Check AI answers with follow-up questions
Get a few specific follow-up questions when an estimate or recommendation deserves a closer look. Each question refers to something in the answer you just received.
You can check AI answers with follow-up questions before acting on a recommendation. With discernment-nudge, the question points back to something the agent actually said. If a delivery estimate assumes two editors, a question about staffing can be more useful than a general request to double-check the schedule.
It adds a short set of prompts after the original answer, at most once in a conversation. The prompts can target a factual claim, a step in the argument or something the agent had to assume about your circumstances.
What it does
- Identify a figure that needs a comparison
- Ask what a recommendation depends on
- Trace the reasoning behind a priority
- Check whether missing context changes a plan
- Choose one issue to explore in the next turn
Install and use
- Install the skill in a compatible agent and ask about a decision you intend to make. It needs no separate fact-checking service or API key.
- For example: “Estimate the editing time for these lessons, assuming one editor is available.”
- Read the answer first. Pick a suggested follow-up if it would change what you do next.
FAQ
Source
Made by Anthropic. Original skill: discernment-nudge, published on GitHub under the Apache-2.0 licence.
View on GitHub