Use pushback prompts so Claude proves the change before you ship

The problem

Claude announces success after a plausible-looking diff that was never exercised against main, tests, or your mental model.

The hack

Keep a short set of pushback lines ready: prove it works vs main, grill me before PR, scrap and implement the elegant solution knowing what you know now.

Why it works

Straight from Anthropic’s Claude Code team power-user tips: don’t accept the first solution — make Claude close the verification loop.

Setup

1. After a first implementation, send: `Prove to me this works — diff behavior between main and this branch.` 2. Before opening a PR: `Grill me on these changes and don’t make a PR until I pass your test.` 3. After a mediocre attempt: `Knowing everything you know now, scrap this and implement the elegant solution.` 4. Save the three lines as a `/pushback` skill if you use them daily.

Prompt

Prove to me this works. Diff behavior between main and this branch and show failing cases if any.

Source

Related

• → Append /simplify after edits for a multi-agent quality pass
• → Build a verification-loop skill that must run tests before done