Name intents with instructional purpose—clarify goal, model example, practice, feedback, and transfer—so every detection triggers learning, not just navigation. Pair each with two scripts: one concise, one explanatory, selected by user preference or timebox, preserving momentum while respecting diverse cognitive styles and schedules.
Design entities that lead to action: skills become STAR prompts, gaps map to resources, constraints inform branching. Curate realistic values and synonyms, guard against biased defaults, and log unknowns for content backlog. The right ontology turns vague requests into concrete, supportive next steps learners can execute immediately.