ai4learning-mvp

name	ai4learning-mvp
description	Use when a learner wants help studying dense text materials such as papers, textbooks, notes, or technical docs and needs a structured AI learning workflow that classifies unknowns into 1/2/3A/3B, converts them into verified understanding, and outputs a reusable learning path instead of ad hoc Q&A.

AI4Learning MVP

Overview

This skill turns a learning session into a stateful workflow:

material -> node map -> 1/2/3A/3B classification -> intervention -> verification -> next path

Use it when the goal is real learning, not just answering the next question.

When To Use

Use this skill when:

• The user is learning from dense text such as papers, textbooks, lecture notes, RFCs, or technical docs
• The user is blocked by multiple unknowns and does not know what to ask first
• The task requires turning confusion into a trackable learning path
• The user wants evidence of understanding, not just a polished explanation

Do not use this skill when:

• The user only wants a direct factual answer with no teaching workflow
• The source material is mainly video, audio, or image-based with no usable text yet
• The main task is curriculum authoring rather than active study guidance

Operating Rules

1. Always build or update a SessionState before teaching.
2. Never mark a node as 1 unless there is observable evidence.
3. Prefer queue order: high-leverage 2 -> 3A -> blocking 3B.
4. Keep the active queue small. Default to 5-12 active nodes per session.
5. If required inputs are missing, make minimal safe assumptions and label them.
6. Avoid long lectures. Each intervention should reduce uncertainty or improve transfer.
7. If a node keeps spawning new hard prerequisites, reclassify it as 3B.

Required Inputs

Build the session around these five fields:

• learning goal
• learning material
• current background
• time budget
• desired outcome

The exact schemas and lifecycle rules live in references/session-contract.md.

Capability Workflow

1. start_session

• Create LearnerProfile, LearningSource, and the initial success condition.
• Normalize scope. If the user gives a full paper or chapter, choose the most relevant subsection first.
• Record assumptions explicitly.

2. map_material

• Chunk the material into concepts, claims, procedures, and prerequisite links.
• Extract candidate KnowledgeNodes.
• Mark obvious foundation nodes that the learner already seems to own.

3. classify_nodes

• Assign each node to 1, 2, 3A, or 3B.
• Use the hard rubric in references/classification-rubric.md.
• If classification is uncertain, explain the ambiguity instead of hiding it.

4. resolve_fast

• Work first on the highest-leverage 2.
• Tactics for 2: definition, contrast, example, short correction.
• Tactics for 3A: analogy, intuitive bridge, simplified reframe, lightweight visualization in text.

5. decompose_hard

• Use this only for true 3B.
• Decompose the node into 1 + 2 + 3A + 3B.
• Expose the dependency tree, then move the new 2 and 3A back into the active queue.
• Do not try to finish the entire tree in one answer.

6. verify_understanding

• Default evidence is:
  • learner teach-back in their own words
  • one transfer or application question
• Confidence alone is not evidence.
• If the learner fails teach-back or transfer, downgrade the node.

7. summarize_path

• End every session with a concise state snapshot and the next best move.
• Use the fixed output format from references/output-templates.md.

Class-Specific Tactics

Class	Meaning	Default tactic	Expected transition
1	Already understood and usable	Use as foundation, do not reteach	stays 1
2	Clarifiable in <= 3 turns	definition, contrast, example	2 -> 1
3A	One explanation can reduce complexity	analogy, intuition, bridge	3A -> 2 -> 1
3B	Requires recursive prerequisite splitting	dependency decomposition	3B -> 1/2/3A/3B

Output Contract

Always structure the response so the user can see:

1. Current objective and assumptions
2. Foundation nodes (1)
3. Active node board with class and rationale
4. Interventions performed
5. Verification result
6. Remaining 3B queue
7. Next best question or task
8. Next-session plan

Use the exact templates in references/output-templates.md when creating a new session or closing one.

Anti-Patterns

• Explaining everything before classifying anything
• Accepting “I think I get it” as mastery
• Over-splitting easy nodes into fake 3B
• Teaching downstream concepts before clearing the blocking prerequisite
• Turning the session into a content dump with no state transition

Reference Loading Guide

• Read references/classification-rubric.md when classifying, reclassifying, or validating mastery.
• Read references/session-contract.md when creating or updating the state model.
• Read references/output-templates.md when formatting a live session or summary.
• Read references/evaluation-protocol.md when benchmarking the skill, designing pilots, or reviewing transcripts.