lingochunk-course

LingoChunk course planner

Turn one rich episode (or a collection) into a coherent SERIES of lessons: a named course with N ordered lessons, each a self-contained lingochunk-lesson document with a DIFFERENT grammar point and a difficulty that ramps across the series. LingoChunk supplies the materials (transcripts with positions and translations, FSRS-graded vocabulary) and renders every lesson natively; your job is the arc - how to slice the source, what each lesson teaches, and in what order - and then to drive the existing lesson flow once per slice.

This skill uses the lingochunk MCP tools. If they are not available, tell the user to add the LingoChunk MCP server (see the plugin README) and stop. This skill ORCHESTRATES lingochunk-lesson: read that guide (get_authoring_guide with topic='lesson') before composing any single lesson - this one only adds the multi-lesson planning on top.

When to use

• "Make a course out of this episode."
• "Build me a series of lessons on the B1 podcast, one per scene."
• "Break this 20-minute episode into a study plan I can work through."
• "Turn this collection into a beginner course."

Options to settle first (ask only what the user left open)

1. Source: one episode (list_library), or a collection you slice episode by episode. A single lesson has ONE source episode in v1, so each lesson in the course draws from one episode (or one slice of it).
2. Length: how many lessons. Default to what the material honestly supports - a 3-8 minute slice per lesson, so a 20-minute episode is ~3-5 lessons, not fifteen thin ones. Say the number you chose and why.
3. Level and ramp: the starting CEFR level and whether it climbs. Infer the floor from the CEFR mix of get_vocabulary(status=known) if the user does not say. State the ramp (e.g. "A2 for lessons 1-2, B1 for 3-4").

Workflow

1. Inventory the source. list_library to find the episode(s). For each, pull the shape: get_transcript (sentence positions, timings, speakers) and get_vocabulary (what is known vs learning/new/due). You are looking for natural seams - scene changes, topic shifts, a run of sentences that share a grammar pattern - and for which grammar points the material can actually evidence.
2. Split into coherent slices. Carve the source into N slices, each a 3-8 minute span that stands on its own AND carries enough evidence for one grammar point. Prefer seams the audio already has (a new scene, a new speaker turn) over arbitrary time cuts. Write down, per slice: its from_time/to_time, its ONE grammar point, its level, and 2-3 archetypes from the lesson skill's menu - deliberately varying the archetype mix across slices so consecutive lessons do not feel identical.
3. Ramp the difficulty. Order the slices so difficulty climbs: earlier lessons lean on recognition (MCQ, matching, sentence reorder) and lower-level grammar; later lessons lean on recall and production (dictation, shadowing, gap-fill, production) and harder points. No grammar point repeats across the course - depth over coverage, one point per lesson, a different point each lesson.
4. Create the course. create_course with a title and a short description naming the arc. Keep its id.
5. Build the lessons in order. For each slice, run the full lingochunk-lesson flow (pull the slice, filter vocabulary, pick the ONE grammar point you assigned it, compose on the scaffold with its archetypes). Before each save, call validate_lesson with {document} and fix EVERY problem it reports; only once it returns valid: true do you save_lesson with {document, course_id, sequence} - sequence = the lesson's 1-based position in the course (ties break by created_at, but pass explicit sequences so the order is deliberate). Stamp generator: {skill: "lingochunk-lesson"} on each document.
6. Deliver. List the course and its lessons in order (title, level, grammar point, app_url), so the user can work through them as a series. Offer add_card for the drill words the course introduced. Confirm the course groups them: list_lessons echoes each lesson's course_id, sequence and course_title.