always-use-askquestion

name	always-use-askquestion
description	Forces the agent to always use the AskQuestion tool for user prompts instead of conversational questions. Apply this whenever asking the user anything - questions, confirmations, choices, or clarifications. Never ask conversational questions when structured options are possible.

Always Use AskQuestion

CRITICAL RULE: Whenever you need input from the user, ALWAYS use the AskQuestion tool. Never ask conversational questions when you can present structured options.

When to Use AskQuestion

Use AskQuestion for ALL of these situations:

• Yes/No questions: "Should I proceed?" "Does this look good?"
• Multiple choice: "Which option would you like?" "Select a file"
• Confirmations: "Ready to commit?" "Create PR now?"
• Selections: "Which branch?" "Pick a component"
• Preferences: "How should I handle this?" "What format?"

⚠️ COMMON MISTAKE: Don't Batch Multiple Questions Into One String

The most common error is cramming multiple distinct questions into a single prompt string:

❌ WRONG:

prompt: "I need several things:
1. What should I build?
2. Should I search the codebase?
3. Should I search online?"

✅ CORRECT - Use separate AskQuestion calls or separate question objects:

Questions:
  - id: "requirements", prompt: "Ready to provide requirements?", options: [...]
  - id: "search_codebase", prompt: "Should I search the codebase?", options: [...]
  - id: "search_online", prompt: "Should I search online?", options: [...]

Or if questions are sequential (one depends on the previous answer), use separate AskQuestion tool calls.

Never Use Conversational Questions

❌ BAD (conversational):

Should I proceed with the changes?

✅ GOOD (AskQuestion):

AskQuestion({
  "title": "Confirm Changes",
  "questions": [{
    "id": "proceed",
    "prompt": "Should I proceed with the changes?",
    "options": [
      {"id": "yes", "label": "Yes, proceed"},
      {"id": "no", "label": "No, cancel"}
    ]
  }]
})

AskQuestion Structure

Every AskQuestion call must include:

1. title: Short header for the dialog (2-5 words)
2. questions: Array with at least one question object
   • id: Unique identifier for this question
   • prompt: The question text to display
   • options: Array of at least 2 choices
     • id: Unique identifier for this option
     • label: Display text for this option
   • allow_multiple (optional): Set to true for multi-select

When to Use Multiple Questions vs Separate Calls

Use multiple question objects in ONE AskQuestion call when:

• Questions are independent (one doesn't depend on the other)
• User can answer all at once
• Example: "Which file format?" AND "Which location?" (both can be answered together)

Use SEPARATE AskQuestion calls when:

• Questions are sequential (one depends on the previous answer)
• Need to show results/context between questions
• Example: "Ready to start?" → (if yes) → "Which option?" → (based on option) → "Confirm?"

CRITICAL: Each Question Must Be Its Own Object

❌ WRONG - Multiple questions crammed into one prompt string:

AskQuestion({
  "title": "Initial Information",
  "questions": [
    {
      "id": "info",
      "prompt": "Let me gather information:
        1. Requirements: What needs to be built?
        2. Should I search the codebase?
        3. Should I search online?",
      "options": [
        {"id": "proceed", "label": "I'll provide requirements now"},
        {"id": "abort", "label": "Cancel"}
      ]
    }
  ]
})

✅ CORRECT - Each question is a separate object in the questions array:

AskQuestion({
  "title": "Initial Information",
  "questions": [
    {
      "id": "requirements",
      "prompt": "What needs to be built/fixed/cleaned up?",
      "options": [
        {"id": "provide", "label": "Let me describe the requirements"},
        {"id": "skip", "label": "Skip to next question"}
      ]
    },
    {
      "id": "search_codebase",
      "prompt": "Should I search the codebase for relevant patterns?",
      "options": [
        {"id": "yes", "label": "Yes, search the codebase"},
        {"id": "no", "label": "No, skip"}
      ]
    },
    {
      "id": "search_online",
      "prompt": "Should I search online for established patterns/solutions?",
      "options": [
        {"id": "yes", "label": "Yes, search online"},
        {"id": "no", "label": "No, skip"}
      ]
    }
  ]
})

Or for sequential questions, use separate AskQuestion calls:

Step 1: AskQuestion({
  "title": "Start",
  "questions": [{
    "id": "start",
    "prompt": "Ready to provide requirements?",
    "options": [
      {"id": "proceed", "label": "Yes, I'll provide requirements now"},
      {"id": "abort", "label": "Cancel"}
    ]
  }]
})

Step 2: If proceed, ask conversationally for requirements

Step 3: AskQuestion({
  "title": "Research Options",
  "questions": [{
    "id": "research",
    "prompt": "Should I research before identifying tickets?",
    "options": [
      {"id": "both", "label": "Search codebase AND online"},
      {"id": "codebase", "label": "Search codebase only"},
      {"id": "online", "label": "Search online only"},
      {"id": "skip", "label": "Skip research"}
    ]
  }]
})

Common Patterns

Yes/No Question

AskQuestion({
  "title": "Confirm Action",
  "questions": [{
    "id": "proceed",
    "prompt": "Would you like to proceed?",
    "options": [
      {"id": "yes", "label": "Yes"},
      {"id": "no", "label": "No"}
    ]
  }]
})

Multiple Choice

AskQuestion({
  "title": "Select Option",
  "questions": [{
    "id": "approach",
    "prompt": "Which approach would you prefer?",
    "options": [
      {"id": "option1", "label": "Option 1: Fast but basic"},
      {"id": "option2", "label": "Option 2: Slower but comprehensive"},
      {"id": "option3", "label": "Option 3: Custom (I'll specify)"}
    ]
  }]
})

With Context in Question

AskQuestion({
  "title": "Lint Errors Found",
  "questions": [{
    "id": "lint",
    "prompt": "Found 3 lint errors in the code:\n\n- Line 45: unused variable\n- Line 67: missing semicolon\n- Line 89: indentation\n\nHow would you like to proceed?",
    "options": [
      {"id": "fix", "label": "Fix automatically"},
      {"id": "manual", "label": "I'll fix them manually"},
      {"id": "ignore", "label": "Ignore for now"}
    ]
  }]
})

Multiple Questions in One Call

Only use multiple question objects when the questions are truly independent and can be answered together:

AskQuestion({
  "title": "Configuration",
  "questions": [
    {
      "id": "location",
      "prompt": "Where should the file be saved?",
      "options": [
        {"id": "local", "label": "Local directory"},
        {"id": "cloud", "label": "Cloud storage"}
      ]
    },
    {
      "id": "format",
      "prompt": "What format should I use?",
      "options": [
        {"id": "json", "label": "JSON"},
        {"id": "yaml", "label": "YAML"}
      ]
    }
  ]
})

Important: If one question depends on the answer to another, use separate AskQuestion calls instead.

Multi-Select

AskQuestion({
  "title": "Select Features",
  "questions": [{
    "id": "features",
    "prompt": "Which features would you like to include?",
    "options": [
      {"id": "auth", "label": "Authentication"},
      {"id": "db", "label": "Database"},
      {"id": "api", "label": "API endpoints"},
      {"id": "tests", "label": "Test suite"}
    ],
    "allow_multiple": true
  }]
})

Converting Conversational to AskQuestion

Example 1: Simple Confirmation

Before (conversational):

Ask the user: "Does this look good?"

After (AskQuestion):

AskQuestion({
  "title": "Confirm",
  "questions": [{
    "id": "confirm",
    "prompt": "Does this look good?",
    "options": [
      {"id": "yes", "label": "Yes, looks good"},
      {"id": "no", "label": "No, let me adjust"}
    ]
  }]
})

Example 2: Open-Ended with Follow-Up

Before (conversational):

Ask the user for the branch name.

After (AskQuestion + conversational):

AskQuestion({
  "title": "Branch Name",
  "questions": [{
    "id": "branch",
    "prompt": "How would you like to specify the branch name?",
    "options": [
      {"id": "provide", "label": "Let me specify it now"},
      {"id": "current", "label": "Use current branch"},
      {"id": "abort", "label": "Cancel"}
    ]
  }]
})

If "provide" is selected, then ask conversationally for the branch name.

Example 3: File Selection

Before (conversational):

Ask which files to include.

After (AskQuestion):

AskQuestion({
  "title": "Select Files",
  "questions": [{
    "id": "files",
    "prompt": "Which files should I include?",
    "options": [
      {"id": "all", "label": "All changed files"},
      {"id": "staged", "label": "Only staged files"},
      {"id": "specific", "label": "Let me specify files"}
    ]
  }]
})

Based on the response, handle accordingly.

When to Combine with Conversational

Some questions require free-form input (file paths, commit messages, descriptions). In these cases:

1. Use AskQuestion to ask HOW they want to provide it
2. Then use conversational follow-up for the actual input

Example:

**Step 1**: Use AskQuestion to ask if they want to provide a custom message

**Step 2**: If yes, ask conversationally: "What message would you like to use?"

Handling Responses

After calling AskQuestion, you receive the user's selection. Document the mapping:

Based on the response:

- "yes" → Proceed with the action
- "no" → Cancel and exit
- "custom" → Ask conversationally for custom input

Critical Rules

1. Never skip AskQuestion when you have 2+ clear options to present
2. Always provide at least 2 options (even if one is "Cancel" or "Abort")
3. Use clear, action-oriented labels ("Yes, proceed" not just "Yes")
4. Include context in the question text when needed (show errors, file lists, etc.)
5. Make titles short (2-5 words max)
6. Keep option labels concise but descriptive

Anti-Patterns to Avoid

❌ Batching multiple questions into one prompt string

Questions:

- id: "info", prompt: "I need:
  1. What should I build?
  2. Search codebase?
  3. Search online?", options: [...]

Fix: Each distinct question should be either a separate AskQuestion call or a separate question object in the questions array.

❌ Asking conversationally when options are clear

Ask the user if they want to continue.

❌ Single option (not really a choice)

Options:

- id: "ok", label: "OK"

❌ Vague labels

Options:

- id: "1", label: "Option 1"
- id: "2", label: "Option 2"

❌ Too many questions at once (limit to 2-3 per AskQuestion call unless they're truly independent)

Examples from Real Skills

From dev-workflow-prepare-commit

AskQuestion({
  "title": "Unstaged Changes Detected",
  "questions": [{
    "id": "unstaged",
    "prompt": "The following files have unstaged changes: [list]. Are these intentionally left unstaged?",
    "options": [
      {"id": "leave", "label": "Yes, leave them unstaged"},
      {"id": "stage", "label": "No, stage all changes"},
      {"id": "review", "label": "Let me review"}
    ]
  }]
})

From dev-workflow-initialize

AskQuestion({
  "title": "Which Ticket?",
  "questions": [{
    "id": "ticket",
    "prompt": "What Jira ticket would you like to start?",
    "options": [
      {"id": "ticket", "label": "Specify a ticket number (e.g., RETIRE-123)"},
      {"id": "unticketed", "label": "Unticketed work (use RETIRE-1908)"}
    ]
  }]
})

From address-pr-feedback

AskQuestion({
  "title": "Comment 3 of 7 - Next Steps",
  "questions": [{
    "id": "action",
    "prompt": "Here's the proposed fix and draft reply. What would you like to do?",
    "options": [
      {"id": "apply", "label": "Apply the proposed fix"},
      {"id": "post", "label": "Post the reply on GitHub"},
      {"id": "both", "label": "Apply fix AND post reply"},
      {"id": "skip", "label": "Skip this comment"},
      {"id": "adjust", "label": "Let me adjust something"}
    ]
  }]
})

Summary

Default behavior: If you're about to ask the user anything, stop and ask yourself:

1. Can I present this as 2+ clear options? → Use AskQuestion
2. Do I need free-form input? → Use AskQuestion to ask HOW, then conversational for the actual input
3. Is this truly open-ended with no predictable answers? → Only then use conversational

Remember: AskQuestion provides a better user experience with clear, clickable options instead of having to type responses.