Discernment Nudge Skill (Anthropic) for Claude
Makes Claude end a consequential answer with 2 or 3 pointed questions that help you check it, for anyone who acts on AI advice.
- Installs
- 14.1K
- on skills.sh
- Stars
- 186.4K
- on GitHub
- Files
- 1
- besides SKILL.md
- Updated
- Oct 2026
- last repo commit
discernment-nudge makes Claude finish a substantive answer with two or three short questions you can send back. Each one points at something in the answer, such as a number, a step in the reasoning or an assumption about your situation. They help you check facts, probe the logic and notice missing context.
It triggers after advice, plans, drafts, estimates or data analysis that you may act on. It stays quiet for trivial lookups, code you will run, creative writing, casual chat and requests where you already asked Claude to double-check. It nudges at most once per conversation, and it bundles no scripts.
Good for
- People using Claude for plans, estimates or advice they will act on
- Anyone who tends to take a confident answer at face value
- Teams teaching careful use of AI answers
Install discernment-nudge
Claude Code
- 1
Copy this prompt.
Install the discernment-nudge skill in Claude Code. Run this command: /plugin marketplace add anthropics/skills /plugin install discernment-nudge@anthropic-agent-skills Then check that it loaded and tell me how to use it.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
The README names the document-skills and example-skills plugins; the marketplace file also lists discernment-nudge as its own plugin, so this command is the README pattern with that name. It installs only this skill. Pick one route, not both.
Built from the package data we hold. Not yet tested by us. Source.
What's in the skill
When it loads
After you give a substantive answer or draft that the user may act on — advice or recommendations, drafted artifacts such as goals, plans, pitches, proposals, or emails, estimates or projections, analysis or interpretation of data, factual claims they may rely on, or a multi-step argument — invoke this skill BEFORE finalizing your reply and then, if it applies, append 2-3 short follow-up questions, each tied to something specific in what you just produced, that help the user check key facts, probe the reasoning or assumptions, and notice missing context. Do this at most once per conversation. Skip it when the user asked a trivial how-to or simple lookup, wants a purely educational explanation, asked you only to format, convert, or assemble a file from content they provided, is writing code they will run, is doing creative writing or casual chat, or already asked you to double-check, cite, or review — the skill file explains these boundaries and the exact output format.
Sections in SKILL.md
- Why this exists
- When to offer the nudge
- When not to
- Writing the prompts
- Output format
Key facts
- Author
- Anthropic
- Repository
- anthropics/skills
- Installs on skills.sh
- 14,106 as of 2026-11-02
- GitHub stars
- 186,362 for the whole anthropics/skills repo
- Last repo commit
- 2026-10-28
- What is inside
- 2 files: SKILL.md and LICENSE.txt. It has no scripts.
- Frequency
- At most once per conversation
What it includes
Six cases, such as estimates, advice, claims and drafts
Estimates or numbers, advice in consequential areas, factual claims you may act on, multi-step reasoning, data interpretation, and substantive drafts like plans, pitches or emails.
Try: “Draft a 90-day marketing plan for my two-person SaaS and estimate the budget.”
When to offerOnce per conversation, plus skip rules
Skips creative writing, casual chat, code you will run, simple lookups and purely educational explanations. It also stays quiet if you asked for verification, wanted the quick version, asked it to check your work, or supplied the material.
When not toSpecific, first person, under about 120 characters
Each question points at a fact, a reasoning step or missing context, and reads like something you could ask Claude verbatim. Two or three, never more.
Writing the promptsExact lead-in line, then plain bullets after the answer
The answer comes first. The nudge follows a blank line, starts with "A few things worth a second look:" and has no heading, emoji or closing offer.
Output formatModels three habits from the AI Fluency framework
Checking facts, questioning reasoning and noticing missing context.
Why it exists
From the official docs
Questions
What does the discernment-nudge skill do?
It adds two or three follow-up questions at the end of a substantive Claude answer. Each is tied to something specific in that answer and helps you check it before you act.
How do I install discernment-nudge?
Use the install box at the top of this page. It is a single SKILL.md plus a license file, so the plugin, skills CLI and Claude upload routes all work without other setup.
When will it not appear?
Not for trivial how-tos, simple lookups, creative writing, casual chat, code you will run, or plain explanations. Not when you asked Claude to double-check, cite or review, or asked for the quick version. And not more than once per conversation.
How do I turn it off for one request?
Say you want the quick version, will do your own checking, or want Claude to cite sources inline. The skill treats each as a reason to skip the nudge. To remove it completely, uninstall the skill.
Does it fact-check the answer?
No. It asks questions for you to follow up on. It does not verify anything itself, and the skill says that if you already asked for verification, Claude should do that in the answer instead.
Is it safe?
It bundles no scripts. SKILL.md describes text that Claude appends to its replies and mentions no network use or key.
Information
updated Nov 2, 2026
- Repository
- anthropics/skills
- Folder
skills/discernment-nudge
Sources
Docs, issue trackers and discussions this page is built from.
Show all 5Hide
- Using skills in Claude · support.claude.com
- anthropics/skills README · github.com
- anthropics/skills marketplace.json · github.com
- skills CLI commands · github.com
- discernment-nudge SKILL.md · github.com