agentleFS
thewyattbrocatothewyattbrocato/design-and-copy-skills/skills/practice-and-assessment/SKILL.md

practice-and-assessment

Skill2 starsChanged 12 days agoMITView on GitHub

Use whenever people must practice, be checked on or retain a skill, in any language and at any length - "write a quiz for new hires", "training that sticks", "people keep making the same mistake", "better feedback for wrong answers", "a practice plan for the first month", "did the training work", "make this course less boring". Covers whether training is the fix, objectives, scenario questions, feedback messages, review schedules, mastery, assessment and measuring results. Not for onboarding screens (teaching-interfaces), quiz software or grading code, form fields (form-design) or reviewing a learning product's interface (design-critique).

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 11 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈11

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write the objective before any activity: who does what, in what situation, to what standard, in terms a bystander could score ("flags a …

    line 37
  2. Distinctive:

    Put a named person in a specific situation and ask what they would do next.

    line 41
  3. Distinctive:

    Write the options from mistakes experienced people actually make.

    line 42
  4. Distinctive:

    Default to a single-decision item: one situation, one choice, one visible result.

    line 45
  5. Distinctive:

    Use only the facts, numbers and policies the user gave.

    line 46
  6. Distinctive:

    No time pressure on first attempts unless speed is the job.

    line 60
  7. Distinctive:

    Start from the claim ("can tell which readings need a recheck a week later, among mixed cases"), then build items that would produce that …

    line 65
  8. Distinctive:

    Do not stall. Ask once only if interactive and cheap (what should people do differently?

    line 75
  9. Distinctive:

    When a count or set is asked for (three items, four weeks, two options), deliver exactly that.

    line 80
  10. Distinctive:

    After the work, at most two one-line notes unless asked for more.

    line 82
  11. Distinctive:

    Every item is a decision in a situation, not a definition.

    line 86

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Practice and assessment

Design what learners do, what they hear back, and when they return. Unguided, the result is definition recall ("what does the acronym stand for"), "which is NOT" items with one joke option, a bare "Correct!" or "Incorrect" with the answer letter, a course for a problem caused by a broken tool, mastery declared after one same-day pass, and a satisfaction survey as proof.

When to use

  • Writing or fixing questions, scenarios, drills, feedback, practice plans, review schedules, pass criteria or a training evaluation.
  • A performance problem where someone has proposed training: check the cause first.

When not to use

  • Onboarding, tutorials and in-product help: teaching-interfaces if installed. Input controls inside a quiz: form-design if installed. Judging a learning product's screens: design-critique if installed.
  • Building quiz software or grading logic, course marketing, exam security policy beyond item design.
  • Text that already works (a good scenario item, sound feedback): say so in a sentence and change at most a word.
  • A first-run screen is not a test; it should lead to a first real success.

Do not produce

  • Items that test recall of definitions, acronyms or slogans; "which is NOT", true/false on a rule, options that differ in length or include a joke.
  • Bare "Correct!" or "Incorrect", the answer letter as feedback, praise of talent ("you're a natural").
  • A course as the answer when the cause is a tool, incentive, time pressure or unclear process.
  • Objectives such as "understand" or "be aware of".
  • Mastery from one same-day score, or a single end quiz with a pass mark as the only check.
  • Satisfaction scores or completion rates offered as proof that training worked.
  • Streaks, public rankings, countdown timers, points for logging in.
  • An invented policy, number, study, pass rate or baseline. Use a visible placeholder in brackets, such as [current error rate], and still deliver the work.

Is training the remedy

Ask what people must do differently, in which situation, and why they are not doing it now: missing knowledge, an unpracticed skill, "I know, but...", habit, or environment (tools, incentives, time, a vague process). Training cannot repair a failing system, so say plainly what to fix first and who owns it. After that, train only the part that is a skill or a hard-to-apply fact, usually a small job aid at the point of use plus practice on the one decision it cannot settle. Do not refuse training when the cause is a skill. For "I know, but..." gaps, pair a small first step with an if-then plan and make the right move easier in the tool than the wrong one; inspiration does not change habits.

Write the objective before any activity: who does what, in what situation, to what standard, in terms a bystander could score ("flags a loan application with mismatched income documents and attaches the two documents, in 9 of 10 practice cases").

Writing items

  • Put a named person in a specific situation and ask what they would do next. Test the judgment, not the policy wording.
  • Write the options from mistakes experienced people actually make. Make the most common one tempting with a realistic detail; one option is best and the rest are worse in different ways. Positive stem, options of similar length and form.
  • Each wrong option maps to one misunderstanding, so feedback can name it.
  • Let people look things up when the job allows it. Put the policy or checklist beside the question, and require memory only for what must be produced with nothing to consult.
  • Default to a single-decision item: one situation, one choice, one visible result. Branch only when an early choice changes what happens or what is known later. Never loop people back until they guess the author's path.
  • Use only the facts, numbers and policies the user gave. If the situation needs one they did not, bracket it.

Writing feedback

  • Open with what happened as a result (the customer's next line, the result on screen), then the principle in one sentence, then the better move. Write it for every option, since people also pick the right answer for the wrong reason. Say what was right in a partly right answer, keep the tone calm, and after a miss give one sentence on what to try next instead of a pep talk.
  • During practice, give feedback right away. For a later retrieval attempt, withhold the answer until the person has tried, then show it.
  • A secure certifying exam may hold back per-item feedback. Give a score report by area, and put full feedback in separate practice tests.

Sequence and schedule

  • Newcomers get two or more worked examples in different contexts, then partly completed problems, then full ones, with support fading as success rises. People who already do the job can start with decisions. Do not open with an empty task and call it discovery.
  • After first exposure, ask people to produce the answer instead of re-reading. Bring items back the next day, with a night of sleep in the gap, then at widening gaps (days, then weeks) matched to how long the skill must last. Do not stretch gaps until people relearn from zero, and do not schedule one-time tasks at all: put the reference in the tool.
  • Mix only categories people confuse, after each has been met alone. Block the first encounter. Mixing unrelated topics is not interleaving.
  • For a move that must become automatic, drill one small piece with success high enough that the right move is what gets stored, redo a miss at once in a different way, then rehearse in a realistic, mildly pressured case. Keep difficulty inside what learners can solve with effort; when most fail, go back to an example.
  • No time pressure on first attempts unless speed is the job.
  • Mastery means success at a stated standard on a later day, in a mixed set, with changed surface details. Choose and state the standard; do not borrow one.

Assessment and results

  • Start from the claim ("can tell which readings need a recheck a week later, among mixed cases"), then build items that would produce that evidence. One same-day score cannot certify it. Complex work needs observed performance scored against a short rubric.
  • Name the metric the organization already tracks and the observable behavior behind it, and write the check before the training. Add closer signals (choices in later practice, error rates, tickets, which aids get used) and compare people who applied the skill with those who did not; ask both why.
  • A satisfaction survey flags unhappiness only. Show ratings beside performed results, and report small groups as "8 of 10", not percentages. Pilot with three to five people thinking aloud, and let them struggle instead of stepping in.

Motivation and access

Keep early stakes low, show progress on the real skill privately, and make the challenge solvable. Send review prompts inside a session the person opened, not as late-night pushes. Give media a text alternative, show consequences in words as well as color, support keyboard use, and use timers only when timing is the skill.

Missing information

Do not stall. Ask once only if interactive and cheap (what should people do differently? who are they?). Otherwise assume a learner, state the assumption in one line, bracket every missing fact, and deliver.

Output discipline

  • Deliver the requested thing first: the items, the feedback, the plan, the verdict. Reasons are a line each.
  • When a count or set is asked for (three items, four weeks, two options), deliver exactly that.
  • Honor any word cap. Count the words and cut the weakest point instead of asserting that it fits.
  • After the work, at most two one-line notes unless asked for more.

Quick checks

  • Every item is a decision in a situation, not a definition.
  • Each wrong option is a real mistake and feedback names its consequence.
  • The cause was checked before a course was proposed.
  • Mastery needs a later day and a mixed set.
  • Success is measured by behavior or an existing metric, with no invented numbers; survey items ask about one recent event each, scores state their base, and scoring was fixed beforehand.

References

  • references/measuring-and-wording.md: load to write a survey, rating scale or self-rating item, set pass criteria for observed practice, design a pre/post check, or report what a score can and cannot show.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.