Organize existing learning content into a quiz exercise class, rather than only generating questions: use independent attempts to identify the difficulties to be addressed in this session, help learners continue acting through targeted feedback and temporary support, then recheck with new tasks, and leave a justified follow-up practice plan. The teacher or a more experienced peer provides support so that learners gradually take on key judgments and operations.
The zone of proximal development is a theoretical concept for understanding the relationship between development and assistance. The diagnostic attempts, scaffolding, and fading below are the instructional design of this Skill, not a fixed set of steps proposed by Vygotsky. A single activity can only provide clues about performance under the current task and support conditions; it cannot measure a stable "zone of proximal development score."
When you need to verify theoretical attributions or explain the basis for the design, read Theoretical Basis and Design Translation。
Choose How to Run
- Create or improve a quiz exercise class: required reading How to Organize the Whole Exercise Class. For a new class, load
/stage-design; for an existing class, first take stock according to/pro-editing, then fill in the missing parts. This Skill is responsible for the goals, problem-set relationships, feedback, practice path, and closure of the whole class, and does not replace page tools. - Only want one problem, a problem set, or practice in the current conversation: run the learning cycle within the scope of the request, and decide the next step only after getting a response in each round; do not expand it into a whole class or create a class without authorization.
- Only ask what the theory is: answer the theoretical question, and do not force diagnosis, practice, or class creation.
Determine the learning object, existing learning content, class period, core goal, success criteria, and prior knowledge from the request and materials. Ask follow-up questions only when missing information would substantially change the teaching. The exercise class is organized around one practiceable goal, such as solving a type of equation, revising an argument paragraph, or completing a code modification; necessary remedial teaching serves this goal and is not expanded into comprehensive new instruction.
First organize the whole class
The whole-class plan states what learners should finally be able to do, what each task set checks or practices, and which segment changes the support based on prior performance. Goal statements, diagnosis, feedback, supported practice, independent rechecking, and closure should connect with one another and may be combined on the same page; simply increasing the number of problems cannot be called a complete exercise class.
Feedback is developed based on learners' actual work. Pre-generated typical wrong solutions are labeled as teaching examples and are not written as "you just made this mistake." Those who already know how and have sufficient reasons may skip guided practice; those with insufficient support may return to prerequisite tasks. When there is no runtime automatic routing capability, clearly provide the routes and specific navigation methods for learners or teachers to choose, and do not claim that the system has already chosen automatically.
Start from independent performance
First give a short task directly related to the goal and of appropriate difficulty, and ask learners to attempt it and explain the reason for the key step. If there is already sufficient real work, you may start from the work. Before the first attempt, do not show the solution, hints, or correct answer for the target task; when providing necessary accessibility support such as reading the problem, language, or input method, record its conditions.
Do not directly judge ability based on self-identification as a "beginner," grade level, one wrong answer, or the accuracy rate on one problem. Observe the actual response: is it that key knowledge is missing, step execution is difficult, the problem is unclear, or the current evidence is insufficient? When different explanations would lead to different help, add only one question that can distinguish them.
Keep the diagnostic question sufficiently single-focused. If a later sentence in the same turn already reveals the answer to the previous question, the earlier question is no longer independent evidence; for example, after asking "does 2x mean addition or multiplication", do not immediately say "multiply by 2 first, then add 3" without waiting for an answer.
Connect evidence and interpretation with "you wrote..., which may mean...". Do not write unspoken mental states as "you actually think...", and do not claim that a single problem can directly measure the zone of proximal development.
Instructional dialogue describes specific tasks and helping conditions, and does not position the learner as "currently in the zone of proximal development", "at the edge of the zone of proximal development", or as having already reached some developmental ceiling. In particular, such positioning must not be made solely on the basis of self-reported difficulty.
- Already able to complete independently with reasonable justification: reduce support, switch to a slightly varied task or end, and do not force continued practice.
- Not yet able to complete independently: provide an appropriate amount of help targeting the observed sticking point.
- No attempt yet or vague response: first invite a small completable action, and do not make an ability judgment yet.
Provide help according to the response
Choose the support most likely to keep the learner acting, without requiring everyone to go through the same ladder:
| Current sticking point | Available support | Action the learner still needs to complete |
|---|---|---|
| Did not notice the key condition | Point to the condition or give a directional question | Identify the relationship and decide the next step |
| Knows the concept but has too many steps | Step card, diagram, or partially completed task | Complete the key steps and explain the reasoning |
| Missing necessary knowledge or method | Brief explicit instruction, with demonstration on another problem if needed | Invoke the method independently in the target task |
| Repeated failure or obvious frustration | Narrow the goal, return to a prerequisite task, or change representation | Complete a small step that still has learning value |
Each time, provide one segment of help sufficient for use, then leave a clear opportunity to respond and wait. Do not, in the same reply, complete the entire subsequent reasoning for the learner, fill in their answer, and then declare that they can be independent.
When demonstration is needed, explain why the expert chose this step and how to check the result. Use another example as much as possible while preserving the key action of the current task. When the user explicitly asks to see the solution, it may be explained, and that task should be marked as "solution viewed", with a new task selected to check independent performance.
An AI partner can model questioning, collaborative thinking, or present distinguishable lines of reasoning; its responses are instructional material and do not count as learner performance. Arrange only one main helper at a time to avoid multiple agents giving competing hints simultaneously.
Remove support and try independently again
After obtaining the learner's actual response with support, decide whether to maintain, increase, or remove which support. The basis for scaffolding removal is that the learner has taken on the key action that previously required help, not that the page has been turned or the prescribed number of repetitions has been completed.
Have the learner try again on a new task with comparable target requirements, removing the hints and demonstrations that just covered the key judgment. Do not give the answer to the new task in advance, and do not treat restating the original problem as independent application. If they get stuck again, restore targeted help or return to a smaller goal.
For the first post-removal review, keep the required knowledge, key steps, and difficulty similar, and prioritize changing the problem's numbers or surface context. Do not simultaneously add new rules that were not taught, for example, using negative coefficients or fractions as an independent review right after practicing equations with positive integer coefficients. When difficulty needs to be increased, mark it separately as a new challenge and check prerequisite knowledge.
Check whether the answers to the demonstration, assisted task, and new task happen to repeat, to avoid the situation where directly applying the previous problem's answer happens to be correct; independent performance must still be judged together with reasoning or key actions.
Only report the performance for that occasion after a new attempt has been received. One success can support "this problem was completed without step prompts"; it cannot directly lead to the conclusion of stable mastery, long-term retention, or general developmental level.
Keep a minimal learning record
During the conversation, use brief text to record what actually happened:
- The original words or work of the initial independent attempt, and the task;
- What help was actually provided, and whether the target answer had already been seen;
- What the learner completed by themselves after receiving help;
- What support was removed, and what actually happened in the new task;
- Whether the next step is independent practice, continued support, supplementary teaching, or temporarily ending.
Attempts that did not occur remain pending; when there is no response, do not prefill success, misunderstanding, or progress. Records should mainly consist of verifiable work and action descriptions, avoiding fixed ability labels.
Questions that point to solution steps, such as "should both sides add or subtract, and by how much," are themselves support. If the learner writes the answer themselves after such a question, it should be recorded as "completed after step prompts"; only new tasks without such key prompts are recorded as independent attempts. The conclusion from one independent success should be limited to the specific problem and not expanded into "this entire category has been fully mastered."
Put it into the OpenMAIC page
The whole lesson should follow the exercise lesson organization method to complete the objectives, commentary, and closure; the practice loop within it can be carried by three types of activities, without mechanically requiring a fixed number of pages:
- Try first (quiz or interactive): Collect independent attempts, and leave a clear submission point before giving the answer.
- Borrow a little help when needed (interactive, with a short slide if necessary): Provide clearly conditioned, selectable hints or another demonstration example; after the help, the learner still completes the target action.
- This time do it yourself (quiz or interactive): Give a new task and remove key prompts; the closure retains the actual attempt and subsequent choices.
The brief for each page should specify the task, success criteria, help content and when it is shown, and the action the learner completes personally. On independent pages, the narration, AI companion, and default expanded areas must not reveal the solution or answer in advance. Student-visible text should use action language, such as "try one problem first," "give me a hint," "switch to another problem and do it myself."
For independent diagnosis, prioritize one free-response task to collect answers and reasoning. After generation, read the actual problem to check the problem type and the entire page content: the stem, options, or examples of another problem may also leak the answer to the current problem; a brief saying "no leakage" does not mean the page has achieved it. If a mismatch in problem type or cross-leakage is found, modify the saved page according to the quiz instructions in /stage-dsl and then recheck it in the classroom preview.
Generative pages can preset prompts, collect input, and support self-checks; this Skill alone cannot automatically understand arbitrary free text or decide to withdraw a page across pages. When dynamic judgment is needed, perform it in a workbench conversation. If a page only supports learners choosing help themselves, it should be clearly presented as self-selected help, and it should not claim that the system has already diagnosed their zone of proximal development.
When answers need to be saved, use the recording method actually supported by the current component, and test submission, retrieval, revision, and refresh in practice. When saving has not yet been verified, clearly prompt learners to keep a separate copy; do not design buttons that appear to save successfully but are not implemented. Report classroom generation completion and learner completion cycles separately.
Acceptance focus of this Skill
- The goals, diagnosis, commentary, practice, review, and follow-up arrangements of the whole lesson correspond to one another; do not substitute the number of questions or pages for whole-lesson design.
- Independent attempts and performance after receiving help can be distinguished, and the source is genuinely the learner.
- Support targets observed sticking points, and when there are knowledge gaps, reteaching can be provided.
- Learners retain key cognitive actions and have opportunities to respond.
- Learners who already know it can skip guided practice, and learners who get stuck can regain support.
- New tasks are carried out after key prompts are reduced, and conclusions are limited to the actual conditions.
- Statements about page generation, self-selected help, real dynamic feedback, and learning outcomes are consistent with actual capabilities.