ask-first
Skill0 starsChanged 9 days agoMITView on GitHub
Has the AI ask the user questions instead of making them read and write long replies: one question at a time, each with a few answers (the AI's recommendation first and labeled) and a free-form answer last, then a summary of the decisions to confirm before the AI goes on. Works at any point: before a task starts, or in the middle of a conversation that has grown long or overwhelming, about the work so far. Use when the user's message contains the command .ask or .askme or the words ask first used as an instruction to the AI, anywhere in the message, in any case: at the start (ask first: plan our launch), at the end (plan our launch. .ask) or alone (.ask, about the work already in this chat). Also when the user plainly asks the AI to interview them or ask them questions. Not when ask first is about someone else, such as a rule for a colleague (new hires should ask first before deleting files).
6 other projects use a version of this.
Adopted by
Most recently taken up by
- nadavnaveh/skillstheir own version, first seen Nov 1, 2026
- MithunWijayasiri/dotclaudetheir own version, first seen Oct 26, 2026
- ymck-x/dsh-skill-startertheir own version, first seen Oct 15, 2026
- icigarette/askFirst-skilltheir own version, first seen Oct 13, 2026
- unix2dos/skillstheir own version, first seen Sep 22, 2026
- nathanonn/agent-skillstheir own version, first seen Jun 5, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- nathanonn/agent-skills★ 22same name
- unix2dos/skills★ 7same name
- MithunWijayasiri/dotclaude★ 4same name
- nadavnaveh/skills★ 2same name
- icigarette/askFirst-skill★ 0same name
- ymck-x/dsh-skill-starter★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 38
Ask one question per message, so each answer can shape the next question.
- Distinctive: line 55
When the app can't show clickable choices, or you aren't sure it can, write the question as plain text, in this layout, with no bold or …
- Distinctive: line 74
When the decisions that matter are made, stop asking.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Ask First
Version: 1.0.0
Author: RosBradley · https://github.com/RosBradley/ask-first
A short request leaves a lot unsaid, and a long conversation piles up more than the user wants to read and answer. This skill turns that around: at any point, before a task or in the middle of one, ask the questions that would change the result, one at a time, each with a few ready answers to pick from. Your recommended answer is always first, so the user can move fast by taking it, and the last answer is always the user's own words.
Commands
| Plain words | Dot style | What happens |
ask first | .ask or .askme | Starts an interview about the task in the same message, or about the work already in this chat if there is none |
- The command can be anywhere in the message:
ask first: write a launch plan,Write a launch plan. .askme, or.askalone. Case doesn't matter, so.Ask,.AskMeandAsk firstcount. - Everything else in the message is the task. Leave the command words out of it.
.askand.askmealways carry the dot.askalone is an ordinary word.ask firstcounts only when the user is telling you to ask before you start. "Write a rule for new hires: ask first before deleting files" is about someone else: help with it without this skill.
Before the first question
- Work out the task. It is the rest of the message; with no task in the message, it is the work in progress in this chat. With neither, the first question asks what the interview is about.
- Learn what can be learned without asking. Read the files, earlier messages and sources the task points to. Never ask the user something you can look up.
- List the decisions that would change the result, most important first. Keep the ones where the user's answer matters: goals, audience, scope, constraints, trade-offs, and things the user probably hasn't thought about. Drop anything with an obvious default; just use the default and say so in the summary.
- Say in one line what the interview is about and roughly how long it is:
Interviewing you about the NERC CIP program: about 6 questions.
Asking
Ask one question per message, so each answer can shape the next question. Each question has:
- A number and the running estimate:
Q3 of ~6. Update the estimate as answers open or close topics; it's an estimate, not a promise. - The question in one sentence.
- 2 to 4 answers. Your recommendation is first and labeled
(Recommended). Each answer gets a short line on what it means or what it costs, so the choice can be made without more research. - A free-form answer last, so the user can write in their own.
Answers are real alternatives, not one good choice with filler around it. When there is no clear best choice, still put first the one you would pick, and say why in its line.
When the app can show clickable choices
If this app gives you a way to show a question as choices the user can click or tap (a question tool such as AskUserQuestion in Claude Code, or the interactive options in claude.ai or ChatGPT), use it. Put the recommended answer first, with (Recommended) at the end of its label, and keep the Q3 of ~6 count in the question or its header. If the app adds its own free-form answer (such as Other), don't add another; if it doesn't, make the last choice Something else - write your own.
Batch option: when the app's question tool can ask several questions in one step and the next 2 to 4 don't depend on each other's answers, ask them together. Ask a question on its own whenever its answer would change what comes next.
Otherwise (text)
When the app can't show clickable choices, or you aren't sure it can, write the question as plain text, in this layout, with no bold or other formatting, so it reads the same in every app:
Q3 of ~6: Who will run the NERC CIP program day to day?
A. An existing staff member, part time (Recommended) - fits a small district's budget
B. A new full-time compliance hire - steadier, but a new salary
C. An outside consultant - fast expertise, ongoing cost
D. Something else - write your own
Reply with a letter, rec, or your own words.Letters, not numbers, so a letter is never mistaken for a number in an answer ("3 days"). One line per answer, so it fits a phone screen. The last letter is always Something else - write your own, and the last line is always the reply hint.
The user answers with a letter, with rec for the recommendation, or in their own words. Read an answer in their own words as it is meant; if it is unclear, ask about it once, briefly, then move on.
Decision summary
When the decisions that matter are made, stop asking. Then:
- Show the decisions as a short numbered list, one line each: the user's answers, and any defaults you chose without asking, marked
(default). - Ask a last question to check you understood:
Yes, that's right (Recommended),Change something, and the free-form answer. Ask it the same way as the interview questions, with clickable choices when the app has them. It confirms the decisions; it isn't a request for permission, so don't switch to text for it. - On
Yes, that's right, do the task, using every decision in the list. On a change, update the list, show it again, and ask again.
The interview is never the end in itself. Keep it to the questions that change the result.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

