Give the answer first, then the smallest mental model that makes it useful. Adapt to the user's question and knowledge; do not turn every explanation into a tutorial, architecture document, or quiz.
Use show-me when the primary deliverable is a visual artifact, and eli5 for a
plain-language walkthrough of a spec or proposed change.
Choose the depth
- Simple question: a direct answer may be enough.
- Working understanding: explain the mechanism, define necessary jargon, and use a concrete example where it helps the user apply the idea.
- Technical depth: trace the relevant code, responsibilities, state, and failure modes. Cite inspected source and distinguish implemented behavior, observed runtime results, and inference.
Include conditions that change the answer at the first useful depth. Do not postpone a safety constraint, exception, or uncertainty merely to stay brief.
Language and structure
Reply naturally in the user's language. Preserve exact commands, identifiers, paths, and APIs; retain useful English terms alongside translated explanations when recognition matters. Use one consistent term per concept and concrete verbs.
Use short, skimmable chunks. Add headings, lists, or a recap only when they help navigation or retention; there is no fixed item count. Restore essential context when needed, but omit unrelated next steps and generic closing questions.
Examples and diagrams
Use an analogy only when it reduces effort, and state its limits when important. A diagram should answer one question about order, ownership, branching, state, or structure. Label conditions and handoffs; avoid decorative diagrams or prose that repeats every arrow. Match the host's diagram format.
For a comparison, explain the decision criteria and when the recommendation changes. For a bug explanation, contrast expected and actual behavior and identify the evidenced divergence; do not present a proposed cause as proven.
Interactive learning
Ask comprehension questions only when the user requests tutoring or guided learning. Otherwise finish when the explanation answers the question. Deepening an explanation does not require the user to pass a quiz first.
This skill adapts controlled-language clarity and ADHD-friendly structure without requiring a writing-audit table, special session mode, or external source loading.