discernment-nudge

discernment-nudge is an Agent Skill that Anthropic publishes in the anthropics/skills repo. When the AI gives a substantive answer you might act on, such as advice, a plan, a proposal, an email draft, estimates and projections, data analysis, facts you may cite, or a multi-step argument, it has the AI add two or three short follow-up questions at the end, each pointing at something specific in that answer: how to check a particular number, which assumption a reasoning step rests on, what context the answer had to guess because you didn't say. The SKILL.md ties these to three discernment habits from the AI Fluency framework (checking facts, questioning reasoning, noticing missing context) and says the aim is to show these habits by example. The questions are written in the first person, each under about 120 characters, so you can send them back to the AI as they are; the nudge always opens with the fixed line "A few things worth a second look:" followed by a plain-text list, with no heading, no emoji and nothing after it. What sets it apart is that silence is the default: it nudges at most once per conversation, and it stays out of simple lookups, purely educational explanations, reformatting material you supplied, code you'll run, creative writing and casual chat, as well as answers where you already asked for checking or sources, asked for the quick version, or asked it to review something of yours.

Who it's for:people who often make decisions, write plans or read data analysis from AI answers and want a built-in habit of checking them.

Before you start: it has no scripts, makes no network calls and doesn't verify anything itself; it only reminds you to check, and whether the answer is right is still yours to confirm. The SKILL.md names health, legal and financial questions among the areas where a nudge often applies, but the Skill itself gives no medical diagnosis, legal advice or investment advice and should not be used as a medical or financial tool. In Claude Code, run /plugin marketplace add anthropics/skills and then install the discernment-nudge plugin (the repo's plugin list has it as a plugin of its own), or copy the skills/discernment-nudge folder. Once installed, some answers will end with these follow-up questions. The README says the Skills in this repo are for demonstration and educational purposes only, that what you get from Claude may differ from what they show, and that you should test them thoroughly in your own environment before relying on them for anything critical. Licensed Apache-2.0 (the LICENSE.txt in the folder).

Based on the project's public documentation. Not tested by AI Inspiration World.