skills/ask-questions-workspace/skill-snapshot/SKILL.md
Ask a genuinely good question — or a composed set of them — in the moment, to discover, clarify, probe, or confirm, in any conversation or domain. Use when about to ask the user something or use the `AskUserQuestion` tool, when a question fell flat or one won't reach and you need a set, when the user's tone or messages signal you have drifted from what they need, or when the user wants questions phrased, generated, or sharpened — "read this spec and generate questions", "ask about the schema", "review this and ask questions", "help me understand what to ask", or to clarify a requirement or idea.
npx skillsauth add synapseradio/ai-skills ask-questionsInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
A question is an instrument: the words you pick decide what comes back. And a good question rarely works alone — it serves one inquiry. Name the question the inquiry exists to answer, and call it the driving question. Every candidate question is then a rung that must earn its place on the ladder toward it. The output is one clean question, or a small set composed to climb together.
One gate holds in the core, from Question Under Discussion theory: would a complete answer to this question at least partly answer the driving question? If not, it's a digression dressed as a rung — cut it, or find the question that isn't.
Every drafted question passes all four before you ask it.
First decide which kind of move the moment calls for — composing the inquiry, going under an answer, or auditing the inquiry itself — then pick the one reference within it. Load that reference and work through its Questions section before you ask.
A. Compose the inquiry. You're building or arranging a set of questions that reaches the driving question.
| The gap | Load | |---------|------| | You don't yet have the rungs and must generate the set that reaches the driving question | ladder | | You already have the questions and are deciding the arc — where to open, when to narrow, what to avoid | sequence-shapes | | You have a qualified set and must pick the single next question to ask | ordering |
When more than one row here matches, load in this order: ladder (generate) → sequence-shapes (arrange) → ordering (pick next). Composing and unsure where to start? Start with ladder.
B. Go under one answer. An answer came back and you need what sits beneath it. Pick the direction you're headed.
| The direction | Load | |---------------|------| | Up — to the value under a stated preference, by asking what matters about each answer | laddering | | Down — to the observable data under a conclusion, belief, or plan | climb-down | | At — one shaky element: clarify a term, test a claim, anchor a generality, label a feeling, or confirm you understood | probe |
C. Audit the inquiry's foundations. Step back from the answers to the inquiry itself.
| The gap | Load | |---------|------| | Some question treats the thing as simpler than it is | pretense | | You're starting out and must name what you accept without justification | grants | | You're carrying a ladder or a question set into a new domain | transfer |
A question lands in a context, and the context decides whether the question is even worth asking. Sometimes the context is already in front of you — the conversation carries it, or you just read the file. More often it isn't, and the honest move is to go read first: skim the thread, open the code, check what's already been said. A question that asks for something you could have found yourself wastes the other person and signals you didn't look. What you're after is the gap that genuinely remains — the thing only they can answer.
Forming a question has two parts: gathering the context and shaping the words. You can keep both, or split them across agents. Pick by where the work is.
data-ai
Locate where a small change shifts the whole system
tools
Surface unstated assumptions at multiple levels and rank them
data-ai
Assess how well evidence supports claims
testing
Spot logical errors in reasoning