Install
invoke this skill BEFORE finalizing your replyWho it's for
Call discernment-nudge after you have just given a substantive answer or draft the user may act on. The official identity is the YAML field name: discernment-nudge, and the description starts from After you give a substantive answer or draft that the user may act on.
It must be called before you finalize. When it applies, append 2-3 short follow-up questions that model the three discernment habits of the AI Fluency framework, rather than lecturing. It fits replies that will be used for decisions or shared outside. Skip trivial how-to, simple lookup, pure explanation, format-only or convert-only work or assembling a file from content the user already gave, code the user will run, creative writing, and casual chat — the body says to skip those.
Page: https://raw.githubusercontent.com/anthropics/skills/main/skills/discernment-nudge/SKILL.md
First try in 3 minutes
The body gives no installer and no local directory. The only official path you can point at is this Skill body: https://raw.githubusercontent.com/anthropics/skills/main/skills/discernment-nudge/SKILL.md
Preconditions: you are about to produce a substantive answer or draft the user may act on; this conversation has not already done this follow-up. Do not invent a copy step, and do not name a directory that is not written down.
Do a read-only check first. Ask the agent: please give a substantive suggestion the user may act on, and call discernment-nudge before finalizing. The success signal is 2-3 short questions after the formal reply, and those questions model discernment habits rather than explaining the framework. If the agent starts lecturing, it is not following the body.
Do this only once in the same conversation. If you next ask a trivial how-to or simple lookup, the agent should skip and not append a second round of questions.
Use cases
Case 1: A substantive, actionable reply. The body requires invoke this skill BEFORE finalizing your reply, and when it applies append 2-3 short follow-up questions. When a user brings a plan they are ready to adopt and asks for comment, write the advice first, then append questions before you finalize. At most once in the whole conversation.
Case 2: A request that must be skipped. The body lists: trivial how-to or simple lookup, purely educational explanation, only format / convert the user's existing content or assemble a file, is writing code they will run, creative writing or casual chat. Hand the agent a trivial how-to or simple lookup; the success signal is that the reply has none of these 2-3 discernment follow-ups.
Usage limits
At most once per conversation: Do this at most once per conversation.
Skip conditions follow the original wording; do not widen or narrow them yourself. The goal is to model three discernment habits, not to lecture.
The body has no installer, no client roster, and no on-disk directory. Do not treat a directory fragment inside the page URL as an install location, and do not add a copy command.
The body does not expand the three habit names, question templates, or a supported-agent roster item by item; do not fill in what is not written.