lecture-to-study-guide

lecture-to-study-guide

Produce a pedagogically structured study guide from lecture material — not a summary, a study-optimized document.

When to use this skill

Trigger on any phrase pairing lecture material + a study/review/exam goal. Concrete cues:

Student voice (lead — higher frequency):

• "turn my notes into a study guide"
• "summarize this lecture for the test"
• "make a study guide from these slides"
• "I have a transcript from class, help me review"
• "convert this PDF of my [subject] lecture into a study guide"
• "here's a YouTube lecture transcript, extract the key points"
• "make me practice questions from this chapter"
• "I need a cheat sheet for tomorrow's exam from these notes"
• "help me review — I'll paste my lecture notes"

Teacher voice:

• "make a study guide for my students from this lecture"
• "build a review handout from my slide deck"
• "give my class a review sheet from these notes"

Inputs

Required:

• Source material — pasted text, PDF, or transcript. May be messy.

Optional (infer if absent — see table below):

• Subject + level — for calibration (6-8 / 9-12 / college / graduate).
• Source type — pasted prose notes / slide deck / transcript / multi-lecture / mixed.
• Exam format — MCQ / essay / problem-set — biases question mix.
• Target length — default: proportional to input.

Inferring level and source-type from phrasing (before asking)

Attempt inference first. Ask ONLY if both level AND source-type are unclear.

Cue in user's phrasing	Likely level
"my 7th grader," "middle school," emoji, simple vocab	6-8
"AP ___," "honors," "SAT prep," "high school"	9-12
"my professor," "syllabus," "midterm," "lecture hall," "TA," intro-course names ("intro micro," "bio 101")	college (default)
"my PI," "qualifying exam," "seminar," "dissertation," advanced domain vocabulary	graduate
No markers, just raw notes	infer from vocabulary density; default college

Cue in user's phrasing	Likely source-type
"my notes," "typed up," "outline"	pasted prose notes
".pdf," "slides," "deck," "slideshow," "lecture slides"	slide-deck PDF
"transcript," "YouTube," "recording," "VTT," "SRT," "captions," "Otter," "Zoom transcript"	transcript
Multiple files listed, "weeks 1-3," "chapters 4-6," "my last three lectures"	multi-lecture series
Both a slide deck AND a transcript mentioned	mixed slide + transcript

If confidence is still low on BOTH dimensions, ask ONE question in this exact shape:

Quick check so I can calibrate this — (a) what's the level (e.g. 7th · 11th · college · grad), and (b) is this pasted notes, a slide deck, or a transcript?

Do not stack questions. Do not ask about exam format, length, framework, etc.

Workflow

1. Ingest. Detect input type. Normalize PDFs and transcripts to plaintext-with-structure-markers (slide boundaries, speaker turns, heading lines).
2. Detect structure. Find natural segmentation: markdown/outline headings, slide boundaries, topic shifts in transcripts, paragraph breaks. See references/input-types.md.
3. Extract terms. Use heuristics in references/term-extraction.md — typographic (bold/italic/"quotes"), structural (slide titles, definition blocks), linguistic ("X is defined as…", "Y, also known as…"). Build a term → definition → source-location map.
4. Group topics. Cluster concepts into 3-8 topic groups. Prefer source-provided groupings (slide sections, chapter headings) over model-imposed ones.
5. Generate practice questions. Mix by level and exam format. Default by level: see table. See references/question-mix.md for templates.
6. Assemble the 7-section markdown guide per references/output-format.md. Validate: all sections present, TOC anchors resolve, Mermaid blocks syntactically valid.
7. Offer follow-ups — optional PDF export and the closing suggestion: "Run the flashcards skill on this guide next for Anki-ready decks." Do NOT auto-invoke sibling skills.

Output

Single markdown file study-guide-<topic-slug>-<YYYY-MM-DD>.md with these sections, in order:

1. Outline / Table of Contents — hierarchical, anchor links.
2. Key Terms + Definitions — glossary table, columns: Term | Definition | Source (slide #, timestamp, or heading).
3. Concept Summaries — one subsection per topic group, 2-5 short paragraphs each (SQ3R framing).
4. Diagrams — Mermaid blocks where content warrants. Inline; do NOT call a separate concept-map skill in V1.
5. Worked Examples — step-by-step solutions for quantitative/procedural content. Use LaTeX math ($...$, $$...$$).
6. Practice Questions — 10-20 questions, mixed types. Answers in <details> blocks per question (default) or a trailing answer-key section (for printable/PDF output).
7. Cheat-Sheet One-Pager — ~1 printed page. Most-testable facts, formulas, top 5 terms. Cornell-style (cue / note + synthesis).

Grade/level calibration

Level	Scaffolding	Question mix (recall/comprehension/application)	Cheat-sheet style
6-8	Heavy. Plain-language definitions for every term. Short sentences. Diagrams liberally.	60 / 30 / 10	Large type, spatial layout
9-12	Moderate. Define technical terms, assume general vocabulary.	40 / 40 / 20	Dense but still visual
College (default)	Terse. Assumes college vocabulary.	30 / 40 / 30	Information-dense single column
Graduate	Terse. Reminder-level summaries. Assumes domain vocabulary.	20 / 30 / 50	Formula/theorem-dense

Exam-format adjustments (stack onto grade default): MCQ-heavy → +10% recall, −10% application. Essay → +15% comprehension, −15% recall. Problem-set → +20% application, −10% recall, −10% comprehension.

Input handling

PDF and transcript ingestion is currently handled by Claude directly; dedicated parsers (pdf_extract.py for slide-boundary-sensitive decks, transcript_clean.py for VTT/SRT normalization) are planned. For V1, rely on Claude-native PDF and do the transcript cleanup in-prompt (strip timing noise, merge fragmented caption lines, preserve speaker labels if present). See references/input-types.md for per-input-shape workflow.

References (loaded on demand)

• references/study-techniques.md — Cornell, SQ3R, Feynman patterns; informs cheat-sheet layout and concept-summary structure.
• references/term-extraction.md — term-identification heuristics + decision table.
• references/question-mix.md — question-type taxonomy, grade-calibrated ratios, templates.
• references/input-types.md — per-input-shape workflow (slide-heavy vs prose-heavy vs transcript vs mixed); slide-vs-prose heuristics.
• references/output-format.md — exact markdown skeleton + conventions.

Load a reference only when its topic is active. Do not preload all references at the start.

Locked defaults

Setting	Default
Level	College (unless inferred otherwise)
Source-type	Infer from phrasing + content
Question mix	Per grade-calibration table
Answer-key style	<details> blocks per question (switch to trailing answer key on PDF/printable request)
Diagrams	Inline Mermaid
Flashcards / concept-map	NOT auto-invoked in V1. Suggest running the flashcards skill as a follow-up.
Output	Markdown file in user's cwd
PDF export	On request only, via shared/scripts/pdf_render.py

Micro-example

User: "Turn these into a study guide — 10th grade bio, cell division. [pastes ~1500 words on mitosis]"

Inferred: level 9-12 (explicit "10th grade"), source-type pasted prose notes. No clarifying question needed.

Skill responds with a file starting:

# Study Guide: Mitosis & Cell Division

_Generated 2026-04-14 from pasted lecture notes. Level: 9-12._

## 1. Outline
- [2. Key Terms](#2-key-terms)
- [3. Concept Summaries](#3-concept-summaries)
  - [3.1 Phases of Mitosis](#31-phases-of-mitosis)
  - [3.2 Cytokinesis](#32-cytokinesis)
...

## 2. Key Terms

| Term | Definition | Source |
|---|---|---|
| Prophase | First phase of mitosis; chromatin condenses into visible chromosomes and the nuclear envelope breaks down. | Heading: Phases |
| Metaphase | Chromosomes align along the cell's equator, attached to spindle fibers at the centromere. | Heading: Phases |

...

## 6. Practice Questions

1. During which phase of mitosis do chromosomes align at the cell's equator? *(Term: metaphase)*

   <details><summary>Answer</summary>
   Metaphase.
   </details>

2. Explain why the spindle apparatus is essential for mitosis. If it failed to form, what would happen? *(Concept: spindle)*

   <details><summary>Answer</summary>
   The spindle pulls sister chromatids apart during anaphase. Without it, chromosomes cannot separate evenly, producing daughter cells with the wrong chromosome count (aneuploidy).
   </details>

Full 7-section guide follows; cheat sheet is ≤ 700 words for grade 9-12. Closing suggestion: "Run the flashcards skill on this file to generate Anki-ready cards from the Key Terms and Practice Questions."

Edge cases

• Very short input (<500 words): produce a proportionally shorter guide; reduce practice-question count to 5-8. Do not pad.
• Very long input (>20k words): warn the user, offer to split. Otherwise chunk by section and synthesize a top-level Outline + Cheat Sheet.
• Slides with no surrounding prose: elaborate bullets into concept summaries; mark model-added content with (expansion) so the student sees what came from the slide vs. what was filled in.
• Transcript with no speaker labels: treat as monologue. Do not invent speakers.
• Scanned/OCR'd notes: flag uncertainty. Ask the user to verify Key Terms before finalizing.
• Non-English content: respect source language; generate guide in source language unless translation is requested.
• Math/science heavy: preserve equations in LaTeX. Worked examples show intermediate steps. Prefer Mermaid for conceptual flow; fall back to prose description for geometry/waveforms.
• Multi-lecture series: ask once whether to produce a unified guide or a per-lecture set (default: unified with per-lecture H3 subsections). De-duplicate Key Terms across lectures.

V1 non-goals (V2+)

• Auto-invocation of flashcards or concept-map skills. V1 embeds inline Mermaid; flashcards are suggested as a follow-up, not generated inline.
• Bundled pdf_extract.py / transcript_clean.py parsers. Claude-native handling for now.
• Explicit translation flag.
• "Quiz mode" output (questions-only); punt to quiz-generator.

Quality bar

Every generated guide must satisfy:

• All 7 numbered sections present and non-empty.
• Key Terms table ≥ 5 rows (target: 8-20 for a 45-min lecture).
• Practice Questions meet the minimums in references/question-mix.md (5-8 for <500 words, 10-15 for 500-5000, 15-20 for >5000 or multi-lecture).
• At least one Mermaid diagram (unless content is fundamentally non-diagrammable — in which case include a one-line justification).
• TOC anchor links resolve.
• LaTeX for math (never unicode substitution for operators).
• Cheat sheet ≤ 500 words (college/grad) or ≤ 700 words (6-8/9-12).
• Grade-band calibration applied to vocabulary, scaffolding, and question mix.

What NOT to do

• Do not auto-invoke sibling skills (flashcards, concept-map, quiz-generator). Suggest, don't invoke.
• Do not fabricate content not supported by the source. Mark any model elaboration beyond the literal source with (expansion).
• Do not over-summarize. This is a study guide, not a recap — keep Key Terms, Worked Examples, and Practice Questions substantive even if summaries are terse.
• Do not skip Worked Examples for quantitative material, even if the source only shows final answers — derive the steps.