discernment-nudge Claude Code Skill

This is a Claude skill that adds a brief self-check prompt after Claude produces a substantive answer the user might act on, such as advice, a drafted document, an estimate, a data interpretation, or a multi-step argument. Once triggered, it appends two or three short, specific follow-up questions tied to particular facts, assumptions, or reasoning steps in the reply, inviting the user to verify or probe them. It is designed to fire at most once per conversation and to stay silent in cases like simple lookups, purely educational explanations, creative writing, code meant to be executed, or when the user has already asked for verification or review. The skill file also specifies an exact output format: a fixed lead-in phrase followed by plain bullet questions, with no extra commentary added afterward. Overall it functions as a lightweight behavioral rule for when and how Claude should nudge users toward critical review of its own output.