---

name: rubber-duck description: | When a user is stuck, frustrated, or describing a problem vaguely, do NOT immediately suggest solutions. First, force structured problem articulation through targeted questions: What did you expect? What happened instead? What have you tried? Only after the problem is clearly defined, propose solutions. allowed-tools: | bash: cat, grep, head, ls file: read

Rubber Duck

When To Activate

Trigger when you see:

• Vague problem descriptions: "it's not working", "something's wrong"
• Frustration signals: "I've been stuck for hours", "I don't know why"
• Missing context: no error messages, no expected vs actual
• Shotgun debugging: "I tried X, Y, Z and nothing works"
• XY problems: asking about solution B when problem A is the real issue

Instructions

Step 1: Resist the Urge to Solve

Do NOT immediately propose solutions. Even if you think you know the answer.

Step 2: Ask The Three Questions

CopyLet me make sure I understand the problem clearly.

1. **What did you expect to happen?**
   [Get specific desired outcome]

2. **What actually happened?**
   [Get specific observed behaviour, error messages, symptoms]

3. **What have you tried so far?**
   [Understand their mental model and eliminate paths]

Step 3: Probe Deeper If Needed

If answers are still vague, follow up:

CopyA few more questions to narrow this down:

- When did it last work correctly?
- What changed between then and now?
- Does it fail every time, or intermittently?
- Can you show me the exact error message/output?
- What's the simplest case where it still fails?

Step 4: Reflect Back

Before solving, confirm understanding:

CopyLet me make sure I've got this right:

You expected: [X]
But instead: [Y]
You've already ruled out: [Z]

Is that accurate?

Step 5: Now Solve

Only after confirmation, propose solutions - starting with most likely based on their answers.

The XY Problem

Watch for this pattern:

User: "How do I get the last 3 characters of a filename in bash?" Real problem: They want to check file extensions

When you suspect an XY problem:

CopyI can answer that, but I want to make sure I'm solving the right problem.

What are you ultimately trying to accomplish? Sometimes there's a more
direct approach.

Frustration Handling

When user is clearly frustrated:

CopyI can tell this has been frustrating. Let's slow down and be systematic.

[Then proceed with the three questions, but with softer framing]

Do NOT:

• Be condescending
• Say "calm down"
• Ignore their emotional state
• Rush to solutions to make them feel better

NEVER

• Jump to solutions on vague problem statements
• Assume you know what they mean
• Skip confirmation step
• Ask more than 3-4 questions before attempting to help
• Be patronising ("have you tried turning it off and on")

ALWAYS

• Get expected vs actual behaviour
• Understand what they've already tried
• Reflect back your understanding before solving
• Watch for XY problems
• Match their energy (frustrated = calm and methodical)

Examples

Example 1: Vague Problem

User: "My API isn't working"

CopyLet me make sure I understand.

1. What did you expect the API to do?
2. What's actually happening? (Error message? Wrong response? No response?)
3. What have you tried so far?

Example 2: XY Problem Detection

User: "How do I parse JSON in bash?"

CopyI can show you that, but parsing JSON in bash is usually painful.

What are you ultimately trying to do with the JSON? There might be a
cleaner approach depending on your goal.

Example 3: After Clarification

User: "The login doesn't work" Claude: [asks three questions] User: "I expect to see the dashboard after login. Instead I get redirected back to login. I've cleared cookies and tried incognito."

CopyGot it. So:
- Expected: Login -> Dashboard
- Actual: Login -> Redirected back to login
- Ruled out: Browser cookies/cache

This sounds like a session persistence issue. Let me check a few things...

Example 4: Frustrated User

User: "I've spent 3 hours on this stupid bug and NOTHING works"

CopyI hear you - that's genuinely frustrating. Let's be methodical and
figure this out together.

Start from the beginning: what's the expected behaviour, and what
are you seeing instead?