CommunityCoding & Developmentgithub.com

chunyun1998-sudo/step-by-step-ask

Step-by-step requirement gathering skill: Ask first, then execute.

What is step-by-step-ask?

This skill, named 'Step-by-Step Ask,' introduces a methodical question-and-answer process for AI agents to gather complete requirements before executing any task. It is designed to prevent misunderstandings, rework, and direction drift by ensuring the AI fully understands the user's needs before taking action. The skill is triggered by keywords like '一步来' or '先问清楚' and enforces a strict workflow: ask one question at a time, continue based on answers, confirm completeness, then rephrase the entire requirement for user confirmation before execution. It is platform-agnostic and can be applied to any AI agent or coding assistant that interacts with users in a conversational manner. The core rules prevent overwhelming the user with multiple questions, making assumptions, or skipping the confirmation step, thus promoting a natural dialogue-like interaction that respects the user's context and constraints.

Works with~Claude Code~Codex CLI~Cursor
npx skills add chunyun1998-sudo/step-by-step-ask

Installed? Explore more Coding & Development skills: steipete/bluebubbles, steipete/eightctl, steipete/blucli · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Step-by-Step Ask — Sequential Requirement Gathering

When to Invoke

When the user says any of the following keywords or similar expressions, you must immediately enter question mode:

  • "Take it step by step"
  • "Hold on, ask clearly first"
  • "One step at a time"
  • "Slow down, let's get it straight"
  • "Ask clearly first"
  • A combination of "Hold on" + "ask clearly"
  • Any expression meaning "Don't rush to execute, first clarify the requirements"

Work Objective

By asking one question at a time, fully collect all the information needed to execute the task. Ensure that the AI fully understands the user's requirements before starting any work, avoiding direction drift or rework.

Core Rules

1. Ask Only One Question at a Time

  • The AI asks the user one question per round.
  • Do not throw multiple questions at once for the user to choose or answer.
  • Questions should be specific, clear, and targeted at the current information gap.

2. Continue Questioning Based on Answers

  • After hearing the user's answer, determine what information is still needed.
  • If information is insufficient, ask the next question.
  • The questioning logic should be natural and fluid, like a conversation, not a survey.

3. Know When to Stop Questioning

Stop questioning when all of the following conditions are met:

  • The AI fully understands the user's goal and intent.
  • The AI has collected all key information required to execute the task.
  • There are no significant information gaps that could lead to direction errors.

4. Confirm by Rephrasing Before Execution

  • After questioning ends, the AI must first rephrase the complete requirement understanding to the user.
  • The rephrasing should cover: goal, scope, key requirements, constraints, and expected deliverables.
  • Wait for explicit user confirmation (e.g., "Yes," "No problem," "Okay") before starting execution.
  • If the user corrects any part of the rephrasing, the AI updates its understanding and re-confirms again.

Workflow

User triggers keyword (e.g., "Take it step by step")
  ↓
① Enter question mode, ask the first question based on known information
  ↓
② Listen to user's answer, evaluate information completeness
  ↓
③ Not enough info? → Go back to ①, ask the next question
   Enough info? → Go to ④
  ↓
④ Rephrase the complete requirement understanding to the user
  ↓
⑤ User confirms? → Start execution
   User corrects? → Update understanding, go back to ④ and rephrase again

Prohibited Actions

  • Asking multiple questions at once.
  • Starting to execute the task before all information is collected.
  • Skipping the rephrase-and-confirm step and acting directly.
  • Making assumptions for the user or filling information gaps on your own during questioning (if assumptions are necessary, they must be explicitly marked and confirmed by the user in the rephrasing step).

Usage Examples

User says: "Hold on, ask clearly first, I want you to help me write a WeChat public account article."

The AI should:

  1. Enter question mode, ask the first question: "What is the topic or subject direction of the article?"
  2. After the user answers, continue asking: "Who is the target reader? Is it your usual public account audience?"
  3. After gradually collecting all information, rephrase: "I understand. You want a long-form WeChat article on [topic] in a casual writing style, targeting [audience], about [word count], emphasizing [key points], and avoiding [pitfalls]. Is that correct?"
  4. Only start writing after user confirmation.

User says: "Take it step by step, help me create a data report."

The AI similarly triggers question mode, confirming the data source, format, time range, key metrics, etc., one at a time.

Related Skills