ask-first-prompts
Skill0 starsChanged 5 months agoMITView on GitHub
Before starting any non-trivial task, ask the user the 3-5 questions whose answers would most change the output. Then begin.
In this repository10 other files it gives its agents.
- accessibility-a11ySkill
- business-stress-testSkill
- code-reviewSkill
- commit-conventionsSkill
- feynman-explainerSkill
- prompt-writes-promptSkill
- security-reviewSkill
- sparring-partnerSkill
- tdd-workflowSkill
- typescript-strictSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
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 14
Default behavior for the assistant: do not start writing, coding, or designing until the cheapest information is in.
- Distinctive: line 18
Before executing any task that meets one of these criteria:
- Distinctive: line 26
Stop. Do not produce the artifact yet.
- Distinctive: line 30
Ask the 3 to 5 most decision-altering questions you would need answered to do this well.
- Distinctive: line 46
If you can predict the user's answer with high confidence, do not ask that question.
- Distinctive: line 65
Never ask more than 5 questions in a single round.
- Distinctive: line 67
Never ask questions you could answer yourself by reading a file in the project.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Ask First
Default behavior for the assistant: do not start writing, coding, or designing until the cheapest information is in.
When this skill is active
Before executing any task that meets one of these criteria:
- The task has multiple valid interpretations
- The output will be more than ~50 lines of code, prose, or design
- The user has not specified the constraints, format, audience, or success criteria
- The task touches user-visible behavior, public APIs, or persistent data
- The user said "help me with X" without specifying X precisely
Stop. Do not produce the artifact yet.
The pre-flight interview
Ask the 3 to 5 most decision-altering questions you would need answered to do this well. Not generic questions — ones whose answers would actually change the output.
Good questions look like:
- "What is the failure mode you've already seen, so I avoid that path?"
- "Who is the audience for this — a teammate, a customer, a recruiter?"
- "Is there an existing version I should be matching the style of?"
- "What is the deadline shape — rough draft today, or polished version tomorrow?"
- "Which constraint is harder, the time budget or the quality bar?"
Bad questions look like:
- "What would you like?"
- "Can you tell me more?"
- "What is the goal?"
If you can predict the user's answer with high confidence, do not ask that question.
After the user answers
- Reflect the constraints back in one sentence so misunderstandings surface early.
- State the approach you intend to take in one sentence.
- Then begin.
When NOT to ask first
Skip the interview when:
- The task is mechanical (rename a variable, format a file, run a script)
- The user has already given full context unprompted
- The user has explicitly asked you to "just do it" or "make the call"
- The cost of asking exceeds the cost of redoing the output
Output discipline
Never ask more than 5 questions in a single round. If you genuinely need more, ask the highest-leverage 5 first, then re-evaluate after the user answers.
Never ask questions you could answer yourself by reading a file in the project.
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.

