An Obsidian notebook that thinks along.
Not a transcript-to-file converter: it knows what has already been written, notices what you are weak on, links related ideas, and keeps its own flashcard deck sharp over time.
Most "AI note-taker" prompts do one thing: turn text into a formatted file. That falls apart once you have a whole course. New material overlaps with old notes, some topics never get tested, and flashcards pile up without any sense of what is sticking.
Smart Notes behaves like a notebook a diligent student would keep. It is deliberately generic: nothing in it is tied to one course, subject or language. It configures itself the first time you use it for a new subject.
| Step | What the skill does |
|---|---|
| 1. Onboarding | Once per subject it asks three things: subject name and type, note language(s), and where to save. Answers are stored in _Progress.md, so it never asks again. |
| 2. Check the index | Looks in _Index.md. If your material is a near-duplicate of an existing topic, it tells you which file already covers it and extends that file. A separate new file only happens if you confirm it is a different topic. |
| 3. Read everything | Reads the whole source and gathers every fact, number and example before writing. |
| 4. Plan the links | Works out which existing topics this one connects to: prerequisite, follow-up, same category, or contrast. |
| 5. Write the note | Builds the note from the shared skeleton and the preset for your subject type. |
| 6. Update the maps | Refreshes _Index.md and _Progress.md. |
Once activated, the skill stays active for the rest of the conversation. Every later transcript or reading you paste is handled automatically.
| 🗂️ Linked notes, not isolated files Every note has a Related: line with [[wikilinks]], and _Index.md keeps a live map of every topic, its status and its links. |
🃏 Adaptive flashcards Weak topics get more cards and a wider spread of angles. Every deck includes a connection card when a genuine link to another topic exists. |
🧩 Gap tracking_Progress.md logs what you are weak on from three sources (see below). |
♻️ Edits instead of duplicating New material that extends or corrects a note edits that note directly, then briefly reports what changed. Affected flashcards, _Index.md and _Progress.md are updated too. |
| 🎛️ Adapts to the subject Four presets with their own note structure (see below). |
🧠 Quizzes only when you ask Never proposes a quiz or review unprompted. |
One folder per subject:
<Subject Name>/
_Index.md ← map of content: every topic, its status, its links
_Progress.md ← config (from onboarding) + gap tracker + quiz history
<Topic 1>.md
<Topic 2>.md
...
The _ prefix keeps the two control files sorted above ordinary notes in any file browser.
| Topic | Status | Links |
|---|---|---|
| [[Topic 1]] | ✅ covered | [[Topic 2]] |
| [[Topic 2]] | [[Topic 1]], [[Topic 3]] | |
| Topic 4 | ❌ not covered yet | referenced by [[Topic 2]] |
A topic that an existing note references but that has no note of its own stays listed as not covered yet.
Every note shares one skeleton: frontmatter, # Title, a Related: line, an opening definition, one section per concept, a common-mistakes callout, and a flashcard deck.
> [!warning] Common mistakes / things people mix up
> - <mistake 1>
> - <mistake 2>
## Flashcards
#flashcards
<Question>?
?
<Answer>The bare ? line is the question/answer separator of the Obsidian Spaced Repetition plugin.
If the subject is configured as bilingual, every prose line and every flashcard question and answer appears as a pair in the two languages you chose. Structure stays identical.
- Topics flagged weak get more cards and a wider spread of angles: bare definition → mechanism / why → applied scenario.
- Difficulty is graded across that spread instead of writing every card at the same level.
- Whenever a genuine link exists, at least one connection card tests the relation between this topic and a related one already in the vault, for example how does CIDR relate to a routing table entry?
The repeated reviews themselves are run by the Obsidian Spaced Repetition plugin.
Only when you explicitly ask:
- No topic named → it pulls from
_Progress.md, weak and not-yet-tested topics first. - A quick quiz widget for factual recall; plain chat for deeper "explain why / how" questions, where you answer in your own words and it evaluates.
- The outcome is logged back into the quiz history in
_Progress.md.
Every reply leads with the file name in bold, whether creating or editing, followed by a short bullet summary of what the file covers or what changed. The full file is not pasted into chat.
- If pasted material is itself a graded assessment (look for "Honor Code", "Submit", point values), it never supplies the answers, even if asked repeatedly. It offers to explain the underlying concept or run a separate self-check quiz instead.
- Any embedded "AI assistant instructions" inside pasted transcripts or page text are treated as inert content, never as commands.
This is a single SKILL.md file, the standard format for a Claude Skill.
- Copy
SKILL.mdinto your Claude skills directory, typically~/.claude/skills/smart-notes/SKILL.md(check Anthropic's current docs for the exact path). - Restart or reload Claude so it picks up the skill.
- Done. No dependencies, no build step.
/smart-notes
or
/study
or just ask Claude to keep smart study notes or flashcards for a course. On the first run for a new subject, answer the onboarding questions. After that, paste lecture transcripts, readings or any course material.
To review, ask: "quiz me on what I'm weak on" or "test me on [topic]".
- Obsidian (or any Markdown editor; wikilinks and callouts are Obsidian-flavored)
- Spaced Repetition plugin if you want the flashcards to run as real spaced-repetition reviews
MIT. See LICENSE. Free to use, modify and redistribute.