/grill-me and /grill-with-docs are changing No more "one question at a time" - now it asks questions in rounds Faster, less token spend, but still keeps dependencies between questions clear. Shipping in v1.2

Learn about my skills here:

cc @marcklingen thanks for the inspiration

Really liked the old one-question-at-a-time flow with AskUserQuestion. It made it much easier to answer without spending forever typing. The new round system is cool when you have time (or can) to speak, but not always practical. Still appreciate the work though! 👍

AskUserQuestion can still handle rounds

i actaully prefer one at a time. its easier to digest and doenst give me anxiety. it solves the problem ai flooding you with too much information at once. please dont do this :)

What is the official way to "update" the skills, meaning also removing old ones? I recently checked the docs but could not find an answer

npx skills update

Keep grill me as grill me/interview me. Do the new skill with some another name. Grill me has it's own usecase btw.

I disagree, I think I have to push what I think are the best practices. It's trivial to adjust the skill to your preferences.

So does this mean a previously answered question no longer shapes the next question?

No

support Codex🥹PLEASE

What support do you need? It works OOTB

May be, it can be integrated into the plugin of Codex

Actually huge! The one question at a time was slow. But also it could use previous answers to inform future questions. Tricky!

Getting overwhelmed by too much AI output is worse than token spend as it reduces quality of human judgement. I stick to one question at a time. Often the question needs my full concentration and other things depend on it.

Can't you make it /batch-grill-me? Love 1 at a time so so much

cant wait! my grilling sessions are usually just 10% clarifications / disagreements and 80% just agree's

I hope questions in rounds don't multiply my 'esprit d'escalier'.

Great update!

One question at a time is superior UX.

Thanks for all the work, I'm relying on this and the rest of your skills every day. Open to trying it, but I'm experiencing more friction when answering multiple questions in a text box. I'm not using Claude Code, so I don't get the AskUserQuestion UI maybe the experience is smoother there.

this is better and best version which stands in between

Batching for token spend makes sense, but what happens when a round-2 answer invalidates an assumption baked into round-1's questions? Re-prompt right away or let it resolve at the end of the round?

I was about to ask about this. Can't you build something, which takes the previous answers into consideration and asks the next one with that in context?

Finally. Did you notice any lack of depth in questions through batches? In the earlier version of the skill it said some about multiple questions being bewildering.

I liked the one question at the time because it's included the previous question context in the next one

I think I like the one-by-one rounds more, but even if you change it, it's easy to work around that by giving an additional instruction when invoking the skill

Intra-round invalidation is the quiet failure mode. Built a similar batched loop once: Q2's answer broke Q3's premise and I didn't catch it until the next round. One-at-a-time sidesteps this because the model pivots mid-stream.

Questions in rounds are better because dependencies surface earlier. One-at-a-time feels thorough, but it hides the shape of the problem and spends tokens on branches that later answers may invalidate.

Why don't you make one that is batch an another that is one by one?

Rounds approach is smart. Collapsing sequential Qs into batches while preserving dependency graphs is exactly the kind of UX-token tradeoff worth shipping. Clean move, Matt.

So much value thanks Matt

/grill-me and /grill-with-docs are such great skills.

Been using a grill-me with 5 questions max per round and the ask question tools from copilot. It can now display a mermaid diagrams it helps a lot. What I added recently is a « rubber duck » reformulation of the problem and the project at the start of this session, so i can see: 1) does the agent has an understand of the project (that means the AGENTS.md and the initial exploration is done correctly) 2) the user (that might not be you) have a small a succinct presentation of what the current session is about.

Now that's a good middle ground

great!!! For anyone who wants to try now, it's there:

Interesting, I'm curious to try it.

Is this live mate?

It's in a PR

Really good idea, it can be really useful when you are dealing with a project that you are pretty sure what you need.

Great! I'm looking forward to the release of v 1.2.

Great!