USE WHEN: teaching complex topics requiring chunked delivery, handling mid-explanation questions, preserving context across interrupts, deep-dive explanations with nested topics. DO NOT USE WHEN: simple one-shot explanations, quick answers, or when @explainer skill is sufficient for the depth needed.
This skill produces chunked, resumable explanations with recursive depth and lineage tracking. Solves the "curiosity interrupt" problem—when learners ask sub-questions mid-explanation, context is preserved and continued.
User provides code or concepts to understand. They may interrupt mid-output with follow-up questions. Track everything; orphan nothing.
Before explaining, plan the thread structure:
>> n sections needed? Each chunk = one digestible concept (3-5 sentences max).[Q₀ → Q₁(>> n) → Q₂] notation.CRITICAL: Output is a resumable stream, not a monolith. If user interrupts at >> 3 asking about a term, answer BRIEFLY (3-5 lines), state the gap filled, then RESUME from >> 4 with remaining chunks. Never orphan pending content. Use PENDING: >> 4 - >> 7 notation. Interrupt mode = Brief answer + Resume. Always.
Every response MUST include these elements in order:
╭─ THREAD ─╮ showing Q₀ → Q₁(>> n) → Q₂ chain, plus PENDING: >> n - >> m if interrupted┌─ GAP ─┐ stating what this answer fills and how it serves parent questions>> 1 ──── through >> n ────, each chunk self-contained, levels marked inline? Check: comprehension question before advancing to next level-cluster>> n - >> m: [topic preview] so learner sees the pathFocus on:
>> n chunk = ONE concept, 3-5 sentences. If longer, split. Chunks enable interruption without context loss. Number every chunk explicitly.Level 1: and Level 2: markers INSIDE chunks when depth is needed. Level 1 = foundation. Level 2 = mechanism. Never jump levels.>> n, trigger interrupt: brief answer (≤5 lines), state gap filled, show PENDING: >> (n+1) - >> m, then RESUME with remaining chunks woven with new insight.[Q₀ → Q₁(>> 3) → Q₂] means Q₁ was asked at chunk 3, Q₂ is current. Never verbose trees—one line shows the chain.? Check: question. Wait for response. If answer reveals gap, address immediately before continuing. No rhetorical questions.NEVER output 90-line walls without chunk breaks, skip the THREAD or GAP boxes, answer interrupt questions verbosely (brief mode only), forget pending chunks after interrupt (always resume), use jargon without recursive Level 1 explanation first, or advance past a ? Check: without learner response.
See references/output-examples.md for complete interrupt-mode and standard-mode output templates.