The AI GROW Coach skill for Claude
Describe a problem and explore Goal, Reality, Options and Will, one question per turn. This is the default route, with earlier phases revisited when new information warrants it. The closing block records your chosen step, date and review, or what remains unresolved. A 1-to-10 commitment self-rating supports discussion; it is not a probability, diagnosis or guarantee of action. Advice requires permission.
the lines above come from the zip on this page · SKILL.md is 20,050 bytes
Problem in, appointment out
reg. G.001This is a shortened illustration of the complete worked example in SKILL.md, not a recorded coaching result. The input supplies four people, sixty weekly working hours and difficulty finding time for new clients. The later answers, senior employee, dates and selected steps are fictional teaching assumptions. In a real conversation, the coach asks and waits for each answer.
What the AI GROW Coach is
The AI GROW Coach is a free skill from our own library, which by now counts 100 skills. A skill is an instruction file, SKILL.md, that gives an AI assistant a fixed way of working for one task. Not software you install, not a subscription, not a login: one text file of 3,223 words that tells Claude how a coaching conversation following the GROW model runs, which questions belong in which phase, what it needs to know before it starts, and what it may never do.
You download the zip at the top of this page, put it in your Claude environment, and from that moment Claude coaches according to this structure. What a skill actually is and how that file format works, read in what Claude skills are NL.
The skill addresses a familiar pattern: a problem is followed immediately by advice, before the person has clarified their goal or situation. That can miss what matters to them, but it does not prove that advice never helps. GROW offers a sequence for exploring the goal, current facts, possible options and a chosen next step; the file does not guarantee a result.
The standard exercise invites at least four of your own options; the explicitly selected ten-minute version invites three. If that is not feasible or wanted, record the limit and ask permission to explore further. The coach never fabricates your answers and labels any permitted suggestions as its own.
There are two kinds of use built in, and the file explicitly asks at the start which of the two it is. Self coaching, where Claude holds the conversation with you. Or preparation, where you are going to coach somebody else and the skill hands you the questions, the order and the closing block. That makes it useful for entrepreneurs who get stuck themselves, for managers with a progress conversation in the diary, for HR, and for people who do the occasional peer supervision session. It belongs to the skill library we make available for free from the AI and automation service, without an account and without a sales email afterwards.
One thing is stated right away in the file, and the page summarises it here: this is not therapy and the skill does not pretend otherwise. At signs of psychological distress, exhaustion that looks like burnout, or thoughts of self harm, the coaching stops and a referral to a doctor or occupational health service follows. That is not a small print line at the end, it is the first action on the list.
Why coaching conversations keep going round in a circle
The review function checks the conversation structure without assuming that either the coach or the other person caused the difficulty. It examines phase balance, question quality, advice moments and how concrete the recorded action agreement is.
The conversation starts at the problem. The default route clarifies Goal before Reality so the situation has a reference point. Starting with a problem does not inevitably produce complaints; return to earlier phases when new answers warrant it.
The session goal and the end goal run into each other. What you want in three years and what you want to have in hand at the end of this half hour are two different things. In the Goal phase they get pulled apart, and the skill asks about both. Without that separation, a twenty minute conversation ends with a life plan and no first step.
The coach fills in the blanks. As soon as somebody says "I have no time", the temptation is to hear "you need to delegate". That is filling in. Hence the summary after every phase, in the other person's own words, with a check question attached: is this right. That keeps the conversation belonging to the person being coached.
Advice gets disguised as a question. Have you ever considered getting a virtual assistant: that is not a question, that is a tip with a question mark. This is stated literally among the things the skill never does, and the assessment function marks every leading question as wrong, because a leading question is disguised advice.
No appointment comes out of it. Discuss a chosen action, date and review, including in the shorter version. If the person declines or details remain unknown, record that instead of forcing an agreement. A missing action is not proof that the conversation had no value.
The diff below illustrates separate question turns. Each replacement is a question to ask and wait on, not a completed answer or a claim about an actual transcript.
Four phases, five closing lines
reg. G.002The default route has four phases and a five-part closing block. Return to an earlier phase if an answer changes the goal or context. The closing block records the person’s own answers, including uncertainty or a decision not to commit; a neat structure alone does not establish success.
What is actually in the SKILL.md
The frontmatter gives a concise description for skill discovery. The body contains the longer trigger examples, including coach me, GROW model, preparing a conversation, feeling stuck, putting things off and having many ideas without a next step. These describe intended use; loading a file does not prove that every model will select or follow it correctly.
The description includes the main exclusions; the body explains them in full. Existing metadata remains version 1.0.0 and MIT. That field is preserved, not a new determination of third-party framework rights.
The seven instructions check suitability, ask one question per turn, use GROW as a revisitable default sequence, and summarise with an accuracy check in a separate turn. Options normally seeks four ideas, or three in the ten-minute version, without inventing answers. Will invites a chosen step, date and review. The final commitment self-rating is followed up separately; the person can revise or decline the step.
Before the conversation starts, the skill runs through a mandatory input checklist of six points: who is being coached, the topic in one sentence the way the other person would say it themselves, the time available, the context (work, career, team, business, personal or study), what has already been tried and why it did not work, and how much freedom the other person actually has. That last point is sharper than it sounds: can the other person decide for themselves, or is there a boss, client or family situation in the way. If something non essential is missing, the skill simply carries on and makes its assumption visible in the summary.
A separate chapter is about language. The you form, never a formal address. No dashes in running text. Short sentences alternate with long ones. Concrete over abstract, a number over an adjective. And explicitly: no coaching jargon without explanation, because words such as reframing, mirroring and congruent mean nothing to most entrepreneurs. That is exactly why a conversation with this skill does not sound like a coaching course.
A second mode is built in too. Give it a conversation record, a list of coaching questions or a transcript, and the skill assesses it on five points: the phase balance (how much of the conversation went to G, R, O and W, and which phase got flattened), the question quality (each question marked as open, closed or leading), the moments the coach filled in instead of asking, the concreteness of the appointment (action, date, owner), and finally a rewritten version with ten to fifteen replacement questions as a question bank in GROW order, for separate turns rather than one live message. Want to learn how to set up an instruction file like this yourself? The method is laid out step by step in writing SKILL.md NL.
Below an 8, the step goes down
reg. G.003The Will phase asks for a commitment self-rating from 1, not committed, to 10, fully committed. The exercise uses 8 as a discussion prompt, not a measured probability or diagnostic cutoff. After a lower answer, a separate turn explores a smaller or more feasible step. The fictional example starts at 6 and later reaches 8; neither number predicts success.
The theory the skill rests on
GROW emerged in British coaching practice in the 1980s around Graham Alexander, Alan Fine and Sir John Whitmore. Whitmore’s Coaching for Performance reached a broad audience in 1992. The download preserves the co-development attribution and links the originating organisations’ accounts. Those references do not establish one exclusive inventor or an undisputed first record; they do not prove that no earlier publication exists.
The underlying idea comes from Timothy Gallwey. The Inner Game of Tennis from 1974 set out the formula that carries Whitmore's view of coaching: performance is potential minus internal interference. This is a conceptual lens, not a measured subtraction or proof that knowledge and skills never need developing. The skill explores what the person knows before offering ideas with permission.
The scaling question comes from solution focused therapy. Steve de Shazer and Insoo Kim Berg are the founders of that approach, and the 1 to 10 scale is a fixed part of it. Here it sits within the Will phase, not as a loose trick.
The file names three alternatives alongside GROW. T GROW puts Topic at the front and is handy when you do not yet know what the conversation is about. OSKAR, developed by Mark McKergow and Paul Z. Jackson and described in The Solutions Focus from 2002, stands for Outcome, Scaling, Know how, Affirm and Action, Review, and may be considered when somebody already has momentum and wants to sustain or accelerate it. CLEAR by Peter Hawkins stands for Contract, Listen, Explore, Action, Review and may be considered in longer processes with several sessions, because there you need to contract explicitly about roles and confidentiality. So the file does not choose blindly for GROW: it names when another model fits better.
And then the most important line from the chapter of bans, because it works against the product itself: the skill may not claim that the GROW model has scientifically proven effect. It is a conversation structure, not a treatment. Anyone who makes more of it is selling thin air. That is also why this page is a good deal longer than a download button. Even without installing the skill, you can improve your own conversations with these principles: start at the goal, keep Reality factual, let the other person name their own options first, and discuss a chosen action, date and review without forcing or inventing commitment. Want to apply that way of thinking more broadly to asking questions instead of filling in answers, there is a separate Socratic Method Coach skill.
What the skill refuses
reg. G.004The SKILL.md contains a list of eight things the skill never does, and for a coaching skill that list matters at least as much as what it does do. A conversation about somebody's work or life can do damage if the person guiding it does not know their role. Those rules play out like this: every rule is a request you could make, with the response the skill gives according to its own instructions.
Installing in Claude Code, Claude.ai or Codex
The zip contains one folder, ai-grow-coach, with the SKILL.md inside. Installing is a matter of putting the file in the right place, and that place differs per environment. SKILL.md has been an open standard since December 2025, so the same skill also works in Codex, Cursor and Gemini CLI. So you are not downloading a Claude file but a working instruction that any modern AI assistant can read.
- Unzip it into
~/.claude/skills/(or.claude/skills/in your project). - Claude then picks the skill up by itself as soon as you start on coaching, goals or feeling stuck.
- Start Claude Code with claude, then enter
/ai-grow-coachinside the session.
- Go to Customize and then Skills.
- Upload the zip there as a skill.
- Or paste the contents of SKILL.md into the project instructions of a Project.
- Open
AGENTS.mdin your repo. - Paste the contents of SKILL.md into it, or put SKILL.md next to it as a separate file and point to it from
AGENTS.md. - Codex reads that along at every session.
Start with your topic and available time. The checklist asks for missing information, one thing at a time. The selected ten-minute version uses a three-option target; action and review still need discussion without inventing a commitment. If you get stuck anywhere while installing, the full step by step plan per environment is in installing Claude skills NL, and the wider explanation of working with AI is in the knowledge base.
When to use it and when not
This skill is at its strongest with a clear topic and somebody being coached who wants something themselves. That can be a choice between two directions, a goal that just will not come closer, or a progress conversation you want to prepare.
T GROW adds Topic when several subjects need sorting. OSKAR is a possible alternative for sustaining or accelerating progress, and CLEAR for a longer programme requiring explicit roles and confidentiality. These are selection suggestions, not proven rankings. An alternative framework is named explicitly rather than described as strict GROW.
In a ten-minute conversation, the exercise reduces the Options target to three of the person’s own ideas. Retain the action and review discussion, while recording missing information or a decision not to proceed. Completion of a structure is not a guarantee of change.
There are also situations where you are better off leaving it, and the file is short and clear about that. With psychological complaints, trauma, grief or addiction you need a therapist, doctor or occupational health service, not a coaching model. With a purely factual question, a technical instruction or a request for a ready made answer, coaching is simply slowing things down: then you want an answer, not four phases. And with somebody who only calls out solutions, the skill deliberately extends the Reality phase and consistently asks further for facts and figures, which is good for the outcome but rarely pleasant halfway through.
One more honest limit: the skill does not change your situation, it changes your conversation about it. If the diary is full because there is not enough revenue to hire somebody, no coaching question is going to fix that. What it does do is stop you running the same round for the fifth time with nothing recorded. If it is about the beliefs underneath, the Growth Mindset Coach fits better; if the goal itself is still too vague, the SMART Goals Coach sharpens that first.
Run it yourself, or have it run
reg. G.005This skill is the free do it yourself version of work we also deliver as a service. It stays complete and without any catches, but understand well what a skill is: it teaches your AI how to do something, while every new session starts empty. A skill is not the engine and not the memory. You prompt, you supply your situation and your previous appointments again every time, you check. Anyone who wants that differently has two next steps: hand over the engine, or sort out the memory.
where you are now The skill: you are the engine You run the AI GROW Coach yourself in Claude, Codex or Cursor. It costs nothing, it works today, and you keep it entirely in your own hands: no trial period, no locked off parts. The limit is your own time: the conversation only happens when you start it, and you have to bring last week's appointment back in yourself.
have it prepared The AI employee: it is ready without you prompting You do not outsource the coaching itself, because that conversation is yours to hold. The work around it, though, you can: preparing a progress conversation, the write up afterwards, the open appointments nobody keeps track of and the review moment that keeps sliding every month. That is what an AI employee sets up daily, with human final review as a fixed part: output stays a draft until a person approves it. We deliver this through Mansotti, the company TheSEO trades as, which today builds three employees: the Quote employee, the Sales employee and the Reporting employee. A coaching employee is not among them yet, so this becomes a role built to measure, and we only set that up once the process is repeatable and checkable.
everything from one source Jarvis: all your AIs work from the same company knowledge The skill teaches the AI, the brain is where the memory lives. If you want all your AIs to work from the same company knowledge, that is Jarvis, the organisation brain. It connects ChatGPT, Claude, Codex and your people to the same projects, core knowledge and decisions, so your next AI session does not start over. For a coaching skill that is the difference between four separate conversations and a process: last month's goal, Thursday's appointment and the number you gave then are all still there. What that delivers in practice, from the plans to your first week, is at Jarvis itself.
What Jarvis actually delivers
reg. G.006Step 3 deserves more than a paragraph, because this is the difference between a clever chat and a system you can build on. Jarvis is the organisation brain: it remembers what your AIs need to know, divides the work and keeps track of what happened. In coaching this stands out extra, because a conversation with no memory of the previous conversation is not a process. You notice it first at the start of a new session.
We have been running on this system ourselves for months. Every agent session, every task and every decision is logged in it and can be read back. A new session therefore does not start blank: it first fetches the recorded decisions, the running projects and the latest changes, and carries on where the previous one stopped. So we are not describing a promise but the way we work every day.
See all four plans at jarvis/prijzen NL. Through the waitlist NL you only pass on your preferred plan, without obligation. That does not create an account, an order or a duty to pay yet. We discuss business bespoke work first.
The skills around it
reg. G.007GROW structures the conversation, but not every problem starts with a conversation. These skills from the same library each take on a different piece: the direction underneath, the questioning technique inside it, and the self image around it.
For when the goal is sharp
the directionGROW starts at Goal. If that goal itself does not rest on anything yet, start here.
Ikigai CoachLooks for the direction underneath the goal: what you can do, what you want and what somebody pays for.SKILL SMART Goals CoachTurns a vague intention into a goal an outsider can check.SKILL Maslow Pyramid CoachLooks at which need sits underneath the problem before you start on goals.SKILLOther conversations, other models
the questioning techniqueCoaching is one conversation form. These skills do related work with a different mechanism.
Socratic Method CoachPresses on your assumptions rather than on your goal: sharper, also more uncomfortable.SKILL Johari WindowFor the blind spot: what others see and you do not, with feedback questions you can ask.SKILL STAR Method CoachStructures looking back at a concrete situation, handy at the review moment.SKILL NLWhen the brake sits elsewhere
the undercurrentSometimes the problem is not the structure, but what sits underneath it.
Growth Mindset CoachFor the belief that it is not going to work out anyway, which shoots down every option in advance.SKILL PERMA Model CoachLooks at wellbeing instead of performance, when that is actually what the conversation is about.SKILL Stoic CoachSeparates what you can influence from what you have to carry, when the situation itself is fixed.SKILL NLThe fundamentals
understand firstNew to skills? These three explain the file format, from understanding to writing your own.
What Claude skills areThe file format, the open standard and what a skill can and cannot do.DOC NL Installing Claude skillsThe full step by step plan per environment, from Claude Code to Gemini CLI.DOC NL Writing SKILL.mdBuilding a skill yourself, following the same method as this library.DOC NL The whole skill libraryAll 100 free skills in one list, sorted by subject.HUBFrequently asked questions
What does the AI GROW Coach skill cost?
Nothing. The skill is free, the file itself names the MIT licence and you do not have to create an account or leave an email address. You download a 7.9 KB zip containing one folder and a single file, SKILL.md, and that is the complete skill. There is no paid version and no sales email follows.
Does this skill work in Codex, Cursor or Gemini CLI?
Yes. SKILL.md has been an open standard since December 2025, so the same file works in Codex, Cursor, Gemini CLI and other tools that follow the standard. In Codex you unzip it into .agents/skills/ in your project, or into ~/.agents/skills/ for all your projects; Codex has supported SKILL.md directly since the open standard of December 2025. Pasting the contents of SKILL.md into your AGENTS.md still works too. The instructions are plain readable text, so any AI assistant that accepts instruction files can handle it.
Can I use this to coach myself, or does a second person need to be involved?
Either works. The first question of the skill is who is being coached: you yourself, or somebody you are going to coach. In the first case Claude holds the conversation with you. In the second case you get the questions, the order and the closing block to use in your own conversation. The content of the phases stays the same, only the role shifts.
Why does the skill ask only one question at a time?
One question per turn is an explicit instruction, including accuracy checks and follow-ups. It gives the person room to answer each point before the next question. It does not guarantee that every question receives an answer, and a prepared question bank is not sent as one live turn.
Does the skill also give advice, or only questions?
The coach asks questions and adds ideas only with permission, clearly labelled. The standard exercise invites four of your own options; the selected ten-minute version invites three. If that target is not feasible or wanted, the coach records the limit and asks permission to explore further. It must not invent your answers or disguise advice as a question.
What exactly is that scaling question from 1 to 10?
It is a self-rating of commitment: 1 means not committed and 10 fully committed. A score below 8 prompts a separate question about a smaller or more feasible step. The person may revise or decline it. The number is not a probability or diagnostic threshold; the skill uses the solution-focused scaling approach within Will.
When should I specifically not use this skill?
With psychological complaints, trauma, grief, addiction or situations that call for a therapist, doctor or occupational health service. That is stated at the top of the file and the skill then stops coaching. Also not for a purely factual question or a technical instruction: coaching there is simply slowing things down, and you are better off just asking for an answer.
First the conversation, then the diary
A structured conversation can help clarify a voluntary next step, but that appointment has to land somewhere. If yours is about too few enquiries, as in the example from the skill, the real work does not sit in the diary but in being found by the people looking for you. The free SEO scan shows in a few seconds where your site stands. Want your team to learn to work with this kind of skill, we do that in the AI training. And want to talk further about what AI can do for your business, from separate skills to full automation, we simply do that in a conversation.
