learn-kit

Pedagogical kit. 3 skills (v3.2.0). Primary skill `three-views`: given a topic + source material (project files / external URL / pasted text), generates a user-chosen subset of 1-3 tier learning markdown (foundation 零基础版 / structural 结构版 / challenge 挑战版). v3.1.0 additive HITL expansion: Step 1.3 视角 multi-select (default all 3 checked, min 1) lets users pick subset; Step 4 5-cell granular multi-select (HTML / NLM audio / NLM video / NLM slide_deck / NLM mind_map; default all unchecked) replaces v3.0.0 3-coarse-cell layout. NLM cartesian = len(generated_tiers) × len(selected view-cycled types) + (1 if mind_map); max 10 when full defaults. v3.1.0 Step 5B re-run guard adds source_corpus_key equivalence check (prevents partial-rerun corpus contamination). Backward-compatible: defaults preserve v3.0.0 product output. HTML uses dual-mode grounding (concept→code via Explore for project file sources; concept→source-section for URL/text). NLM preserves all nlm-studio dogfood防护: per-step refresh_auth, real auth gate via notebook_list, source_add validation via notebook_get, 4-way re-run guard, quota right-sizing 4-way gate (3-way when degenerate), bounded polling 12×10s. Requires notebooklm-mcp MCP server (bundled) + one-time `nlm login` only when NLM opt-in is selected. Slash: `/learn-kit:three-views <topic>` (auto-discovered). v3.2.0 adds two pure-prompt in-chat explanation skills: `glossary` (`/learn-kit:glossary <term>`, 六槽 150–250字 速记卡) + `concept` (`/learn-kit:concept <concept>`, 六节 500–800字 深讲，2 跨域正例 + 失效边界) — no tools/files, filling the Q&A niche three-views disclaims; see [ADR]_LearnKit_Explanation_Skills_Addition.md. v3.0.0 baseline BREAKING: removed 4 skills (scaffold-learning, locate, scan, nlm-studio); renamed generate-tier → three-views. locate/scan algorithms preserved as manual recipes in plugin-internal [GUIDE]_LearnKit_Discovery_Recipes.md.