Re: [PATCH 0/4] Update maintainer-pgp-guide.rst Chinese translation

On 8/21/26 9:11 PM, Weijie Yuan wrote:
> On Fri, Aug 21, 2026 at 08:57:46PM +0800, Dongliang Mu wrote:
>> On 8/21/26 8:48 PM, Weijie Yuan wrote:
>>> On Fri, Aug 21, 2026 at 08:17:06PM +0800, Dongliang Mu wrote:
>>> [...]
>>>>> (I definitely won't admit that the log file annoyed me a few times
>>>>> because I accidentally included it with 'git add .' and then had to 'git
>>>>> commit --amend' right after) ;-)
>>>> My experience is to not use 'git add .'. In Linux, hidden files are not
>>>> listed by default.
>>> Yes, I was being lazy to type the full pach/full file name, oops.
>>>
>>> While hidden files are not listed by default on Linux, thankfully the
>>> kernel source tree ignores .* via .gitignore. But 'git add .' is indeed
>>> a bad habit.
>>>
>>> Btw, git commit -v, or setting commit.verbose = true, can really save
>>> one at the last minute. (as mentioned in our how-to doc) And sadly I
>>> actually happened to use another computer to write patches last night,
>>> which I hadn't configured it. XD
>> Actually these labor-tensive works should be done by LLM. This is one of the
>> right directions of LLM usage.
>>
>> You can write a simple skill to help you complete these works.
> That makes sense. I'll give it a try. Still, given that the kernel
> community tends to be a bit conservative, I've generally tried to do
> things myself. Of course, the most important thing is that we supervise
> what the LLM is doing and make sure it follows our actual intent rather
> than hallucinating.

My experience is to always double check the result of LLM.

However, to be honest, translation is not a complicated task. Sometimes 
LLM proposes greater translation than me.

And LLM is really good at capturing some common problems in the patchset.

>
>>>>>>> Btw I've been trying to know more about Sphinx and kernel-docs, hope my
>>>>>>> mistake won't happen again.
>>>>>> That change can fix problem. The underlying problem is RST syntax. Maybe you
>>>>>> can have a look at this or query it in the LLM.
>>>>> Absolutely. I did have relatively little experience writing rst before.
>>>>> But that's definitely no excuse - I'll try to pick it up quickly.
>>>> LLM is really useful to capture some syntax and semantics issues. Therefore,
>>>> before submitting patch, use LLM to scan problem in this issue.
>>> Got it, thanks.
>>>
>>>> I am writing a draft skill for reviewing patches of kernel Chinese
>>>> documentation.
>>> Nice! I've been trying "b4 review tui" these days and I also made some
>>> modifications to its pre-configured prompt to fit our specific
>>> workflows:
>>>
>>>     https://git.kernel.org/pub/scm/utils/b4/b4.git/tree/misc/agent-reviewer.md
>>>
>>> b4 review tui 'agents' - AI-assisted review:
>>>
>>>     https://b4.docs.kernel.org/en/latest/reviewer/getting-started.html#ai-assisted-review
>>>
>>> Sometimes my agent makes false alarms, but at other times it does manage
>>> to identify the issues that we have overlooked. This is probably also
>>> one of the reasons why the community is gradually becoming more
>>> accepting of LLMs.
>> Ah, this is interesting.
>>
>> My skill is to use LLM to search the patch name, use b4 to apply patches and
>> finally ask LLM to refer to how-to.rst and review the patch or patchset.
> As Chen-Yu mentioned before, I´ve basically trained myself to find
> almost any email in the lore archives manually. ;-)
>
> And yes, every time before I send a patch, I have an LLM check whether
> it follows the instructions in the how-to.rst.

Please take a look at my review:

https://lore.kernel.org/all/[email protected]/

The first comment is from Codex + ChatGPT.

And my skill complains the following warning:

https://lore.kernel.org/all/[email protected]/

>
>> And I think we could try to add our subsystem to sashiko. What we need to do
>> is to write some subsystem-related prompts.
> I was thinking about this a few days ago. I'd actually planned to talk
> to the Sashiko maintainer about it first, but I haven´t gotten around
> to it yet. As far as I know, though, Sashiko can currently only be
> enabled based on mailing lists; it can't filter or match patches
> specifically for the Chinese translation subsystem.
>
> Overall, though, I´m in favor of enabling Sashiko for us. Having an
> extra pair of automated eyes certainly doesn't hurt. That said, some
> settings may be worth discussing - for example, whether the Sashiko bot
> should be allowed to send review comments directly to the original patch
> thread.
Maybe we can do some tricks in the prompt. docs/zh_CN: is an obvious 
signal for us.
>
> Thanks!