clear-explanation

Clear explanation (ChatGPT-style study sheets for Claude)

Many people find Claude's explanations dense: long paragraphs, small headings, undefined jargon, diagrams typed as text in a code block. ChatGPT's answers to the same question often read like clean study sheets: big headings, one idea per line, plenty of white space, real images, a short quiz at the end.

This skill makes Claude answer that way, without losing accuracy. The problem is density, not length: a long answer is fine if it breathes.

Built from:

• public skills that tackle the same problem: "explain-concepts" (thepexcel/agent-skills), "feynman-technique-explainer" and "education-tutor";
• a side-by-side comparison of Claude and ChatGPT answers to the same question, with a student's feedback after each version.

0. Before writing (silently)

1. Start from the user's own material. If their notes, course pages, slides or textbook are available (uploaded files, project documents, a linked folder), read the relevant part first. Reuse their words, their values and their teacher's angle. Say where it comes from: "In your course…", "Your teacher points out that…".
2. Check what you are not sure of. If a textbook says something different from their course, add one line: "Good to know: …".
3. Pick the mode.
   • Sheet mode: a precise question ("what is…", "why…", one mechanism). 150 to 400 words.
   • Lesson mode: a broad question ("explain everything", "start from the beginning", a chapter, several linked ideas). As long as needed, always airy.
4. One concept at a time. In lesson mode, one part per concept.
5. Bridge. Start from what they already know. If a prerequisite is missing, explain it in one line first.

1. Layout (both modes)

Headings: big, bold, clearly separated. The reader must see at a glance that a new part starts.

• Lesson mode:
  • each part gets a level-1 heading: # <emoji> <n°>. <title>;
  • sub-parts get level-2 headings (##);
  • never use ### for a part.
• Sheet mode: each block gets a level-2 heading: ## <emoji> <title>.
• Before each part or block: a blank line, a --- line, a blank line, then the heading.
• When it is a line of reasoning, the heading is a question ("Why is the inside negative?").

White space:

• One sentence per line, with a blank line between sentences. Without the blank line, the chat merges them into a paragraph.
• Never more than two sentences stuck together.

Lists and emphasis:

• Lists as one-line bullets; values and comparisons as tables.
• Bold only the key word, at most one per line.
• Give each concept a color dot (🟢 🟠 🟣 🔵), always the same one, and use the same color in the images. Three at most.

Markers, each on its own line:

• 📌 Key takeaway: one sentence, at the end of each part (lesson mode) or after "How it works" (sheet mode).
• ⚠️ Watch out: the classic misconception, at most one per part.
• ⭐ Key point of your course or Comes up in exams: only when their material shows it.

2. Sheet mode (fixed order)

• ## 💡 The idea: one sentence, 15 words at most, no jargon.
• ## 🪢 Picture it: <the comparison in 3 words>: pick the emoji that fits (🪢 🚪 🔋 🚰 ⚖️).
  • 2 to 4 bullets that show how it WORKS, not just what it looks like.
  • The last bullet says: "The comparison stops here: …".
• One visual (section 4), right after the comparison.
• ## ⚙️ How it works:
  • 3 to 5 numbered steps, one line and one fact each.
  • Put → between a cause and its effect.
  • End with 📌 Key takeaway.
• ## 🌍 A real example: a concrete, checkable case, written as a chain of →. If there is an excess, also give the shortage (too much / too little).
• ## 📖 The words: a table "Word | In plain words". Latin, Greek or foreign terms are always translated.
• ## ✅ Check yourself: one application question, then "Answer: …".

3. Lesson mode

1. Opening: one sentence announcing the 2 to 4 questions of the plan: "The goal: that you understand why…, why… and why…".
2. The chain of "why". Number the parts in the order of the reasoning: each part answers the question the previous one raises. Ask it out loud ("So why −70 and not −95?").
3. In each part:
   • the explanation, in short sentences;
   • one visual (section 4);
   • 📌 Key takeaway.
4. Both cases, always: excess and shortage (high and low, open and closed), each with its consequence, ideally side by side in a table.
5. "The other ideas in this chapter", if there are some: 2 to 4 lines each, with what the teacher expects (understand or calculate).
6. "The recap to memorize": 3 to 6 chains of arrows. Not a summary in sentences.
7. The mini-quiz (section 5).

4. Visuals: in the middle of the text, the user's own material first

Rule n°1 (blocking): every visual appears right under the sentence it illustrates. Never all grouped at the end.

• Show a visual only with a tool that displays it exactly where you call it: the chat's drawing tool (the Visualizer: show_widget, after its read_me) or image search.
• Never an attachment for a visual inside an explanation (file-sending tool, PNG or JPG file): the app places attachments wherever it wants, often all at the end, and the reader no longer knows when to look at them.
• Work order:
  1. Prepare all the visuals first, silently (figure copies, searches).
  2. Then write the answer piece by piece: the text of one part, then the call that displays its visual, then the text of the next part, and so on.
• If your environment only shows text sent through a messaging tool (for example SendUserMessage), send each piece of text with it; otherwise text written between two tool calls may be summarized and not shown.
• Right before the visual, one line announces it ("The diagram from your course:"). Right after it, one line 👀 What to look at: …

Take the first source that exists, in this order. Never modify an original file: work on a copy.

A. A diagram from the user's own study pages, copied as is (best choice). It is already right and familiar: do not redraw it.

• For an HTML page with SVG diagrams, use scripts/figure_to_chat.py: python3 figure_to_chat.py page.html lists the figures, python3 figure_to_chat.py page.html <n> fig.svg writes one as a standalone SVG with its colors inlined (usually 3 to 10 KB).
• Display it with show_widget, in HTML mode, inside a white card (it was made for a light background): <div style="background:#fff;border-radius:12px;padding:8px 4px">, the SVG copied exactly, then </div>.
• Only SVG figures. A photo or scanned figure cannot be displayed in place: use B, C or D.

B. A figure from their slides or textbook, redrawn and annotated.

1. Find the right slide or page by its text (python-pptx or pdftotext), convert to PDF (soffice --headless --convert-to pdf), render it (pdftoppm -r 110) and look at it.
2. Redraw it as a simplified SVG with show_widget (same elements, same layout), then annotate it: colored rounded frames around the important elements, numbered badges, the legend of the numbers right below, and a source line: "Redrawn from slide of your course".
3. Why redraw: an image file can only be shown as an attachment, so not in its place. The drawing shows up exactly where you call it.

C. A web image (image search tool, which displays in place), only if A and B do not exist, or as an extra.

• Pick the simplest one, in the user's language if possible.
• Below it, write 👀 What to look at: …, and translate any foreign labels.

D. A drawn diagram with the chat's drawing tool (the Visualizer: show_widget, after its read_me, "diagram" module). Use it for a mechanism no figure shows, or for the comparison picture (tug of war, door, scale).

• Show the mechanism through shapes, not a list of boxes.
• Use 2 or 3 colors, the same as the dots.
• Show strength or quantity through thickness or size.
• At most 6 labels of 5 words, plus a one-line caption.

How many:

• Sheet mode: one visual (A if it exists, otherwise D). No more than 3 tool calls to get it.
• Lesson mode: at least one visual per part.

Never:

• a diagram in a code block, in ASCII, or in hand-aligned characters;
• an image attachment inside an explanation;
• all the visuals grouped at the end.

5. Quiz

Sheet mode: the "Check yourself" block.

Lesson mode: an interactive 5-question multiple-choice quiz with the chat's quiz tool if it exists (otherwise 5 questions in text, answers at the bottom).

• One correct answer per question.
• At least 2 reasoning questions ("what happens if…"), not just definitions.
• Plausible traps: classic misconceptions and the "Watch out" lines of the answer.
• For each question: a hint and a short explanation.
• Only about what the answer explained.

6. Words

• Simple verbs (go in, go out, push, pull, open, close) rather than abstract nouns.
• No acronym or notation without translating it the first time.
• Jargon check: every word a 14-year-old would not know is either translated at its first use or removed.
• Concrete before abstract.
• Simplification check: if someone believed this version 100 %, would they get an exam question wrong? If so, add the exception in one line.
• At most one nuance, never stacked caveats.
• Never: "simply", "obviously", "just", "the answer is simple", "great question".
• Answer in the user's language, in a calm tone.

7. Review before sending (blocking)

1. The mode is chosen and respected.
2. Headings: # for lesson parts, ## for sheet blocks, each with its emoji, and a --- between blank lines before each one.
3. One sentence per line, separated by a blank line.
4. Every visual is displayed in the middle of the text, right under what it illustrates, by the drawing tool or image search. No attachment, no visual grouped at the end, no diagram in a code block. Sources follow the order A → B → C → D.
5. Both cases are covered whenever there is an excess or a shortage.
6. Every technical word and acronym is translated.
7. In lesson mode: a 📌 per part, an arrow recap, the quiz.
8. At most one question to the user, at the very end (the quiz does not count).

8. If the user does not understand (or gets it wrong)

• "I don't get it": do not make it longer. Ask in one line where it breaks ("part 2 or part 3?"), go back to that point, change the comparison, cut it smaller, then reconnect.
• Wrong answer: say what is right in their answer, name the error and why it seems logical, show wrong and right side by side in a table, then ask a new question of the same kind.
• "OK, I got it": ask them to restate the idea in one sentence, in their own words.

9. Example (sheet mode)

Question: "What is the resting membrane potential?"

💡 The idea

At rest, a nerve cell is negative inside, mainly because potassium leaks out.

---

🪢 Picture it: a tug of war

• 🟢 Potassium (K⁺) pulls toward −95 mV, with many players.
• 🟠 Sodium (Na⁺) pulls toward +67 mV, with very few players.
• The marker stops near the strongest team: −70 mV.
• The comparison stops here: the "players" are the open channels.

(Here, a visual: a figure from the user's course, or the tug of war drawn with the Visualizer.)

---

⚙️ How it works

1. Each ion has its equilibrium value: K⁺ at −95 mV, Na⁺ at +67 mV.
2. At rest, mostly K⁺ channels are open.
3. So K⁺ wins → the resting value is close to −95.
4. Neuron: −70 mV. Heart muscle: −90 mV.

📌 Key takeaway: the more easily an ion crosses, the harder it pulls the resting value toward its own.

---

✅ Check yourself

Suddenly, many Na⁺ channels open. Does the marker move toward −95 or +67?

Answer: toward +67, because the Na⁺ team suddenly has many players. That is the start of an action potential.