/ask-me-questions
Gather requirements through adaptive questioning before executing any task
> /plugin marketplace add glittercowboy/taches-cc-resources > /plugin install taches-cc-resources@taches-cc-resources
How it fires
How this command gets triggered: by you, by Claude, or both.
- Fires itselfClaude auto-loads it when your prompt matches the work.
- You can call itInvoke it directly when you want it.
- Slash command
/ask-me-questions
Context preview
What this command does when you run it.
Gather requirements through adaptive questioning before executing any task
Command definition
ask-me-questions.mddescription: Gather requirements through adaptive questioning before executing any task
argument-hint: [task or leave blank]
<objective> Use the Intake & Decision Gate pattern to gather requirements through adaptive questioning before executing a task.
This prevents premature execution, captures nuance, and creates a collaborative context-building flow where you maintain control over when work begins. </objective>
<intake_gate>
<no_context_handler> IF $ARGUMENTS is empty or vague: → **IMMEDIATELY use AskUserQuestion** with:
- header: "Task"
- question: "What would you like help with?"
- options:
- "Write something" - Create a document, email, post, or other written content
- "Build something" - Create code, a feature, system, or technical artifact
- "Figure something out" - Research, analyze, or help me think through a problem
- "Other" - Something else entirely
Then proceed to context_analysis with their response.
IF $ARGUMENTS provides clear context: → Skip to context_analysis </no_context_handler>
<context_analysis> Analyze $ARGUMENTS (or conversation context) to extract what's already provided:
- **What**: The task, deliverable, or outcome requested
- **Who**: Target audience, recipient, or stakeholders
- **Why**: Purpose, goal, or motivation
- **How**: Approach, constraints, or requirements
- **When**: Timeline, urgency, or dependencies
Only ask about genuine gaps - don't re-ask what's already stated. </context_analysis>
<initial_questions> Use AskUserQuestion to ask 2-4 questions based on actual gaps:
**If "what" is unclear:**
- "What specifically do you want?" with domain-appropriate options
**If "who" is unclear:**
- "Who is this for?" with options: Myself, My team, External stakeholders, Public audience, Other
**If "why" is unclear:**
- "What's the goal?" with options relevant to the task type
**If "how" is unclear:**
- "Any constraints or preferences?" with domain-appropriate options
Skip questions where the context already provides the answer. </initial_questions>
<decision_gate> After receiving answers, use AskUserQuestion:
Question: "Ready to proceed, or would you like me to ask more questions?"
Options: 1. **Start working** - I have enough context, proceed with the task 2. **Ask more questions** - There are details I want to clarify 3. **Let me add context** - I want to provide additional information
If "Ask more questions" → generate 2-3 contextual follow-ups based on accumulated context, then present decision gate again If "Let me add context" → receive input, then present decision gate again If "Start working" → proceed to execution </decision_gate>
</intake_gate>
<process> 1. Check if context was provided via $ARGUMENTS 2. If no context: use AskUserQuestion to determine task type 3. Analyze provided context to identify what's already known 4. Ask 2-4 initial questions about genuine gaps only 5. Present decision gate 6. Loop (ask more / add context) until user selects "Start working" 7. Execute the task with full context gathered </process>
<success_criteria>
- No questions asked about information already provided
- User maintains control over when execution begins
- Context accumulates through multiple rounds if needed
- All AskUserQuestion calls use structured options (not plain text questions)
- Task executes only after user explicitly chooses to proceed
</success_criteria>
Read more
description: Gather requirements through adaptive questioning before executing any task argument-hint: [task or leave blank]
<objective> Use the Intake & Decision Gate pattern to gather requirements through adaptive questioning before executing a task.
This prevents premature execution, captures nuance, and creates a collaborative context-building flow where you maintain control over when work begins. </objective>
<intake_gate>
<no_context_handler> IF $ARGUMENTS is empty or vague: → **IMMEDIATELY use AskUserQuestion** with:
- header: "Task"
- question: "What would you like help with?"
- options:
- "Write something" - Create a document, email, post, or other written content
- "Build something" - Create code, a feature, system, or technical artifact
- "Figure something out" - Research, analyze, or help me think through a problem
- "Other" - Something else entirely
Then proceed to context_analysis with their response.
IF $ARGUMENTS provides clear context: → Skip to context_analysis </no_context_handler>
<context_analysis> Analyze $ARGUMENTS (or conversation context) to extract what's already provided:
- **What**: The task, deliverable, or outcome requested
- **Who**: Target audience, recipient, or stakeholders
- **Why**: Purpose, goal, or motivation
- **How**: Approach, constraints, or requirements
- **When**: Timeline, urgency, or dependencies
Only ask about genuine gaps - don't re-ask what's already stated. </context_analysis>
<initial_questions> Use AskUserQuestion to ask 2-4 questions based on actual gaps:
**If "what" is unclear:**
- "What specifically do you want?" with domain-appropriate options
**If "who" is unclear:**
- "Who is this for?" with options: Myself, My team, External stakeholders, Public audience, Other
**If "why" is unclear:**
- "What's the goal?" with options relevant to the task type
**If "how" is unclear:**
- "Any constraints or preferences?" with domain-appropriate options
Skip questions where the context already provides the answer. </initial_questions>
<decision_gate> After receiving answers, use AskUserQuestion:
Question: "Ready to proceed, or would you like me to ask more questions?"
Options: 1. **Start working** - I have enough context, proceed with the task 2. **Ask more questions** - There are details I want to clarify 3. **Let me add context** - I want to provide additional information
If "Ask more questions" → generate 2-3 contextual follow-ups based on accumulated context, then present decision gate again If "Let me add context" → receive input, then present decision gate again If "Start working" → proceed to execution </decision_gate>
</intake_gate>
<process> 1. Check if context was provided via $ARGUMENTS 2. If no context: use AskUserQuestion to determine task type 3. Analyze provided context to identify what's already known 4. Ask 2-4 initial questions about genuine gaps only 5. Present decision gate 6. Loop (ask more / add context) until user selects "Start working" 7. Execute the task with full context gathered </process>
<success_criteria>
- No questions asked about information already provided
- User maintains control over when execution begins
- Context accumulates through multiple rounds if needed
- All AskUserQuestion calls use structured options (not plain text questions)
- Task executes only after user explicitly chooses to proceed
</success_criteria>
A growing collection of custom Claude Code resources built for real workflows.
Other commands on taches-cc-resources.
audit-skill
Audit skill for YAML compliance, pure XML structure, progressive disclosure, and best practices
audit-slash-command
Audit slash command file for YAML, arguments, dynamic context, tool restrictions, and content quality
audit-subagent
Audit subagent configuration for role definition, prompt quality, tool selection, XML structure compliance, and effectiveness

