The AI Maslow Pyramid Coach skill for Claude

The problem the skill solves is that motivation questions are almost always answered with the first measure that comes to mind. Somebody stops engaging, so a bonus follows. A customer does not buy, so a discount follows. A team goes quiet, so a team outing follows. In all three cases a remedy is chosen before the need has been established, and if the remedy does not land, the conclusion that sticks is that the people are the problem. The pyramid forces the reverse order: first work out which level is blocked, only then think of a measure that fits that level.

Six contexts are named where the skill belongs: marketing, HR and leadership, personal coaching, product development, sales and education. That list is broad because the question itself is broad. Why is somebody not doing what you would expect, and what need lies beneath that. Whether that is an employee who has gone quiet, a customer who still does not sign after three conversations, or you yourself putting off a plan for months.

What the skill is not: a measuring instrument. It delivers a structured hypothesis based on what you tell it, and that hypothesis is exactly as good as your description. If you only write down what somebody does and not what is going on around that person, you get an analysis resting on too little. Not a shortcoming of the model, then, but of the input, and it is the same reason the skill checks all five levels in step 1 instead of jumping straight to the most obvious one.

Aimed too high. The classic mistake is a measure at level four or five while level two is shaky. Offering a development budget and a fine growth path to someone who does not know whether their department will still exist next year is an offer that does not land. Not because it is a bad offer, but because it addresses a need that is not yet in view at that moment. This is where the skill catches it, because step 1 checks all the levels and step 2 pulls out the blockage that sits deepest.

Money used as a universal answer. Income sits at level 1 and security at level 2, which makes money a strong instrument for precisely those two levels. If the problem sits at connection or recognition, an amount of money changes nothing about the cause. It is also hard to reverse, so you have taken an expensive measure that does not touch the problem and that you can no longer swap for a measure that does.

Tackling everything at once. Anyone who makes an analysis and pulls eight actions out of it usually carries out none of them. That is why the file sets a hard ceiling of three recommendations, and step 5 explicitly asks which of them has the most impact and should go first. Not a limit on the analysis, but a choice about what actually happens on Monday morning.

The answer disappears into model language. An analysis that comes back in terms such as deficiency needs and self-actualisation potential is useless to anyone, and certainly in a conversation with the person it is actually about. Step 3 in the file is therefore literal: put the analysis into plain language, no jargon. That is the difference between an analysis you can read straight out loud and one you first have to translate.

In diff form, with the reflex taken out and the question put in:

Next come the five levels, each with a short description. Those descriptions are the skill’s real working material: they determine where it recognises a signal. Level 2, for example, lists contracts and predictability, so a reorganisation or an open vacancy lands there. Level 3 lists working together and a sense of team, so a colleague who left or a meeting that disappeared lands there. The levels appear in FIG.02 above with exactly the text that is in the file.

The heart of the file is the five-step method. Step 1: scan all five levels using the description you are given. Step 2: detect the primary blockage, singular. Step 3: put the analysis into plain language, no jargon. Step 4: give a concrete recommendation per relevant level, a maximum of three in total. Step 5: prioritise, and answer the question of which recommendation has the most impact and should go first. There is no more engine than that. There is no scoring model underneath it, no weighting and no point count, and that is deliberate: the model is a reading lens, not a ruler.

Then comes the list of six application contexts, and the output format with five fixed parts. That format appears in FIG.03 below. It closes with the attribution, a reference to us and the licence. It states MIT License, with the addition that attribution is appreciated but that you are free to use and adapt the skill. If you want to learn to write a file like this yourself, the structure is laid out step by step in writing a SKILL.md NL.

What is not in it is at least as important. There is no separate list of things the skill refuses, unlike some other skills in the library. The boundaries are built into the instructions themselves, as a clause attached to a step, and we have written those out in FIG.04 so you can check them. There is also no worked-out example conversation in it, and that is exactly why the example in FIG.01 on this page is labelled schematic.

Two things are worth knowing before you turn the model loose on your team. The first is that the 1943 article contains no pyramid. Maslow described a hierarchy of needs in text; the triangle by which almost everyone knows the model was drawn later by others for teaching and management books. More than a piece of trivia, because that drawing suggests a strict step-by-step order that the article itself states less absolutely. People can need something at several levels at once, and the skill accounts for that by showing the status of all five levels in the output format, not just the blockage.

The second is that the model is descriptive, not predictive. It gives you a way to order behaviour, not a formula that arrives at the correct answer. In psychology, the strict order of the levels has therefore remained a subject of debate. That does not make the model useless: as an ordering tool in a conversation it is useful precisely because it is simple and because everyone at the table understands it within a minute.

For using the skill, this mainly means: treat the outcome as a hypothesis you are going to test in a conversation, not as a diagnosis you are going to carry out. The fifth part of the output format helps with that, because it describes what should change if you carry out the first recommendation. If that does not happen, the hypothesis was wrong and you go back to the pyramid. That is not a failure but exactly how a thinking model is supposed to work.

If you want to improve the conversation itself rather than just the analysis, you can look further in the library. The GROW Coach skill runs a coaching conversation in four phases, and the Nonviolent Communication Coach skill is about how you keep an observation and a need separate in what you say. Both connect to the moment that comes after this analysis: the actual conversation.

If unzipping or recognition goes wrong, the full route with the common mistakes is in installing Claude skills NL. After that, using it is simple: describe the situation the way you would tell it to a colleague. The more context it contains about what has changed over the past few months, the sharper the scan in step 1 becomes. Maslow terminology is not required, and there is no fixed form.

It is also useful as a counter-argument to a decision that already seems made. If there is a plan for a team outing, a bonus or a new tool on the table, first hold the situation up against the five levels and check whether the measure lands at the same level as the signal. That costs ten minutes and it is the cheapest test there is.

There are limits, and they are not in the file because the file only describes the method. So we add them here ourselves. The first: this is not a diagnosis. The skill only reads what you tell it and has never spoken to the person involved. What comes out is a hypothesis about a situation, and therefore not a judgement on a human being, and it is wise to describe it that way when you share the outcome.

The second: if it is about somebody’s health, safety or wellbeing, a thinking model is not the right instrument. What belongs there is a conversation with the person themselves, and a professional where needed. An AI reading a piece of text cannot replace that and should not try to.

The third: the analysis is never better than your description. If you only write down the behaviour, you get an analysis about behaviour. What makes the difference is the context around it: what has changed, what has already been tried, what people themselves have said. And the fourth: this is a lens, not proof. If you bring the outcome to somebody’s table, the opening line is a question, not a conclusion.

If you are looking for something other than motivation, you are probably better off elsewhere in the library. For prioritising your own calendar there is the Eisenhower matrix NL, and for sharpening a cause rather than a need, the 5 Whys Analyst skill is the logical choice.

Do this more than once, and you notice the limit of a stand-alone skill: the previous analysis is gone the moment you close the window, while it is exactly the comparison between two moments that tells you whether anything is moving. That is what Jarvis is for, from €9 per month including VAT. And if you are looking for more tools of this kind, the rest is in the library of 100 free skills NL.