The AI Deliberate Practice Coach skill for Claude

The problem the skill solves is the problem almost everyone has who wants to get better at something. You practise faithfully, you do roughly the same thing every week, and after a while nothing changes any more. The file calls that the OK plateau: the point where a skill feels good enough, becomes automatic, and growth stops while the practising just carries on. What is missing is almost never time or motivation. What is missing is a specific goal just beyond your current ability, and feedback sharp enough to correct something with.

This coach is meant for individual use, and the frontmatter is clear that the domain is broad: presenting, selling, programming, writing, sport, music and any other skill with measurable feedback. It picks up the request on clear asks such as making a practice plan, a session plan for your training or breaking through a plateau, but just as well on sighs of frustration. I practise a lot but I am not getting better. I have been at the same level for years. I do not know exactly what to practise. My presentations stay mediocre. I do the same round every week. I lose focus after half an hour. No one ever gives me feedback. Say something like that to Claude while the skill is loaded, and it comes up by itself.

It belongs to the skill library NL of 100 free skills we make available from the AI and automation service, with no account and no sales email afterwards. If you are not yet sure exactly what a skill is, start with what Claude skills are NL. Want to learn to write a file like this yourself, then writing a SKILL.md NL explains the structure you see reflected in this file.

One thing up front, because the skill says so itself: this is not a replacement for a coach and not a promise. The coach itself is not allowed to promise mastery or guarantee a level within a set time, and it honestly states that the evidence for deliberate practice is stronger in domains with a clear yardstick than for open skills. What you get is a way of working with a known origin, not a path to the top with a date underneath it.

The four traits are stated literally in the file, and they decide everything the skill does after that. A specific goal just beyond your current ability. Full concentration. Immediate and accurate feedback. And direct correction, followed by repetition. Remove one and the whole thing collapses. The third point in particular is the fault line between what most people call practising and what this method means, because without feedback you reinforce your mistakes instead of fixing them. You then literally get better at doing it wrong.

The 10,000 hour rule is a misunderstanding with a long tail. Malcolm Gladwell made the research famous in Outliers in 2008, but Ericsson himself distanced his work from it: repeating something naively for 10,000 hours makes no one an expert. The file says the skill may not present that rule as settled science, and instructs it to explain the difference as soon as you refer to it. Rarely does an instruction file contradict the popular version of its own subject as explicitly as this file does here.

The practical difference sits in how you write down a practice goal. In diff form, with the wording the file itself uses:

Two findings put a boundary around every plan you get. The first: top performers rarely practise more than 4 to 4.5 hours a day with real purpose, because full concentration cannot be sustained for longer. The second: almost everyone ends up on the OK plateau. Both are in the file, and both have been turned into behaviour. It never plans sessions longer than 90 minutes, it puts exactly one sub-skill in a session, and it recognises a plateau by scores that stand still for three sessions.

If you want to widen this line of thinking beyond a single skill: the growth mindset coach sits on the belief side of the same question, and the Feynman technique coach tackles the same problem for knowledge instead of skill: only once you can explain something simply do you know where the gap is.

Next comes the theory, with its sources. Ericsson and the violin study from 1993, Peak from 2016, the four traits of deliberate practice, the 10,000 hour rule and why it does not hold up the way it gets retold, the limit of 4 to 4.5 hours a day and the OK plateau. At the end of that block are the three rules of thumb that keep coming back after that: four steps per cycle, a maximum of 90 minutes per session, and exactly one sub-skill per session.

The heart of the file is a list of eight things the skill always does. It asks for the skill, the current level, the practice time available per week and the goal. It splits the skill into at least four practisable sub-skills. It has you give a score from 1 to 10 per sub-skill. It designs an exercise per sub-skill with an exact duration and a concrete feedback mechanism: video recording, self-scoring card, measurement, coach or client reaction. It builds a week schedule that fits within your time. It makes a feedback moment compulsory after every session. It diagnoses a plateau when the scores stand still for three sessions. And it guards the limits, even when you yourself want to go past them.

After that come the four prompts, and that is the practical part that decides what you ask for. The Mastery Plan delivers the full map with sub-skills, exercises and a four week schedule.

The Session Design delivers a complete session of, say, 60 minutes: goal, a five minute warm-up, core exercise with exact instructions, feedback protocol and a correction note for the next session. The Feedback Analysis is for afterwards: you describe what you did and what the feedback was, and the skill names what is already going well, what the most critical point to improve is, which exercise addresses that and what the priority is for the next three sessions. And the OK Plateau Breaker is for when things stand still: a diagnosis of the likely cause, three concrete breakthrough exercises, a new feedback system and the ideal practice session for the next two weeks.

It contains a compulsory input checklist with six points, with the instruction to ask only about what is missing. The skill, and why that one specifically. Your current level, ideally with a concrete example of a recent performance. Your available practice time per week, realistic rather than aspirational. Your goal and the deadline, if there is one. Which feedback sources are available: camera, coach, colleague, measurement data, audience. And your earlier attempts at practising, with the point where they stalled.

The output format is fixed in five parts. The sub-skills map. The exercises, per sub-skill with duration, feedback mechanism and the criterion you read your progress from. The four week schedule with rest days built in. The feedback log, as a fill-in format per session with date, sub-skill, what measurably improved, the weakest point and the correction for next time. And the plateau signal: how you recognise that growth has stalled and when the breakthrough prompt gets used. Below that template is the shortened worked example you saw in FIG.01 above.

At the back of the file is exactly what the skill never does, together with five sources named explicitly. Those two blocks are below in section 04 and DOC.04.

Ericsson, Krampe and Tesch-Römer, 1993, is the main source. The paper The Role of Deliberate Practice in the Acquisition of Expert Performance appeared in Psychological Review and is based on research into violinists at the Berlin academy of music. This is where the four traits come from, and the finding that the way you practise weighs more heavily than the number of hours.

Ericsson and Pool, Peak, 2016, is the accessible version. In it, Ericsson brought his work to a wide audience and put his own correction of the 10,000 hour rule on paper. For the skill that is the source of the practical side: how you design an exercise, why feedback is the hinge, and why concentration sets the upper limit of a practice day.

Ericsson as editor of The Cambridge Handbook of Expertise and Expert Performance, 2006, is the broad reference work. It bundles research into expertise across many domains, and it is the reason the skill dares to limit its own scope: the strength of the evidence differs per domain, and you can only know that if you know the overview.

Gladwell, Outliers, 2008, is listed as the source of the misunderstanding. It is literally named there as the source of the popular but incomplete 10,000 hour rule. That is not a dig but a necessity: there is a good chance you picked that rule up somewhere, and the skill then needs to know where it comes from and what is wrong with it.

Macnamara, Hambrick and Oswald, 2014, is the critical note. Their paper Deliberate Practice and Performance in Psychological Science is in the file with the reason attached: practice does not explain all differences in performance. So it may not act as if enough good practice brings everyone to the same point. Combined with the honesty rule about open skills, that produces a coach that does not make its own method bigger than it is.

What is deliberately left out is at least as relevant. There are no success percentages in it, no number of weeks after which you reach a level and no figures about how much faster you grow with this method. So we do not repeat any of that either. For a second framework that makes learning measurable without making promises, you can look at the Kirkpatrick evaluation coach or the Bloom taxonomy coach.

After that, using it is simple: name the skill, your current level with a concrete example, your available time per week and your goal. The more honestly you state that time, the more usable the schedule, because the skill builds within what you say you have. If you do not know how to describe your level, just tell it what happened last time: the skill gets enough from that to suggest a score, which you then confirm or adjust yourself.

A practical tip that comes straight from the file itself: work out in advance which feedback sources you have. A camera, a coach, a colleague, measurement data or an audience. You get asked for this, and without a feedback source the whole plan falls back on your own feeling, which is exactly the mistake the method wants to prevent.

If you get stuck on the installation itself, the detailed explanation per environment is at installing Claude skills NL. And if you want to understand why files like this work the way they do, what Claude skills are NL is the best starting point. The wider articles about working with AI are in the knowledge base.

It is also strong in a situation that is less obvious: when you have been practising for a long time and it stops paying off. That is what the OK Plateau Breaker is for. Instead of suggesting you work harder, it looks for the cause of the standstill, gives three breakthrough exercises that push you back outside your comfort zone, suggests a new feedback system and designs the practice session for the next two weeks. That is a different answer to more hours, and usually the right one.

There are also situations where you are better off leaving it alone, and the file names them itself. The first is in the frontmatter: do not use it as a replacement for a qualified coach, teacher or physiotherapist for technically or medically risky skills. Strength training with heavy weights, singing technique, anything where practising incorrectly causes harm: there you need someone watching while you do it.

The second limit sits right behind it and is sharper than it looks: do not use it for motivation problems without a practice question. Someone who does not want to practise does not need a session plan but a different conversation. A practice schedule for someone who never starts on it is just another way to set yourself up to fail. If your issue is more about sticking with it than about technique, the Atomic Habits coach is the logical neighbour, because that one builds the system in which you show up.

The third limit is that of the evidence. Deliberate practice is best researched in domains with clear yardsticks. For open skills such as leadership or creativity the evidence is thinner, and feedback is harder to get objectively there. That does not mean you cannot use the method there, but it does mean you have to organise the feedback source yourself, and that you should not expect a score from 1 to 10 to mean the same thing there as with a measurement.

And the last honest limit: the skill does not practise for you. It designs the session, it guards the limits and it forces the correction, but you have to put in the hour yourself. What it does do is prevent that hour being spent repeating what you can already do.

If you want to talk further about what else AI can do for your business, from separate skills to an organisation brain your whole team works on, we will just do that in a conversation. No sales pitch, but an honest verdict on what makes sense in your situation.