The AI Flow State Coach skill for Claude

The problem the skill solves is the gap between knowing and doing. Everyone knows by now that distraction is bad for concentration. That does not help, because it is a generality and you have a specific task, on a specific afternoon, with specific people around you. Here it does get specific. It asks about the task, your experience with it, the time you have, who can interrupt you and the last moment it did work. That last answer usually holds the clue. What follows is not a list of tips but a schedule with actual times, and a first action small enough that it cannot fail.

It is written for people who earn their money with their attention: the self-employed, developers, writers, designers, consultants, students, and team leads who want to arrange the same thing for a group. It is part of the skill library NL that we make available for free from the AI and automation service, with no account and no sales email afterwards.

There is one thing you should know in advance, and the skill says so itself: flow is not a switch. The circumstances can be arranged, the state cannot. What this skill does is increase the odds by arranging the things you can arrange. It does not promise you flow, it does not promise you a number of focused hours per week, and it is not a tool for persistent concentration problems. If the conversation heads that way, it does not finish the protocol and refers you to a GP, psychologist or occupational health doctor.

Nobody starts with a diagnosis. "I cannot concentrate" is a symptom, not a cause. A task you are avoiding because it feels too big calls for something completely different from a task you are avoiding because it is boring, and the usual advice only fits one of the two. For a task that is too hard, breaking it up helps; for a task that is too easy, breaking it up makes things worse, because it just chops the boredom into finer pieces. That is why the quadrant you are in gets established first, and only then does a proposal follow.

The advice is about you and not about the situation. More discipline, less scrolling, just start. That is not advice but a judgement, and it produces guilt instead of a session. It looks at the task and at the circumstances. In the example above, the conclusion is not that the user lacks discipline, but that two sources of interruption are sitting right next to him and that his one successful session was one without them. That is something you can actually do something about, this afternoon even.

The conditions stay abstract. "Make sure you have a clear goal" is true and useless. A clear goal is: by the end the opening and the phases are down, rough, in your own words. You can check that afterwards. The same goes for feedback: not "get some feedback" but "after each block you check whether the client's own words appear in it". All four conditions are made concrete enough that you can tick them off.

The first step is too big. This is the point where most protocols fall apart. "Start the chapter" is not a first step, that is the whole task with a friendlier word in front of it. Only one rule applies here: the first action cannot fail. Opening the file and typing the heading, that is a first action. And every session gets a break, because working through until it is finished is never suggested.

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

Not only for the obvious questions like getting into flow, deep work or planning a focus block, but also for complaints such as: I have been switching between things all day and nothing gets finished, I open the file and then do something else, I do not dare start because it is too big, I work for hours and have no idea whether I am making progress. Anyone who says something like that to Claude while the skill is loaded gets it automatically. The same frontmatter also sets the limit: the skill is not there to make or confirm a diagnosis, and not to enforce performance targets on other people.

The core is a five step method, and none of them get skipped. Step 1 is asking: which task, how experienced you are with it, how much time you have, where you work and what went wrong last time, and it does that in no more than three questions at once. Step 2 is the diagnosis: the task gets placed in a quadrant, in no more than five sentences, stating what the diagnosis is based on, namely your own words. If the complaint clearly falls outside what this model covers, the skill stops there and refers you on.

Step 3 is adjusting, and it works differently per quadrant. For a task that is too hard, the skill breaks it up until the first step can be done within ten minutes, and it proposes a rough first version instead of the final result. For a task that is too easy, the bar goes up with a time limit, a higher standard or a different method. If challenge and skill are both low, the question of why this task exists at all comes first, because breaking it up would be pointless. Step 4 fills in the four conditions and step 5 turns the whole thing into a schedule with actual times.

Four fixed rules run through every step. The block length comes from you, not from the skill: ask for advice and you get a range with the reason behind it and a note that this varies by person and by task. Every session gets a break. The first action is always small enough that it cannot fail. And whatever you have not told it does not get invented, it gets marked as [FILL IN].

The output format is fixed in seven parts: diagnosis, adjustment, the four conditions filled in, the session schedule with actual times, the first action, the close with three evaluation questions, and finally notes for the user naming the assumptions built into the protocol and the point where the schedule is likely to break. That last block is never left out, and that is what sets the skill apart from a list of tips: it names its own weak spot before the session does. The file closes with a list of seven sources, from Csikszentmihalyi to Zeigarnik.

Flow comes from Mihaly Csikszentmihalyi. From the 1960s onward, the Hungarian-American psychologist studied what people were doing and feeling at random moments of the day, using a method where participants noted down what they were doing whenever an unannounced beeper went off. Beyond Boredom and Anxiety appeared in 1975, the book Flow: The Psychology of Optimal Experience in 1990, and Finding Flow in 1997, which translates the theory to everyday work. The description the skill holds to comes from 1990: the state in which someone is so absorbed in an activity that nothing else seems to matter any more.

It deliberately uses only part of that theory. Csikszentmihalyi names a range of characteristics of flow experiences, and some of them are a consequence you cannot arrange: the distorted sense of time, the disappearance of self-consciousness, the experience that the activity is its own reward. The skill leaves those characteristics alone, because there is nothing to steer there. What is left are the four conditions from FIG.03, which you can set up in advance. Jeanne Nakamura and Csikszentmihalyi later reordered those characteristics in a 2002 chapter, and that ordering is the reason the skill holds to exactly these four.

Raising the bar comes from the work of Anders Ericsson. His 1993 research with Ralf Krampe and Clemens Tesch-Romer describes deliberate practice: progress comes from working at the edge of your ability, with feedback attached. The same move, in other words, as the adjustment in the boredom quadrant. Anyone who wants to go that way, with a practice programme rather than a single session, can continue with the Feynman Technique Coach skill and the Growth Mindset Coach skill.

Planning the block comes from Cal Newport, closing it from Bluma Zeigarnik. In Deep Work from 2016, Newport describes why concentrated work needs a block you schedule in advance and not a gap that happens to be left over. And in 1927 Zeigarnik studied why unfinished tasks stick in memory better than finished ones, which explains why it helps to close a session at a point where the next action is already fixed. Both are built into step 5 of the skill.

Even without installing the skill, you can use these principles to set your afternoon up differently: first establish whether the task is too hard or too easy, make the four conditions concrete instead of well-meant, and choose a first action that cannot fail.

After that, using it is simple: describe the task you are putting off, say how experienced you are with it and how much time you have. Name the last moment it did work too, because that is usually where the clue sits. If you want to learn to write this kind of instruction file yourself, for a way of working that fits only you, for instance, writing a SKILL.md NL explains how the frontmatter and the structure work. The broader explanation of working with AI is in the knowledge base.

For recurring work you build the protocol once and reuse it, with only the goal adjusted each time. For a team it fills in the four conditions at group level: a shared goal, a visible sign of progress, an agreed meeting-free block and a clear first step per person. The diagnosis does stay individual, because challenge and skill differ from person to person.

There are also situations where you are better off leaving it be. The most important one is the most honest: if the conversation is actually about exhaustion, low mood or persistent complaints, the skill does not finish the protocol. So not a disclaimer in the small print, but an instruction in the file itself. A second limit: if you have too many tasks and do not know which one comes first, this is not the skill you need. A timer, a blocking app or an even tighter schedule will not solve that. That is about ordering, not concentrating, and the Eisenhower Matrix Coach skill or the Getting Things Done Coach skill is the logical first step.

One more honest limit: the skill arranges your circumstances, not your calendar. If your best hours are structurally eaten up by meetings, nothing useful comes out of this protocol, because there is nothing left to schedule. That is then not a focus problem but a planning problem, and you solve that with the people who plan those meetings. And if the task itself is not sound, for instance because nobody knows why it exists, the skill's answer is not a schedule but a question back.

The second block only shrinks once work actually leaves your hands. That is what an AI employee does: the preparatory work is ready before you even sit down at your screen, and a person approves it before anything goes out. If you mostly get stuck on supplying context over and over again, Jarvis is the far cheaper route, from €9 per month including VAT.