Discernment nudge

Discernment nudge gives the agent a short closing step for answers a user is likely to act on. After it gives advice, an estimate, an analysis of data, factual claims or a drafted plan, pitch, proposal or email, the agent decides whether a nudge fits. If it does, the reply ends with the lead-in line "A few things worth a second look:" and two or three plain bullets. Each bullet is a short first-person question tied to a specific number, step or assumption in the answer, written so the user could send it back as is.

The questions model three habits that the SKILL.md takes from the AI Fluency framework: checking facts, questioning reasoning and noticing missing context. The nudge appears at most once per conversation, and nothing follows it.

Much of the file covers when to stay quiet. Creative writing, casual chat, simple lookups, purely educational explanations and code the user will run get no nudge. The same goes for replies where the user asked for verification or sources, wanted the quick version, asked the agent to review their own work, supplied the material being summarized, or asked for an opinion. The result is a normal answer that sometimes ends with a few targeted questions.