Skip to content
SkillModel-led2,224 tokens in use

discernment-nudge

After a substantive answer the user may act on, Claude appends two or three specific follow-up questions that help check facts, probe reasoning and notice missing context.

197,763 stars on the repositoryanthropics/skills · skills/discernment-nudge

Skill label

measured from its files

What installing discernment-nudge costs

203tokens idle
2,224tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Model-led

Script-ledMixedModel-led

Mostly instructions, so how well it works depends on the model that follows them.

This describes the skill's structure, how much of it is scripts. It does not say how well it works.

Touches

Nothing detected: no shell, network, file writes, git or credentials in its files.

Declared tags come from the skill's frontmatter; detected ones from its scripts and code blocks. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Good
  • Frontmatter15/15
  • Structure11/25
  • Examples6/15
  • Output15/15
  • Size13/15
  • Resources10/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Excellent

No findings.

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity72
  • Maintenance32
  • Community68
  • Maturity36
  • Momentum40

discernment-nudge is an agent skill from Anthropic’s public skills repository that adds a brief moment of reflection to answers people are likely to act on. When Claude has given advice, an estimate, a factual claim, an interpretation of data, a multi-step argument or a drafted plan, pitch or email, it closes the reply with two or three short questions the user could send back, each tied to something concrete in that answer.

The skill models three discernment habits from the AI Fluency framework: checking facts, questioning reasoning and noticing missing context. Most of it is about when to stay silent. It skips creative writing, casual chat, code the user will run, simple lookups and purely educational explanations, and it stays out of the way when the user already asked Claude to verify or cite, asked for the quick version, asked Claude to review their own work, supplied the material being summarized, or asked for an opinion. It nudges at most once per conversation, in a fixed plain-text format.

What You Get

  • A trigger rule for answers worth scrutinizing: estimates and projections, consequential advice, factual claims the user may act on, multi-step reasoning, data interpretation and drafted plans, pitches, proposals or emails
  • A list of cases where the nudge is left off: creative writing, casual chat, code the user will execute, simple lookups and purely educational or definitional explanations
  • Four patterns where the user has already opted out: asked for verification or citations, asked for the quick version, asked Claude to check their work, or supplied the material
  • A once-per-conversation limit so the suggestion never turns into nagging
  • Guidance for writing the questions: specific to the answer, first person, question form, under about 120 characters, two or three at most
  • An exact output format: a fixed plain-text lead-in line and plain bullets, with nothing added after them

Common Use Cases

  • Following a cost estimate or timeline with questions about how the figures compare to the user’s own situation
  • Ending business, health, legal, financial or career advice with prompts about the assumptions it rests on
  • Closing a drafted plan, pitch or email with questions about context the draft had to guess at
  • Adding a check after Claude interprets data or research on the user’s behalf

What Makes It Unique

The other skills listed from the same repository, such as frontend-design and pdf, help Claude produce something: a UI or a processed document. discernment-nudge produces nothing of its own; it shapes how Claude ends an answer and is aimed at the user’s judgment rather than the task. Its weight sits in the exclusions: it spells out when a nudge would be noise or would override what the user already said, from a request to double-check to a request for an opinion, and it treats silence as the default. Where it does apply, it rejects generic prompts like “Can you verify those facts?” in favor of questions tied to a specific number, step or assumption.

Under The Hood

The skill is a single SKILL.md of instructions with no scripts or reference files; its folder also holds the license text. The frontmatter description asks Claude to invoke it before finalizing a substantive reply, and the body sets out the cases that qualify, the cases to skip, the once-per-conversation rule and the exact format. Claude applies it in its own writing: it decides whether the answer qualifies, then appends the lead-in line and two or three bullet questions. It runs no commands, calls no services and writes no files.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers