Design or review educational applications by connecting learning goals, learner needs, activities, feedback and evidence of learning. Use for new learning tools or substantive changes to their teaching design. Do not apply to cosmetic changes or ordinary bug fixes without a learning-design issue.
Installs into .claude/skills of the current project.
Sponsors
SSkills DirectorySkills Directory
Know which skills are safe — weekly.
SSkills DirectorySkills Directory
Is that skill safe to install?
Are you the author of learning-methodologies?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/siberius-learning-methodologies)
---
name: learning-methodologies
description: Design or review educational applications by connecting learning goals, learner needs, activities, feedback and evidence of learning. Use for new learning tools or substantive changes to their teaching design. Do not apply to cosmetic changes or ordinary bug fixes without a learning-design issue.
metadata:
version: "1.4.0"
license: "MIT"
repo: "https://github.com/siberius/learning-methodologies"
author: "Luděk Sušický"
format: "native bundle with generated standalone Markdown"
---
# Learning methodologies
Help design, implement or review educational applications whose learning activities, support and assessment serve the intended learners. Adapt the depth of work to the request, consequences and uncertainty. A small classroom activity does not need a full evaluation study or a new planning document.
## Didactic methods for applications built with AI
Use the methods to turn educational goals into learner actions, feedback rules and evidence in the generated application. For example, retrieval requires an attempt before revealing the answer, while scaffolding requires controllable help and a distinction between assisted and independent performance. AI can help implement these decisions; its output still needs subject and teaching review. This applies to offline applications and to runtime AI when the learning goal needs it.
Read [didactic rationale](references/didactic-rationale.md) when explaining why a method fits or tracing it into implementation. It connects methods to product decisions and the evidence needed to assess them.
## Guide the creator through decisions
At each substantive learning-design decision, revisit the goal, learner action, evidence, support and constraints. Use existing choices. For an unresolved consequential choice, offer a small relevant selection, usually two or three options, each with a plain-language description and a concrete application here. Recommend one and explain the main tradeoff; ask which approach the creator wants. Read [design dialogue](references/design-dialogue.md) when starting or revisiting such a choice. This governs the creator conversation, separately from the learner's tutor interaction.
Do not reopen settled or declined choices without a relevant change. Honor explicit delegation of routine decisions and continue with a short rationale. Silence does not approve a scope expansion. While waiting, work on independent parts. A cosmetic change does not trigger this process.
Load details by need: [methods](references/methods.md) for learning activities and support; [games](references/learning-games.md) for motivation and mechanics; [assessment](references/assessment.md) for evidence and scoring; [cognitive interface](references/cognitive-ux.md) for representations, multimedia and attention; [adaptive practice](references/adaptive-practice.md) for repeated sessions or difficulty selection. Use the existing conditional routes below for simulation, tutor state and runtime AI. Read the relevant method before offering it; do not load every reference.
Translate the chosen method into observable product behavior and a meaningful check. Distinguish offered, selected, implemented and verified decisions. Technical verification does not establish learning effectiveness.
## Working route
Use this route at the depth the task needs: establish the goal and constraints, choose a method for the learning obstacle, connect the learner's activity to evidence, implement within scope and check the result. For a review, inspect the existing design instead of implementing. Reuse known context; do not require a planning document for every step.
Read [decision examples](references/decision-examples.md) when help, alternative answers or uncertain curriculum alignment needs clarification. They illustrate decisions, not scripts to copy into every lesson.
## Establish the learning context
Use information already present in the request, project and supplied materials. For context gaps, ask only about information that materially changes the design. For minor uncertainties, state a reasonable assumption and proceed. Avoid a fixed questionnaire or repeated approval for ordinary choices already covered by the request.
Identify the context that changes the design: learners and prior knowledge, subject, intended outcomes, classroom or independent use, available time and devices, language, accessibility needs, and any supplied curriculum. Follow a curriculum when relevant to the setting; do not assume one national framework. Choose tone from the context and author preference. Do not infer attention, motivation, ability or preferred pedagogy from generation labels.
Include the surrounding teaching activity where useful: preparation, individual or group work, teacher intervention, discussion and follow-up. A teacher-led demonstration may need a good discussion prompt; accounts and individual scores are not required.
Separate age, educational role and experience with the topic; child, pupil, student and adult participant are neither interchangeable categories nor ability levels. When establishing or changing the audience, read the learner profile in [design dialogue](references/design-dialogue.md). Adapt terminology and support to the setting; use neutral terms in general guidance. The creator, learner and accompanying teacher or parent can be different people.
## Connect goals to observable activity
For each important goal, connect the learner's action, evidence, success criteria, assistance conditions and next instructional decision. Keep this concise for small tasks. For open inquiry, the evidence may concern question quality, use of evidence or revision of an explanation; one predetermined answer is not the only option.
Example: the learner predicts how changing resistance affects current, checks a model, explains any discrepancy, and solves a new case independently. Moving a slider or completing a screen alone does not establish understanding.
Use the observed learning obstacle to choose an intervention that fits it. Treat an explanation of an error as a provisional hypothesis, not a diagnosis. Select a small coherent set of methods; do not add features merely because they appear in the catalog. Read [methods](references/methods.md) when selecting or implementing a method, support sequence or progression rule.
For game mechanics, motivational features or educational-game reviews, read [learning games and purposeful gamification](references/learning-games.md). It covers playable learning loops, mechanic selection, evidence limits and implementation from small tools to longitudinal products. Adapt to primary, secondary, university and adult learners across countries using prior knowledge and local context.
The method reference explicitly covers Socratic learning, the Feynman technique, the Zone of Proximal Development and scaffolding, worked examples, retrieval and spaced practice, error-based learning, metacognition, inquiry, role-play, peer and cooperative learning, projects, flipped teaching and optional gamification. These are selectable approaches, not required features of every application.
Bloom's taxonomy describes cognitive goals and knowledge types. Keep difficulty and prerequisites separate; do not require six locked levels. For practical, interpersonal or other outcomes, choose evidence that fits the outcome and acknowledge what an on-screen task cannot establish.
## Choose a realistic realization
Distinguish AI used to build the application from AI used by learners in the finished product. Choose among reviewed static content, deterministic interactions or simulations, and runtime AI according to the goal and operating constraints. Hints, feedback and spaced practice can work without an AI service.
Before adding a substantial feature, consider its learning contribution, interaction burden, teacher workload, maintenance, data needs and running cost. Explain material tradeoffs. Do not add accounts, analytics, dashboards, a backend, paid services or external data flows merely to make the product appear complete. Confirm scope-changing decisions when not already authorized.
Read [simulations](references/simulations.md) when modeling a system, running an experiment or designing role-play. Clearly distinguish real observations, model calculations and fictional reconstruction. Never present invented AI output as a measurement or pretend an experiment ran.
Read [runtime AI](references/runtime-ai.md) only when the finished product calls an AI service. Define the teaching role, source boundaries, human oversight, data flows, cost limits and failure behavior. A prompt is not a substitute for application security.
Read [tutor modes and continuity](references/tutor-state.md) when implementing multi-step tutoring, practice/assessment transitions or resuming learning. Keep state and evidence in the application; these features do not require runtime AI.
## Make feedback and decisions trustworthy
Provide a useful next step after an error. In practice, allow help and a route to direct explanation or a worked solution when needed; avoid endless questioning. Reduce assistance based on independent performance, while keeping requested support available. An assessment may restrict assistance under explicit conditions.
Distinguish independent success, supported success, exposure to the answer and uncertainty. One correct guess, a confidence statement or two supported successes do not automatically establish mastery. Do not infer stable ability, motivation or a disorder from a short interaction.
Read [assessment](references/assessment.md) when designing scored tasks, rubrics, adaptivity or learning evaluation. Check subject correctness, plausible alternatives, numerical units and tolerances, and whether the task really requires the intended performance. Preserve valid alternative explanations and methods. Keep game rewards and speed separate from achievement unless they serve an explicit goal.
## Preserve access and teacher agency
Design for the WCAG 2.2 AA criteria that apply to the activity and the learners' actual needs. Check keyboard operation, visible focus, understandable labels and feedback, alternatives to color or sound alone, and justified or adjustable timing. Choose ways to perceive information and express understanding that fit the learner; do not assign learners fixed learning styles. Seek specialist accessibility help when the implementation warrants it.
Collect only data needed for the activity and identify who sees it, where it goes and how it is retained. Check provider, school and local requirements for the actual deployment; do not invent universal legal rules. A local anonymous activity need not acquire accounts or persistent tracking.
Give the teacher or responsible author a way to review content, correct judgments and intervene in inappropriate adaptation. An editable content file is enough; a management system is not required. Significant grading or placement decisions need accountable human review.
## Coordinate expertise without inflating the workflow
Use these perspectives as needed; one agent can cover them for a small task:
- Methodologist: intended outcomes, teaching context and instructional choices.
- Subject didactician: content accuracy, prerequisites, examples and misconceptions.
- Assessment designer: evidence, criteria, alternatives and limits of inference.
- Interaction designer: learner flow, help, accessibility and classroom use.
- Developer: data contracts, calculations, state, persistence and failure behavior.
- Reviewer: discrepancies between the goal, design and observed behavior.
Carry their decisions into implementation. Do not label a convenient technical choice pedagogically correct without evidence. Resolve material disagreements through source checking, a targeted check or an explicit unresolved question. Use independent review when complexity or consequences justify it; do not start a team for every task. Invoke other skills only for relevant work and when available.
## Verify and report at the right level
For implementation, check the affected learner path, the errors learners actually make, the help route and the failure states that matter. For analysis-only requests, report findings and proposals without changing the product. Adapt checks to the application: a simulation needs model checks; runtime AI needs service-failure and output-handling checks; a simple static activity needs neither by default.
Keep three conclusions separate: functional correctness, didactic suitability and observed learning outcomes. Technical checks alone do not establish educational effectiveness. A small pilot can reveal problems without proving causal benefit. Report actual checks and unresolved assumptions; do not invent learner observations or efficacy results.
At handoff, summarize the learning goal, selected approach, important operating assumptions, checks actually completed and remaining limits. Include a goal-to-evidence table only when it makes the result easier to use. Match the form to the user's request.
## Write learner-facing text without AI slop
The application's text carries the teaching. Apply the same standard to prompts, instructions, feedback and explanations:
- Cut empty adjectives ("modern", "engaging", "interactive") unless the next sentence shows what makes them true.
- State the claim directly; avoid "it is not just X, but Y" and "this is not X, it is Y".
- Skip announcements such as "in this activity you will learn"; say what the learner does.
- Feedback names what is correct, what is missing and the next concrete step.
- Keep numbers, units and conditions specific; no "significantly improves" without a measure.
- No emoji in buttons, headings or labels. Use text and a consistent icon set.
- Match the vocabulary to the learners and keep one term for one thing.
Read the text as a learner would. If a sentence could sit unchanged in another product, cut it or make it specific.
## Integration
When selecting a model, effort or speed, or adapting the loading mechanism, read [model and host profiles](references/model-profiles.md). This includes documentation-based guidance for OpenAI, Claude and Gemini. Keep one shared educational contract and report actual testing separately from intended compatibility. Do not load the profile for an ordinary educational task that does not need it.
Use it for educational games, quizzes, tutorials, simulations and reviews of their teaching design. The core requires no packages, API, specific AI model or student data. References are conditional. The native bundle uses this file and its references; `learning-methodologies.md` is the generated standalone edition. Installation, deployment and other external changes require authorization for those actions. This skill does not replace project instructions or authorize additional scope.
Design references: [WCAG 2.2](https://www.w3.org/TR/WCAG22/), [CAST UDL Guidelines](https://udlguidelines.cast.org/). These guide design and review; citing them does not certify conformance or educational effectiveness.