Discernment Nudge
Add a thoughtful checkpoint to consequential AI answers
Install
Officialhttps://github.com/anthropics/skills/tree/main/skills/discernment-nudge
Activates after consequential advice, analysis, projections, or drafts that may benefit from a brief review of facts and assumptions.
- Author
- Anthropic
- License
- Apache-2.0
Discernment Nudge gives an agent a lightweight way to encourage reflection after producing material that could influence a real decision. Instead of interrupting the main response with broad warnings, it identifies specific claims, estimates, assumptions, or reasoning steps that may deserve closer examination.
Highlights
- Distinguishes consequential guidance from simple lookups, casual conversation, creative work, and routine formatting.
- Generates a short set of questions tailored to details in the agent’s latest response.
- Draws attention to uncertain figures, hidden assumptions, missing context, and fragile reasoning chains.
- Avoids redundant prompts when the user has already requested verification, citations, or a critical review.
- Requires no external tools, MCP servers, scripts, or executable helpers.
The skill is useful for teams building assistants that handle planning, research interpretation, business decisions, professional communication, or other material users may act upon. It offers a restrained alternative to generic disclaimers by making the review step specific to the answer at hand. The folder can be obtained directly from Anthropic’s public skills repository and added to a compatible Agent Skills environment.
Similar skills
Brainstorming
SkillSettles intent and design with the user before implementation is allowed to start.
Receiving Code Review
SkillRequires understanding and checking review feedback before acting on any of it.
Code Review
SkillMulti-agent pull request review with confidence scoring to filter out false positives.

Comments(0)
Sign in to comment