| name | lecture-notetaker-adhd |
| description | Generate ADHD-optimized Obsidian markdown study notes from lecture slides and transcripts — short capped chunks, per-section time estimates, static lecture-through percentages, session/break maps, micro-checks, and rotating interaction styles. Same learning techniques as lecture-notetaker-abbr with delivery tuned for sustained attention. |
| disable-model-invocation | true |
Lecture Notetaker (ADHD-optimized)
Invocation
Manual only. Apply this skill only when the user invokes /lecture-notetaker-adhd or explicitly selects it. Do not load or run it based on conversation context alone. For standard notes without ADHD-oriented delivery, the user should invoke lecture-notetaker-abbr instead.
Same learning techniques as lecture-notetaker-abbr (retrieval practice, spacing,
pre-questions, elaborative interrogation, interleaving). What changes is delivery —
pacing, chunking, and feedback — to keep re-recruiting ADHD attention.
No Obsidian plugins required beyond what the core notetaker skills already need (e.g. PDF++).
Before starting
- Read
/mnt/skills/public/pdf-reading/SKILL.md.
- Read
references/output-template.md — follow it exactly.
- If
.pptx slides, also read /mnt/skills/public/pptx/SKILL.md.
ADHD design principles
- Cap the chunk. Max ~5–7 bullets per callout. Split longer content. ADHD WM ≈ 3–4 items.
- Make progress visible. Static ~X% through lecture on the Game Plan and each objective (from time estimates) — position in the material, not checkbox completion.
- Time-box it. Minute estimates per objective + session map. Fights time blindness.
- Offload WM. One-line "gist" per objective; "You are here" recaps between chunks.
- Inject novelty. Rotate interaction type across objectives — sameness kills attention.
- Tighten feedback. Micro-check after every 1–2 slides, answer immediately foldable.
- Keep the scaffold. Objective-level structure + whole-objective recall at end. Chunk the reading, not the retrieval.
- Cheap re-entry. Clear stop points, "resume here" cues, explicit small first step.
- Respect the reader. Adult tone. No children's-worksheet energy.
Inputs
| Input | Format | Required? |
|---|
| Slides | .pdf | Yes |
| Transcript | .txt/.srt/.md/pasted | Yes |
| Learning objectives | List | Yes — ask if missing |
| PDF filename in vault | String | Yes — ask if unclear |
| Session length | Minutes | Optional (default 25) |
.pptx → ask for PDF export or convert: soffice --headless --convert-to pdf.
No objectives → infer 3–5, label as inferred. No transcript → suggest references/transcriber-guide.md.
Workflow
1. Extract slide text
python3 scripts/extract_slides.py <path-to-slides.pdf>
Outputs /tmp/slides_extracted.json. Fallback: pdftotext slides.pdf /tmp/slides-text.txt.
2. Map slides → objectives
Each slide → one or more objectives. Unmapped slides (title, admin) → Supplementary.
3. Segment transcript → slides
Walk linearly; assign by shared terms, topic shifts, marker phrases ("on this slide", "moving on"). .srt → use timestamps.
4. Extract abbreviations
Build ABBR → full term map from slides + transcript. Add discipline-standard expansions from own knowledge.
Tag with <abbr title="full term">ABBR</abbr>. First occurrence: <abbr title="full term">full term (ABBR)</abbr>. Don't tag inside quotes or the Abbreviations table.
5. Filter redundancy
Keep: why, examples, analogies, caveats, cross-topic links, exam flags, definitions the slide only names, processes the slide only labels.
Drop: slide restatements, filler, duplicates. Purely redundant transcript → one-line note.
6. Size segments
Check each slide's content against chunk cap (~5–7 bullets). Over → split into labeled callouts ("Slide 9 — Myelination (1/2)", "(2/2)").
7. Estimate time
~150–200 words/min reading + ~1 min/micro-check + ~2–3 min/recall check. Round to nearest 5.
8. Build session map + lecture-through %
Group into sessions (default ~25 min). Break at each boundary. Pick one tiny first step → Game Plan.
Assign ~X% through lecture per objective: cumulative estimated minutes before that objective ÷ total study minutes (objectives + Self-Test only), rounded to nearest 5. Put % on Game Plan lines and each objective’s ⏱ line. Self-Test heading ~85% (adjust if few/many objectives).
9. Generate markdown
Follow references/output-template.md exactly. Key elements:
- Game Plan — sessions, time estimates, first step,
· ~X% on each objective line
- Per objective —
⏱ + ~X% through lecture; gist; pre-question; capped slide callouts; micro-checks as - [ ] inside [!question]+; rotating interaction; answer block; sample answer; recall as - [ ] inside [!success]
- Break markers at session boundaries
- Self-Test — interleaved
- [ ] inside [!question]+ groups; heading includes ~% through lecture
- No
apb blocks, Progress Tracker section, or #obj / #selftest / #review tags
- Abbreviations, Key Terms, Gaps, Review Schedule (plain checkboxes)
10. Deliver
Save .md, present it, point reader at Game Plan + first step.
Quality rules
Pre-questions: Must create a curiosity gap requiring relational understanding. ❌ "What are glial cells?" ✅ "Why can't neurons function without glial cells?"
Micro-checks: One question per check, answerable in a sentence, targeting the slide just covered. Vary the verb. Every objective gets ≥1 micro-check — the rotating element is in addition, never a substitute. Place as - [ ] inside the [!question]+ callout.
Lecture-through %: Must be consistent between Game Plan and objective headers. Static only — never imply they update when boxes are ticked.
Checkboxes: Use normal Markdown checkboxes for tasks the reader may tick in Obsidian. Do not generate apb blocks or hashtag task tags.
Redundancy filtering: Good = adds to slide (analogy, mechanism, context). Bad = restates slide bullet.
Sample answers: 3–6 sentences synthesising the section's slides + transcript. Names mechanisms, addresses the relational part. Not a hint or restatement.
Rotating interactions (one per objective, rotate type):
- Predict-then-reveal — predict before unfolding next slide
- Fill-in-the-blank — cloze of key mechanism
- Diagram-to-complete — Mermaid with blank nodes
- Speed round — 60s free recall
- Spot-the-difference — two confusable concepts
- Real-world hook — connect to something outside the lecture
Match type to content. Don't repeat the same type consecutively.
Edge cases
| Situation | Action |
|---|
| Transcript only | Topic-based structure, no embeds |
| Slides only | Embed + expand, flag missing verbal context |
| Objective too long | Split across sessions, flag |
| 50+ slides | Multiple files + index |
| Poor transcript | Correct from slide text, flag with ⚠️ |
| Uncovered objective | Include with gap warning |
| Unknown abbreviation | Infer if standard + mark (inferred), else note in Gaps |
| More/less gamification | Adjust micro-check count / encouragement — not dynamic progress UI |
References
| File | When |
|---|
references/output-template.md | Always — before generating |
references/adhd-design.md | ADHD reasoning + citations |
references/evidence-base.md | Learning-science base |
references/transcriber-guide.md | If user needs transcript help |
