Hi Aleksandr,

Sorry, the AI agent was quite fast and sent a reply before I could validate
it, instead of adding a todo for tomorrow :)

This is an important point: we have not discussed an AI code policy at all
yet, even though we are already using AI in our work. I will prepare a
separate discussion on the AI code policy tomorrow.

For example, a user-facing skill could run pre-flight checks before taking
a node out of service. It would look at the cluster state and warn whether
stopping the node is safe.

It could also help investigate a problem using the available logs. There is
already a proof of concept developed outside the Apache Software
Foundation: https://gitverse.ru/dpavlov/ignite-firefighter/  - It is just a
vibe-coded tool for now, but hopefully we can create something better
together.

Best regards,
Dmitriy

ср, 2 сент. 2026 г. в 23:59, Teamcity-tasks Dmitriy Pavlov <
[email protected]>:

> Hi Aleksandr,
>
> Thanks, this is a useful point. By user-facing skills I do not mean a
> generic agent that can already search docs or source code. The idea is
> to package a narrow workflow with project context and checks.
>
> For example, the docs/release-note skill I mentioned would look at a
> concrete change, find the affected docs and release-note conventions,
> and help prepare the update. It should be easy to see its inputs,
> expected output, and limitations. I will publish that first so we have
> something real to discuss.
>
> I also agree that we should not publish a "make a PR" skill without
> project-level guardrails. The contributor still owns the patch, tests,
> and design; the agent is only a tool. A short AI usage policy and an
> AGENTS.md with the rules you listed sound like a good starting point.
>
> I can prepare a small initial proposal around that together with the
> first skill. Then we can review the rules and decide what belongs in
> the main repo.
>
> Best,
> Dmitriy
>

Reply via email to