The AI Growth Mindset Coach skill for Claude

The role is sharply defined: a Dutch-speaking coach who analyses language for mindset and rewrites it into growth language. Not a therapist. Someone who does not excuse real problems, does not use forced positivity and never says you should just see it differently. Those three denials sit in the very first paragraph of the skill itself, before a single technique has been explained, and they set the tone for everything that follows.

The field of use the file names is broad and concrete: self-talk, feedback to staff, appraisal conversations, job listings, teaching material, team agreements and internal communication. In practice that means three kinds of users. The manager who has a conversation coming up and notices his draft sentence is about the person. The teacher or parent who wants to give compliments that do not backfire. And the business owner who hears himself say he is just not commercial, and realises that sentence is a conclusion, not an observation.

It belongs to the library of 100 free skills we make available through the AI and automation service, with no account and no sales email afterwards. One thing you should know upfront: this is a language tool, not an organisation tool. Growth language in a company where mistakes have consequences changes nothing. Then you are dressing up a culture instead of changing it. The file says so itself, and further down this page you can read why.

How you praise someone does something different from what you think. In Dweck's research line with Claudia Mueller, it turned out that praising process and strategy has different consequences from praising ability. Praise someone for being clever, and the next task becomes a test they can fail. Praise the approach, and that same task becomes practice material. That is why the skill has a separate variant that rewrites compliments: from you are clever to you adjusted that approach three times until it worked.

The word not yet moves a verdict onto a timeline. Dweck describes the power of yet: the difference between I cannot do this and I cannot do this yet. That is not a language trick, because the first sentence is a statement about who you are and the second about where you stand. That is why not yet is an anchor point in step 4, but only where it fits. A sentence where it does not fit does not get the word stuck onto it.

An approach can fail, a person cannot. That is the fifth fixed step in the file and the reason a rewrite is never kinder than the original, only more precise. The expectation is allowed to stay hard. In the example above, it still literally says the junior needs to have this team's pace under control within two sprints. What disappears is the verdict on his suitability, because that verdict makes the conversation impossible to answer.

Dweck has publicly qualified her own concept. She called it the false growth mindset: saying the words growth mindset without adjusting behaviour, feedback and appraisal. This skill takes that nuance seriously, which is exactly why it never skips the block What is not a mindset problem here. Growth language without an organisation that allows mistakes is an empty promise.

In diff form, with the sentences from the skill's own example: what goes out and what comes back in its place.

And for laments such as I put everything off because I am afraid it will not be good enough, I do not dare ask for feedback, every rejection feels like proof I cannot do it, my team does not take on difficult projects, everyone plays it safe. Anyone who says something like that to Claude while the skill is loaded gets it automatically. The same frontmatter also states straight away what it is not for, and we come back to that at DOC.06.

Then comes the core: seven steps the skill always carries out. It asks for the literal text or statement, because without real sentences there is no real analysis. It marks each sentence or fragment as fixed, mixed or growth language, with the reason attached. It names which mechanism sits behind the fixed wording. It rewrites every marked sentence into growth language, with not yet as an anchor point where it fits. It separates behaviour from person in every rewrite. It delivers one concrete action for today or tomorrow. And it checks whether the problem is actually a mindset problem: if it is about time, resources, an unclear assignment or a lack of safety, it says so and does not rewrite.

Six questions sit behind those seven steps. The mandatory input checklist asks for the literal text, for who is saying it and to whom, for the situation and what is at stake, for the goal (do you want to move someone into action, or mainly get yourself unstuck), for the relationship and the history (first conversation or the fifth about the same point), and for the question of how much room there really is: can the recipient actually change something, or is it beyond their control. Only what is missing gets asked afterwards, so you get at most two or three targeted questions rather than a form.

A striking number of rules concern the language of the answer itself. Ten writing rules fix that the answer is written in clear Dutch, in the informal je form rather than the formal u form, without dashes in running text, with AI always capitalised, with small language quirks so it sounds human, without rule-of-three lists where every list has exactly three items, without AI turns of phrase such as in the world of, discover, seamless or dive into, without hollow superlatives, with a number over an adjective, and with short sentences alternated with long ones. The tenth rule is the most important: avoid toxic positivity. Sentences like everything happens for a reason or just stay positive have no place here.

The file also holds five variants alongside the standard mindset scan, a separate mode for reviewing existing material in five steps, and a closing list of seven things the skill never does. Those three parts are covered below, because together they mark where this skill stops. If you want to learn to write a file like this yourself, the guide to writing SKILL.md NL explains how that structure works.

The foundation is Mindset from 2006 and Self-Theories from 1999. In the first book, Dweck translated her research for a broad audience; the second holds the academic foundation: implicit theories of intelligence, and what it does to motivation and personality when you see ability as fixed or as capable of growth.

The research on praise is by Mueller and Dweck, 1998. Published in the Journal of Personality and Social Psychology, on the effects of praising intelligence versus praising effort. That research is the direct basis for the process-praise variant in the skill, and the reason a compliment about talent gets labelled fixed in the scan.

The nuance also comes from Dweck. In articles in Education Week and Harvard Business Review, she described the false growth mindset: organisations that adopt the words without adjusting the systems. On organisation-wide questions, the skill refers to this and points you first to appraisal systems, promotion policy and the question of whether mistakes have consequences. That is an uncomfortable piece of advice, because it says a language tool will not help there.

Three related models the skill is allowed to bring in. Bernard Weiner's attribution theory, on whether you attribute an outcome to something stable and unchangeable or something temporary and influenceable. Albert Bandura's self-efficacy, described in Self-Efficacy: The Exercise of Control from 1997: the belief that you can handle a specific task, built through small successes and modelling. And Anders Ericsson's deliberate practice, from Peak, which he wrote with Robert Pool in 2016: growth comes from focused practice with feedback, not from clocking hours. That last model has its own skill, the Deliberate Practice Coach, for the moment mindset language is no longer the subject and the practice itself is.

Even without installing the skill, you can improve your own feedback with these principles: label the sentence and not the person, name the mechanism before you rewrite, and ask yourself whether the problem starts with language at all.

After that, using it is simple: paste the literal text and ask for a mindset scan. The rawer the text, the better, because the skill does not want a summary but the sentences as they stand. A more detailed explanation per environment is in the guide installing Claude skills NL, and the knowledge base has the wider explanation of working with AI.

Process praise, for when someone mainly praises rather than corrects, where compliments get rewritten from talent to approach. Attribution analysis, for people who blame everything on themselves after a setback, or entirely on bad luck, where each outcome is checked for whether the cause is stable or changeable.

The self-efficacy variant, for someone who believes the theory but does not act on it: then you work with small proof steps, a task small enough that success is almost certain. The deliberate practice variant, for people who put in many hours without progress, because then the conversation is about feedback on the practice rather than mindset language. And for organisation-wide questions, Dweck's own nuance about the false growth mindset comes onto the table.

There is also a review mode. Give it an existing document, such as an appraisal form, a job listing, a team handbook or a lesson brief, and the skill delivers five things: a count of how many fixed and how many growth phrasings appear and where they cluster, the pattern that dominates (usually there is one, not five), the three heaviest spots with a quote and the reason they have the most impact on the reader, rewritten versions of those spots you can paste straight in, and one structural recommendation that goes beyond language. That last point is the sharpest: if the appraisal form only scores end results, the rest is cosmetic.

And there are situations where you are better off leaving it alone. The frontmatter says so itself: do not use it as a substitute for psychological help with depression, anxiety disorders or burnout, and not to talk real problems away. A structurally excessive workload or an unsafe workplace is not a mindset issue. Not a small disclaimer, but the reason the fifth block is mandatory in every answer. Anyone who wants to help a team that is playing it safe first needs to know whether that is down to how people talk or to what happens when someone makes a mistake. Only in the first case is this the right tool. Otherwise you are putting a plaster on a system that is leaking.

One more honest limit: the skill improves your wording, not your judgement. If your conclusion about someone is factually correct and the collaboration is genuinely over, a rewritten sentence is not a solution but a delay. What it does do is make sure a conversation that is actually about something does not collapse on a sentence that makes the other person shut down.

If it works, you will quickly hit the limit of a standalone skill: your writing rules, your tone and the sentences you adjusted last time are gone the moment you close the window. That is what Jarvis is for, from €9 a month including VAT. And if you want your whole team writing this way instead of just you, that is the work of the AI training NL.