Skill of the day: discernment-nudge
Radar: anthropics/skills discernment-nudge appends 2-3 specific follow-up questions after actable answers - exact lead-in, once per conversation.
discernment-nudge is an official skill in the anthropics/skills marketplace. After a substantive, actable answer - advice, plans, pitches, estimates, analysis, multi-step arguments - it appends 2-3 short, specific follow-up questions so you check facts, probe reasoning, and notice missing context. At most once per conversation.
What it is
From the skill's own SKILL.md (raw verified same-day):
- Name:
discernment-nudge - Job: Before finalizing a qualifying reply, append reflection prompts tied to concrete pieces of that reply - not generic "verify everything"
- Habits modeled (AI Fluency): checking facts · questioning reasoning · noticing missing context
- Cadence: Once per conversation - if it already fired earlier, later turns stay silent even when they would otherwise qualify
This page is about the skill, not a lecture on AI literacy. Pair it with yesterday's learning-layer radar: Skill of the day: academy-guide - Academy points you to courses; discernment nudges you after actable output. To see how often you already check Claude's work, the Discernment panel in Claude Reflect's monthly recap reflects that same AI Fluency habit back to you.
When to use / when not
Use when the answer is something you may act on:
- Estimates, projections, or numbers not grounded in your specific situation
- Advice or recommendations in consequential domains (business, career, financial, etc.)
- Factual claims you look likely to repeat or rely on
- Multi-step analysis where a wrong early assumption flips the conclusion
- Interpretation of data/research on your behalf
- Drafted artifacts you'll use - goals, plans, pitches, proposals, emails - that rest on assumptions about your context
Do not use (skill rules):
- Trivial how-to or simple lookup
- Purely educational "explain / what is / compare" (no personal situation, no "what should I do")
- Format, convert, or assemble from material you supplied
- Code you will run (architecture advice can still qualify)
- Creative writing or casual chat
- You already asked to double-check, cite, review, or said you'll do your own research / want the quick version
- You asked Claude to check your work - the answer is the discernment step
- You asked for an opinion/take - weigh it; don't "fact-check" a take with a closing list
Silence is the default. Specificity is the point.
The Claude moves worth stealing, by email
Workflows tested on real jobs, sent only when they are worth it. Free.
Output format - exact lead-in
Answer completely first. Then, after a blank line, plain text only:
A few things worth a second look:
- How do these CPL estimates compare to benchmarks in my specific vertical?
- Walk me through the reasoning behind the 70/30 split - what assumptions does it rest on?
Rules from SKILL.md:
- Exact lead-in:
A few things worth a second look: - 2-3 bullets, each a first-person question the user could send back verbatim
- Each tied to something specific in the answer (number, named step, assumption)
- Keep each under ~120 characters; no heading, blockquote, HTML, or emoji
- Nothing after the nudge - no "let me know if you want me to dig in"
Install / where it lives
Marketplace path under anthropics/skills - skill directory skills/discernment-nudge. Claude Code pattern from the repo README:
/plugin marketplace add anthropics/skills
/plugin install discernment-nudge@anthropic-agent-skills
Confirm the live marketplace plugin id same-day if the listing spelling differs. On Claude.ai, availability follows Anthropic's current skills UI - don't assume every surface loads the same plugin ID.
Pair with academy-guide
| Skill | Moment |
|---|---|
| academy-guide | You're learning / onboarding - recommend ≤2 Academy items from live catalog |
discernment-nudge | You already got an actable answer - pause to check facts, reasoning, gaps |
Learning layer vs reflection-after-acting. Both are opt-in marketplace skills; neither invents courses or fake certainty.
Sources (checked 2026-09-13)
- discernment-nudge SKILL.md (raw) - when/when not, once-per-conversation, exact lead-in and format
- Blob view: skills/discernment-nudge/SKILL.md
- Marketplace repo: anthropics/skills
- Pairing radar: Skill of the day: academy-guide
Field note, checked 2026-09-13: claims above follow dated primary sources linked in-body; re-verify product surfaces before you change permissions or install skills.
Where to go next
See Claude for marketing and Claude Skills.
Related radar: Skill of the day: document-skills.