The Johari Window skill for Claude

No software to install, no subscription, no login: one text file of 2,446 words that tells Claude how to fill the four windows of the Johari Window, which source belongs with each line, what it needs to know before it starts, and what it must never do under any circumstances. You download the zip at the top of this page, drop it into your Claude environment, and from that moment Claude works according to this structure. What a skill actually is and how that file format works, you can read in what Claude skills are NL.

The problem the skill solves shows up in almost every filled in Johari Window you come across online. Somebody fills in four boxes with words like committed, perfectionist and sometimes too direct, and afterwards nobody can say where those words came from. Did a colleague say that, is it a self-image, or did the person filling it in make it up because the box would otherwise stay empty. This goes wrong especially with the Blind Spot, because that is by definition the window you cannot see yourself. Anyone who fills it in anyway is inventing how others see them. The skill inverts that: every line gets a marker, and without feedback from somebody else that window simply stays empty.

That also contains the second fixed choice in the file: one window per relationship. The model is not a photo of you as a person but of one specific relationship. That is why the skill starts by asking which relationship or group you are filling it in for, and the file gives a reason for that which needs no explanation: your manager sees something different than your brother in law. Fill it in without context, and according to the skill the result is a horoscope.

It is useful for anyone with something concrete ahead of them. A feedback conversation or performance review you want to prepare for. A new team where you still have to find out how you come across. A 360 degree round where the separate answers now sit in a document without anyone seeing a pattern in them. Or a collaboration that runs stiffly while nobody can name exactly why. It belongs to the skill library we make available for free through the service AI and automation, without an account and without a sales email afterwards.

One thing is stated immediately in the file, and we take that over here literally: this is not a psychological assessment, not a personality test and not a performance appraisal. In cases of mental health problems, trauma, bullying, harassment or a workplace conflict, the skill stops, names what it sees and refers you to a confidential adviser, mediator, GP or occupational health physician. That is not a small print line at the back, it stands at the front among the cases in which the skill must not start.

Lines appear with no source. According to the file, this is the biggest problem in almost every filled in window. A line with no origin cannot be checked, cannot be asked about and cannot be refuted. It stands there as fact while nobody knows who wrote it down. The skill solves that with three markers, further down in FIG.03, and with one hard rule: lines without a source do not exist.

Labels appear instead of behaviour. Dominant, chaotic, committed, driven: those are judgements about a person, not descriptions of something that happened. You cannot do anything with them, because you do not know which moment is meant. That is why it asks about your self-image in behaviour, and in the example converts analytical into something checkable: in meetings you ask for figures first before forming an opinion. If somebody did literally say such a label, it comes between quotation marks with the source next to it, exactly as it was said.

One remark becomes a pattern. Somebody hears once that they talk too fast and writes in the window that they talk too fast. The file counts differently: once is a signal, not a pattern. When ordering a feedback round the skill therefore notes, per point, how many people mentioned it, and sets points that only come up once to one side.

The window becomes a score. People want to know how big their Open Arena is, preferably in percentages. That is not possible, because the model has no unit of measurement. It therefore refuses percentages, just as it refuses window ratios in numbers and typings, and it is not allowed to claim that the Johari Window is a validated measuring instrument either. It is a conversation model from 1955.

No action comes out of it. Four filled boxes and then nothing. For each window it therefore gives one action that fits the movement of that window, it closes with three feedback questions you can ask tomorrow, and a next step comes with a date plus a moment to look back. Without that block, the report is not finished according to the skill.

In diff form, with the difference the assessment function flags in almost every window it is given: what goes out and what comes back in its place.

Not just on the obvious terms such as johari window, blind spots, self-insight or preparing a feedback conversation, but also on sentences people actually use: I never get honest feedback, I think I am being clear but nobody else thinks so, my manager says something different from my colleagues, new team how should I position myself, everyone thinks I am strict and I do not see it myself, I keep too much to myself, I do not dare ask what they think of me. Anyone who says something like that to Claude while the skill is loaded gets it automatically. The same frontmatter also lists the cases in which the skill should specifically not start, and below that is the licence: MIT, version 1.0.0.

After that comes the core: seven fixed actions. First it asks which relationship or group the window is being filled in for. It asks about the self-image in behaviour and not in traits. It asks for the feedback received as literally as possible, including who said it and when, and marks paraphrases as paraphrase. In every window it separates three kinds of lines and marks them. It leaves the Blind Spot empty if no feedback from others has been supplied, and delivers questions instead. For each window it gives one action that fits the movement of that window. And it closes with three feedback questions you can ask tomorrow, a next step with a date, and a moment to look back.

Before the windows are filled in, the skill runs through a mandatory input checklist of six points: which relationship or group the window applies to, the self-description in behaviour, the feedback that actually exists with its source, what you deliberately do not share and why not, the trigger, and how much room there is in this context to share. That last point is sharper than it looks, because the file adds a sentence rarely seen in this kind of model: not every workplace is a safe place. If something that is not crucial is missing, the skill carries on and makes its assumption visible in the report.

The output has a fixed structure of six blocks. First a frame of two or three lines: which relationship the window applies to, what the trigger is, which sources were used and how many people gave feedback, because one opinion is one opinion.

Then the four windows in fixed order, each with short lines and a source marker. Then two or three sentences about the relationship between the windows, in descriptive language, with no diagnosis or advice about who you should be. Then three feedback questions, concrete and answerable, aimed at behaviour in a situation: not how do I come across, but what did I do on Monday in the meeting that made you stop talking. Then the next step. And finally a block that never lapses: the limit of this report, which states what could not be filled in and why.

A separate chapter deals with language. Clear British English, with the reader addressed as you throughout. No dashes in running text. Short sentences alternate with long ones. Describe behaviour, not character: you interrupt is allowed, you are dominant is not, unless somebody literally said it that way. And no therapist language or jargon without explanation, because words such as mirroring, projection and congruent mean nothing to most people. That is exactly why the output does not sound like an assessment report.

There is also a second mode built in. Hand it a filled in Johari Window, a feedback round or a collection of loose remarks, and the skill assesses it on five points: the source separation per line, behaviour versus label with a rewrite attached, the source count per point from the Blind Spot, the action focus per window, and finally a rewritten version with eight to twelve replacement feedback questions. If you want to learn to set up this kind of instruction file yourself, the method is in writing a SKILL.md NL.

The ladder of inference explains why feedback so often lands as an attack. The thinking model by Chris Argyris, made more widely known through The Fifth Discipline by Peter Senge from 1990, describes how a judgement grows out of one observation in a few steps. Somebody sees you cut a conversation short, infers from that that you have no interest, and ends up saying you are dominant. The observation and the judgement are then already four rungs apart. Anyone who knows that asks different feedback questions: not how do I come across, but what exactly did I do at that moment. The same way of thinking is in the Ladder of Inference Coach skill.

Radical Candor is about the quality of the feedback you gather. Kim Scott's book from 2017 is in the source list because the Johari Window only helps you further if the feedback you get is honest and specific. If you only get pleasantries back, your Blind Spot stays exactly as big as before. The Radical Candor Coach skill NL takes on that side of it.

Nonviolent Communication supplies the phrasing. Marshall Rosenberg described in 1999 how you name behaviour without sticking a label on it. That exact writing rule is what the skill imposes on itself: you interrupt is allowed, you are dominant is not. For conversations where that distinction makes the difference, there is a separate Nonviolent Communication Coach skill.

A 360 degree round is the organisational form of the same idea. Multiple sources that together make the Blind Spot visible. The skill can order that kind of round, but it does not replace it: it sorts the answers by behaviour, counts the mentions and sets the one off points aside.

And then the three warnings the file itself gives about using the model, because they go against its own product. The window is not a personality test and delivers no score. The window cannot be established once and for all, because it looks different per relationship and per context. And a large Hidden Area is not automatically wrong, because not everything belongs at work. Even without installing the skill, those three rules alone let you set up your own feedback round better: ask about behaviour in a situation, note who said what, and count before you draw conclusions.

After that, using it is simple: say which relationship or group the window applies to and what the trigger is, paste in your self-description and the feedback you have, as literally as possible and with the source attached. Whatever is missing, it asks about itself, one thing at a time. If you only have a stack of loose answers from a feedback round, just paste them in as they are: the second mode sorts them for you. If you get stuck anywhere installing it, the full step by step plan per environment is in installing Claude skills NL, and the broader explanation of working with AI is in the knowledge base.

Always fill in one window per relationship, so separately for your team, your manager and your clients, because those windows really differ. For a team as a whole there is a team version: what does the team know about itself and what do outsiders see that the team does not, where you explicitly ask for sources outside the team. For a 360 degree round you use it to order loose answers, sort them by behaviour and count them. And without anyone giving feedback you only get the questions, which according to the skill is not half a result but the real work.

For a performance review or appraisal a separate rule applies: use the window as preparation, not as a report. A judgement from your employer does not belong in a self-insight model. That distinction is easy to forget if both conversations fall in the same week, and it is exactly the point where this model goes wrong in organisations.

There are also situations where you are better off leaving it alone, and the file is brief about that. Not as a substitute for a psychological assessment, a personality test or a performance appraisal. Not for mental health problems, trauma or conflicts that call for a confidential adviser, mediator or occupational health physician. And not to map out somebody else who is not taking part in the conversation.

One more honest limit: the skill does not gather feedback. It orders what exists and formulates the questions with which you can gather the rest, but asking those questions remains human work, and the answer only comes once the other person feels safe enough to give it. If you then want to have a conversation about what you do with that feedback, the GROW Coach connects to that with four phases and an appointment with a date. If the underlying question is more about direction than self-image, the Ikigai Coach is more logical, and if it is about how a team develops, the Tuckman Team Stages Coach fits better.

What is possible is ordering the round around it: setting the question up, bundling the answers and not letting the look back moment slip. Do that for a whole team, and that is work an AI employee is for. If you want your people to have this kind of conversation better themselves, with or without AI alongside them, that is the AI training.