we hook up sashiko.dev?

[Eric Sandeen]

On 4/15/26 1:40 PM, Carlos Maiolino wrote:
> On Wed, Apr 15, 2026 at 11:49:50AM -0500, Eric Sandeen wrote:
>> Hi all - 
>>
>> "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."
>>
>> This is something built by Chris Mason and others, with Google providing the LLM and tokens.
>>
>> The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.
>>
>> We could send a PR to request adding linux-xfs, and I think we should try it.
>>
>> However, if you look at the config, for example for linux-ext4:
>> https://github.com/sashiko-dev/sashiko/commit/d921c59
>>
>> there are options to:
>>
>> 1) reply back to the patch other (or not), and
>> 2) automatically cc: one more more people on every review if desired
>>
>> so before single-handedly making this decision for every future patch author
>> and/or possible cc: individuals, what do we think?
>>
>> (One can drive Sashiko from the command line, too, with or without the list
>> subscription mentioned above, see the github repo for details.)
>>
>> Oh, and I guess we could add xfs-specific review prompts, too, see
>> https://github.com/masoncl/review-prompts/blob/main/kernel/subsystem/btrfs.md
>> for example. EXT4, OTOH, does not have this. (and the btrfs prompt looks quite
>> limited right now.) Maybe we could distill the xfs documentation into prompts...
>> but I'm getting ahead of myself now.
>>
> 
> I think it's a fair idea. I've been talking with Mason about it and it's
> on my todo list pile to look into it for a while, but I didn't come to
> that yet. I have an initial concern on who will be reviewing its output,
> but I didn't really put any thoughts about it yet.

I don't think there's any obligation for anyone to review its output, but it's
there for whoever would like to. I would expect that the patch author would be
first to look, and probably anyone who decided to do a human review might
find it useful as well.

FWIW I am also playing with using claude to generate a set of xfs-specific
review prompts based on documentation, code, and ~500 prior "fixes" commits
that fixed things presumably missed on review. That in itself probably also
would need review. ;)

-Eric

>> Thanks,
>> -Eric
>>
> 

Re: Should we hook up sashiko.dev?

[Andrey Albershteyn]

On 2026-04-15 11:49:50, Eric Sandeen wrote:
> Hi all - 
> 
> "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."
> 
> This is something built by Chris Mason and others, with Google providing the LLM and tokens.
> 
> The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.

I've tried Chris's prompts [1] with Claude for review (the ones
sashiko is based on), and it didn't produce much of a value. But
maybe Gemini models/sashiko is better with them.

I think it's still worth trying to enable it for xfs, maybe it will
give some results. But without Cc imho.

1: https://github.com/masoncl/review-prompts

-- 
- Andrey

> 
> We could send a PR to request adding linux-xfs, and I think we should try it.
> 
> However, if you look at the config, for example for linux-ext4:
> https://github.com/sashiko-dev/sashiko/commit/d921c59
> 
> there are options to:
> 
> 1) reply back to the patch other (or not), and
> 2) automatically cc: one more more people on every review if desired
> 
> so before single-handedly making this decision for every future patch author
> and/or possible cc: individuals, what do we think?
> 
> (One can drive Sashiko from the command line, too, with or without the list
> subscription mentioned above, see the github repo for details.)
> 
> Oh, and I guess we could add xfs-specific review prompts, too, see
> https://github.com/masoncl/review-prompts/blob/main/kernel/subsystem/btrfs.md
> for example. EXT4, OTOH, does not have this. (and the btrfs prompt looks quite
> limited right now.) Maybe we could distill the xfs documentation into prompts...
> but I'm getting ahead of myself now.
> 
> Thanks,
> -Eric
> 

Re: Should we hook up sashiko.dev?

[Eric Sandeen]

On 4/15/26 11:49 AM, Eric Sandeen wrote:
> Hi all - 
> 
> "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."
> 
> This is something built by Chris Mason and others, with Google providing the LLM and tokens.
> 
> The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.
> 
> We could send a PR to request adding linux-xfs, and I think we should try it.

Ok, we're connected: https://sashiko.dev/#/?list=org.kernel.vger.linux-xfs

for now it is not sending any emails; it's web-only. Depending on the
usefulness and accuracy of the reviews, we could consider more than web-only.

I'd like to set up some xfs-specific review prompts, but apparently we need
to be somewhat sparse on that due to token usage for a large set of prompts.

LMK what you think of the review quality!

-Eric

Re: Should we hook up sashiko.dev?

[Brian Foster]

On Tue, Apr 21, 2026 at 06:47:01PM -0500, Eric Sandeen wrote:
> On 4/15/26 11:49 AM, Eric Sandeen wrote:
> > Hi all - 
> > 
> > "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."
> > 
> > This is something built by Chris Mason and others, with Google providing the LLM and tokens.
> > 
> > The website sashiko.dev shows these review results for mailing lists it is subscribed to, so you can browse around and see how it works. I've heard reasonably positive feedback from some who've used it already.
> > 
> > We could send a PR to request adding linux-xfs, and I think we should try it.
> 
> Ok, we're connected: https://sashiko.dev/#/?list=org.kernel.vger.linux-xfs
> 
> for now it is not sending any emails; it's web-only. Depending on the
> usefulness and accuracy of the reviews, we could consider more than web-only.
> 
> I'd like to set up some xfs-specific review prompts, but apparently we need
> to be somewhat sparse on that due to token usage for a large set of prompts.
> 
> LMK what you think of the review quality!
> 

I just took a quick look at this one and it seems like it's accurate:

https://sashiko.dev/#/patchset/20260420061630.62077-1-changfengnan%40bytedance.com

The flaw isn't actually with the patch, but rather the surrounding code
it touches. It also feels a little weird to me to have a tool that tries
to communicate like a person, but maybe that's just something to get
used to I guess. ;)

Fengnan,

Any chance you want to try to fix that up in your patch, since it's
outstanding and you've created the helper to facilitate it? ;) If not
nbd we can fix it up separately. I don't think it's a pressing issue.

Brian