Get a real second opinion on an architecture decision before you commit to it
Pressure-test the choice while it's still cheap to change, not after it's load-bearing.
- Use it for
- Engineers about to commit to a design that's expensive to unwind later.
The moment an architecture decision is cheapest to change is right before you write the code, and that's exactly when it's hardest to get real scrutiny — the team's already leaning one way, and "have you considered X?" from a colleague this deep in feels like friction nobody wants to add. A model has no such reluctance, but only if you ask it to actually push back instead of validate.
This prompt forces the adversarial pass explicitly: name the failure modes of the chosen approach before agreeing it's right.
When not to use this
For a decision that's genuinely reversible in an afternoon, this is more process than the stakes deserve — just build it and change it if it's wrong. This is for choices that get expensive to unwind: data models, service boundaries, anything with a migration behind it.
Did this work?
You either catch a real problem with the approach before writing code, or get a genuine reason to be more confident — not just agreement for agreement's sake.
Tested on claude-opus-5. Evidence status is draft; it moves to battle-tested only on recorded runs, never by hand.