Home / All Skills / Workflow & Core Tools / discernment-nudge
English (EN)简体中文 (ZH)日本語 (JA)
Workflow & Core Tools · anthropics/skills

discernment-nudge

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.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit
Author anthropicsGitHub anthropics/skills ↗Stars 175386License Apache-2.0（自带 LICENSE.txt；Copyright 2026 Anthropic, PBC.）Commit 41bbe19d1a
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

The mechanism functions as a 'post-response self-check hook': before finalizing a substantive response that the user might act upon (recommendations, drafts, estimates, factual claims, multi-step arguments), the model evaluates whether to append 2-3 follow-up verification questions.

skills/discernment-nudge/SKILL.md
The goal is to *model* three discernment habits from the AI Fluency framework, not to lecture about them
Three modeled habits: Checking facts / Questioning reasoning / Noticing missing context. Rather than generic 'please verify' remarks, follow-up questions instantiate these habits by targeting specific numbers, reasoning steps, and assumptions in the current response.

Silence is the default, and execution is limited to at most once per conversation: the vast majority of responses receive no nudge, and once added, it is never repeated within the same conversation.

skills/discernment-nudge/SKILL.md
**Once per conversation.** Offer the nudge at most once in a conversation. If you have already offered it on an earlier turn, stay silent on later turns even when the new answer would otherwise qualify
'Silence is the right default' and 'repeating it turns a light suggestion into nagging' serve as the core rationale for frequency control.

Triggering and suppression rely on an 'exclusion checklist' rather than arbitrary thresholds: consequential domains, numerical estimates, citeable facts, multi-step reasoning, data interpretations, and substantive drafts are positive triggers; exemptions apply to creative writing, casual banter, code the user will execute, simple lookups, purely educational explanations, and four 'user-already-covered' scenarios (asking to verify, asking for a quick version, asking to review, or providing source materials).

skills/discernment-nudge/SKILL.md
Offer it when your answer contains content the user would benefit from scrutinizing before acting on it. The clearest cases:
Exemption examples from the prompt: 'Code the user will execute — running it is the verification.' and 'The user asked you to verify, cite, or flag uncertainty. If their question included "double-check,"…they've already put themselves in a critical frame.' The design intent is to avoid patronizing users who are already adopting a critical mindset.

Follow-up questions must be tethered to specific content: each question points directly to a number, named step, or assumption in the preceding answer, strictly forbidding generic template questions.

skills/discernment-nudge/SKILL.md
Generic prompts ("Can you verify those facts?") defeat the purpose; the value is in the specificity.
Distinct example phrases are provided for three prompt types (fact/figure checking, reasoning step/assumption probing, missing context cueing); users can send these questions back to the model verbatim.

Output format is hardcoded: a fixed lead-in line followed by plain text bullet points (2-3 items, ≤120 characters each), placed immediately after the completed response, with nothing appended afterward.

skills/discernment-nudge/SKILL.md
Use that exact lead-in line — "A few things worth a second look:" — followed by the prompts as plain bullets. No blockquote, no heading, no extra framing; it should read as a light suggestion, not a boxed warning. Plain text only — no HTML, no headings, no emoji.
The formatting template section provides example code blocks; 'Don't add anything after the nudge' ensures the nudge always acts as the definitive conclusion to the response.

2Core Capabilities

01Automatically appends 2-3 verification follow-up questions to actionable, substantive responses (estimates, advice, factual claims, multi-step reasoning, data interpretations, drafts)
02Models three types of follow-up inquiries: fact verification (how to check / comparing against user's own data), reasoning critique (underlying assumptions), and missing context prompts
03Do-Not-Disturb exemption filter: skips trivial how-to questions, educational explanations, executable code, and cases where the user requested verification, asked for brevity, or provided source materials
04Anti-fatigue rate limiting strictly enforced at at-most-once per conversation
05Ready-to-resend user follow-up questions (first-person voice, interrogative form, ≤120 characters, free of formatting noise)

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Can be auto-loaded
For Administrators
Register source and version
Evidence & Triggers (File Paths & Excerpts)
  • SKILL.md: Pure prompt instructions and rule text; involves no script execution or external side effects.
Audit Notes: Purely instructional conversational skill: no script execution, no outbound network requests, no credential reading, no local file read/write operations, and zero code assets (contains only SKILL.md + LICENSE.txt). Triggering relies entirely on the model's internal assessment during generation, without altering host permission boundaries. The only consideration is maintaining conversational tone and moderation, presenting no code-level security risks.

5Second-Pass Verification

  • [ok] 资产面：仅 SKILL.md + LICENSE.txt — glob 全量（含隐藏文件）确认无其他文件。
  • [ok] 零脚本/零网络/零凭证/零写入 — 全文 token 扫描零命中；无任何文件操作动词。
  • [ok] at-most-once 与 silence-default 存在 — 'Offer the nudge at most once in a conversation.' 与 'Silence is the right default' 原文均在。
  • [ok] 豁免清单与 description 一致 — description 列的 skip 场景（trivial how-to、educational、format/convert、code、creative、已要求 double-check）全部在 When not to 节有展开原文。
  • [ok] 输出格式硬规则 — lead-in 'A few things worth a second look:' 与 'Plain text only — no HTML, no headings, no emoji'、'under ~120 characters' 原文均在。
  • [ok] 夸大检查 — description 未承诺任何工具/外部数据能力，全部行为为对话文本追加——与正文能力相符，无越界。
  • [ok] commit/license/元数据 — 本地 HEAD==pin 41bbe19；LICENSE.txt Apache-2.0 © 2026 Anthropic PBC。

6Verdict & Scenarios

  • Lightweight with high leverage: adds a zero-cost layer of 'key assumption self-checking' to counter AI overconfidence and fake clarity.
  • User-respecting design: the exemption list excludes scenarios where users already asked to verify, wanted a quick version, or supplied source materials, eliminating patronizing lecturing.
  • High operational utility: each question targets specific numbers, steps, or assumptions and can be sent back verbatim, minimizing user cognitive friction.
  • Extreme restraint: capped at once per conversation, strictly formatted, and silent by default to prevent noise accumulation.
    • Best for: Suitable for high-stakes conversations where the user is likely to act on the response: estimates, budgeting, health, legal, financial, or career advice, critical email and proposal drafts, data interpretations, and multi-step arguments. Especially relevant when users might cite the AI's output as an authoritative source to others (e.g., reports, decision-making meetings).
    Not for: Not suitable for educational or curiosity-driven Q&A (explaining concepts, defining terms), code-execution tasks (where running the code is the verification), scenarios where the user explicitly asked for brevity or self-verification, creative writing, or casual banter—adding nudges in these contexts only introduces unwanted noise.
    Install Agent-Ready
    ① Verified Mirror 2026-09-09
    Option A · Direct DownloadDownload discernment-nudge.tar.gz
    sha256: 35e91c7b3cb30136…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 41bbe19d1a.