Append targeted reflection prompts to substantive AI answers.
npx skills add https://github.com/anthropics/skills --skill discernment-nudge
Please help me install this Agent Skill. Skill: discernment-nudge Source: https://github.com/anthropics/skills/tree/main/skills/discernment-nudge Command: npx skills add https://github.com/anthropics/skills --skill discernment-nudge
Users often accept AI-generated answers at face value, missing bad assumptions or missing context before acting on consequential advice, estimates, or drafted artifacts. This Skill adds a brief, specific reflection moment without disrupting the answer itself.
Ask the AI for a business strategy recommendation and review the appended reflection prompts before acting on the advice.
High-intent search queries and answers about installing and using this skill.
Use a skill that triggers after substantive answers and appends two or three short follow-up questions tied to specific claims, reasoning steps, or assumptions in the response. The prompts should be phrased as first-person questions the user could send back verbatim.
Include them when the answer contains estimates, consequential advice, factual claims the user will act on, multi-step reasoning, data interpretation, or drafted artifacts. Skip them for trivial lookups, creative writing, code execution, educational explanations, and when the user already asked for verification.
Yes, by limiting the nudge to once per conversation, keeping prompts under 120 characters each, and tying each prompt to something specific in the answer rather than generic questions like 'can you verify those facts?'
AI models generate plausible-sounding text based on patterns rather than verified facts, and they often fill in missing context with assumptions. Surfacing those assumptions as reflection prompts helps users catch errors before acting on the answer.
No, caveats should be reserved for substantive, actable answers. Adding them to trivial lookups, creative writing, or educational explanations creates noise and can override explicit user requests for quick, unhedged responses.
cosmicdreamsGathers web evidence into a NotebookMLab notebook for brainstorming preparation.
OneZero1aiQuery a local knowledge commons to validate guidance before acting.
meiiieCross-reference factual claims against multiple authoritative sources with citations.
zhou210712Draft litigation document sections with citation and verification tags.
