Hi,
I've been experimenting with an LLM skill that answers Incubator
questions. It routes each question to a real source rather than answering
from memory, using the existing MCP servers for podlings.xml, the health
reports, Incubator reports, general@ and the ASF policy documents.
Things you can ask it are questions like:
- When is our next report due?
- We're preparing our first release, what do we have to do differently
from a normal Apache project?
- Can we name our project Apache Foo?
- Is Pegasus still incubating?
- What does a podling actually have to show to graduate?
- Which of the podlings I mentor haven't reported recently?
- Does a podling need a DISCLAIMER in every release?
It tries to pitch the answer at whoever is asking, so the same question
gets a different reply for a first-time contributor than for a mentor as long
as it knows who you are.
It's in the Incubator repo under tools/concierge. See:
https://github.com/apache/incubator/tree/master/tools/concierge
You need an MCP client and the Incubator MCP servers configured, then:
cd incubator/tools/concierge
make package
Install the generated .skill file in your client and ask it things in plain
English.
There are some eval suites in evals/ if you want to see what it is tested
against.
This is an experiment, so it can be wrong, so please don't expect it to be
100% authoritative. As always, our policy pages are still the authority.
Comments and feedback welcome.
Kind regards,
Justin
