rapid_skill_acquisition v1.0.0

name: "rapid_skill_acquisition" description: "Builds a twenty hour plan to learn a new practical skill to a defined target performance: deconstructing the skill into subskills, scoring which subskills carry the target, removing practice barriers before starting, structuring sessions for fast feedback, and testing on unseen material at hour twenty. Triggers when someone wants to learn a skill fast, get good enough at something in weeks, plan the first twenty hours of practice, or break a skill into what to practise first." metadata: { "includeInPrompt": true }

Rapid Skill Acquisition

Purpose

Turn a wish to learn a skill into a twenty hour practice plan with a test at the end. The plan defines a target performance, breaks the skill into subskills, scores which ones carry the target performance, removes the barriers that stop practice from happening, and lays out sessions that each carry a feedback loop and an exit check. One complete worked plan ships with the skill, for basic Spanish for a restaurant meal, in references/worked-plan.md with a machine readable copy in examples/.

Treat twenty hours as a planning budget. The figure carries no guarantee. It comes from Josh Kaufman's The First 20 Hours (Portfolio and Penguin, 2013), where it is a practical commitment for getting past early frustration to noticeable competence for personal use. It is not a peer reviewed threshold and it does not produce expert performance. Ericsson and colleagues studied averages of over 10,000 hours of deliberate practice by age 20 in elite track violinists. That is a different goal at a different scale. This skill aims at good enough for a named use, tested honestly, with a repair rule for learners who miss the target at hour twenty. Full sources and limits are in references/research-brief.md.

Workflow

1. Define the target performance. Write what the learner will do, on what material, under what conditions, with no help, in one or two observable sentences. Write the scope boundary beside it, naming what the plan leaves out. Write the safety boundary too: any part where a beginner error can cause harm, damage, loss, or legal trouble leaves the self-taught plan and goes to a qualified teacher. Use the worksheet in references/subskill-worksheet.md.
2. Check three to five resources, then stop researching. Use books, courses, and working practitioners to find the subskills. Cap this step at the first one or two sessions of the budget. Research past that point is a barrier wearing a productive disguise.
3. Deconstruct into five to nine subskills. Walk the target performance in order and list each distinct thing the performer does. Add the failure view: picture the worst realistic failure inside the performance and list what would have prevented it. Each subskill gets a beginner error, a beginner level correct version, a practice task, a feedback source, and a self-correction cue.
4. Score priority, then argue with the score. Score each subskill on frequency inside the target performance, blocking power, and early learnability, each from 0 to 2, following references/subskill-worksheet.md. Handle safety separately: a safety score of 2 removes the subskill from self-teaching at any total. Totals of 5 to 6 are critical, 3 to 4 are useful, 0 to 2 are deferred with a written reason. Then run the removal challenge on every critical call: if the final test would still pass without this subskill, it is not critical.
5. Remove barriers before session 1. List at least three barriers across physical setup, time, mental friction, and emotional avoidance, with a concrete removal for each. Materials ready, a fixed practice slot, and a format the learner will actually do. A removal still listed as an intention at session 3 is the likely cause of a stall.
6. Learn enough to self-correct, per subskill. For each critical and useful subskill, name the model, checklist, measurement, recording, or person the learner will compare attempts against. A beginner starts in the cognitive stage described by Fitts and Posner, with frequent errors and heavy attention load, so early sessions need models and outside reference points. Self-correction is built by the plan, not assumed at the start.
7. Lay out the sessions to exactly 1200 minutes. Use the templates in references/session-templates.md. The default is twenty sessions of 60 minutes: cold recall first, then a model and a single target, short practice rounds with one change per round, integration back into the whole performance, and a three line log. Every session names one primary subskill, a practice task, a feedback loop, and an exit check. Place at least two repair and rehearsal sessions in the second half, driven by the error log.
8. Spread the hours. Put the sessions across days and weeks when the goal is to keep the skill, in line with the distributed practice finding in the research brief. A crammed plan for an immediate deadline must say plainly that retention will be weaker. Count only time spent performing with feedback toward the twenty hours. Reading and watching about the skill do not count unless the learner is performing inside that time.
9. Keep one error log for the whole plan. One line per notable error: subskill, what happened, what the feedback source showed, what changes next time. The log picks the repair session topics, the rehearsal weak points, and the diagnosis when the final test fails.
10. Validate, test, and apply the retest rule. Save the plan as JSON matching the examples and run bin/plan_check.py on it. Fix every reported problem. At hour twenty, run the final test on material the learner has not practised on, scored item by item against the target performance. One failed subskill gets four extra focused hours and a full retest. Three or more failed subskills mean the deconstruction was wrong, so redo the worksheet before spending more hours.

Output Contract

A finished plan always contains:

• A target performance, a scope boundary, and a safety boundary
• Three to five resources checked, with a stated point where research stopped
• Five to nine subskills, each with priority, why it matters, a practice task, a feedback source, and a self-correction cue, plus deferred subskills with written reasons
• At least three practice barriers, each with a removal that is done or dated before session 1
• A session list totalling exactly 1200 minutes, each session with a primary subskill, focus, practice task, feedback loop, and exit check, including at least two repair or rehearsal sessions in the second half
• An error log format and a final test on unseen material, with the retest rule stated
• A validation line showing bin/plan_check.py passed on the JSON version

Operating Rules

1. Accuracy rule: every factual claim about learning research must trace to references/research-brief.md. Never invent hours, percentages, stages, or study results. Unverifiable means cut it or label it as a design choice, as this skill does with its priority scoring and session templates.
2. Never present twenty hours as a guarantee for any skill or any learner. Present it as a budget, a target performance sized to fit early gains, and a test with a repair rule.
3. Never present the 80 and 20 figures as a law or as a method for finding the important subskills. Priority in this skill comes from analysing the target performance itself: frequency, blocking power, safety, and early learnability.
4. Never let a safety critical subskill be self-taught from a worksheet. Driving, swimming in open water, electrical work, medical procedures, work at height, and similar skills need qualified instruction in person. The plan can cover the classroom parts and the practice structure around lessons, and must say that is what it is doing.
5. Never count passive time toward the twenty hours. If a session has no performance and no feedback, it is preparation, and the plan logs it as preparation.
6. When a learner stalls, diagnose before adding hours: wrong subskill split, a feedback loop that cannot show the fault, a barrier that was never actually removed, or a target performance sized too large for the budget. Each cause has a different fix, and the plan names which one applies.