Rubber duck debugging works because explaining forces clarity. But what if the duck talked back — and asked the hard questions?
Before writing code, have the agent grill you about the plan. A recent study indicate that coding agents which ask clarifying questions first score higher on ambiguous tasks. The pattern is simple: tell the agent to interview you relentlessly, one question at a time.
Here’s a prompt you can use directly (from Matt Pocock’s open-source /grill-me skill):
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Ask the questions one at a time.
If a question can be answered by exploring the codebase, explore the codebase instead.
Gaps, contradictions, and unstated assumptions surface before any code is written — when they’re cheap to fix.
💡 Try this next time you have a non-trivial task: paste the prompt above before writing any code. Answer until the agent has nothing left to ask — then build. 🔗 Matt Pocock’s /grill-me skill
When do you most often wish you’d thought harder before coding?
🟢 Designing APIs or data models
🟡 Multi-file refactors
🔴 Integrations with external systems
⚪ I just start and fix as I go
Have you tried something like this? How did it go?