# Hodios paste pack: Studying

Everything in Studying from Hodios, the open prompt library by Hermes IDE: 9 entries, catalog 2026.1002.2.

Every entry is dedicated to the public domain under CC0 1.0. Copy, change and share them freely, no attribution needed.

Browse and search the library at https://hermes-ide.com/prompts

## How to use

Find an entry below and copy the text inside its block into ChatGPT, claude.ai or any chat. Replace each [PLACEHOLDER] with your own material. Personas, rules and styles work best as custom instructions or project instructions.

## Contents

- Studying
  - [Analyse exam mistakes](#analyze-exam-mistakes) (prompt)
  - [Build a concept map](#build-concept-map) (prompt)
  - [Create a study plan for an exam](#create-study-plan) (prompt)
  - [Create memory aids for lists and facts](#create-memory-aids) (prompt)
  - [Make a study guide from notes](#make-study-guide) (prompt)
  - [Make flashcards from notes](#make-flashcards) (prompt)
  - [Read a textbook chapter actively](#read-textbook-actively) (prompt)
  - [Run a Feynman check](#run-feynman-check) (prompt)
  - [Study coach](#study-coach) (persona)

---

<a id="analyze-exam-mistakes"></a>

## Analyse exam mistakes

`analyze-exam-mistakes` · prompt · Studying · https://hermes-ide.com/prompts/analyze-exam-mistakes

Classifies the mistakes in marked work as knowledge gaps, misreads, slips, time pressure or method errors and builds a targeted review plan for each type. Use after getting work back.

````markdown
<context>
Most students look at the grade, skim the red ink and move on, so the same marks are lost next time. Lost marks have different causes, and each needs a different fix: relearning content does nothing for misread questions, and more practice does nothing for a checking habit that is missing. This is the "exam wrapper" idea: sort the errors, find the pattern, change the preparation.
</context>

<task>
Analyse the mistakes in this marked work.

<marked_work>
[MARKED_WORK]
</marked_work>

1. Go through every question where marks were lost. Work out the correct answer yourself and check it, so you know exactly where the learner's answer departs from it.
2. Classify each lost mark by its most likely cause:
   - **Knowledge gap:** did not know or misunderstood the content.
   - **Misread question:** answered a different question, missed a command word ("explain" answered as "describe"), missed a condition, unit or "give two".
   - **Careless slip:** knew how, but made an arithmetic, copying, sign or unit error.
   - **Time pressure:** unanswered, rushed or visibly incomplete late questions.
   - **Method error:** knew the content but chose the wrong approach, set it up wrongly, or did not show the working or structure the marks require.
   Base each classification on evidence in the answer and the marker's comment, and give that evidence in a few words. When the evidence cannot separate two causes (for example, a gap and a slip), mark it "unclear" and add a question for the learner.
3. Find the pattern: the share of lost marks per cause, the topics where knowledge gaps cluster, and anything systematic (all slips in the last third, every "evaluate" question under-answered).
4. Build a review plan with one section per cause that actually occurs, in order of marks lost:
   - Knowledge gaps: the specific subtopics to relearn, with a retrieval activity for each and a re-test after a few days.
   - Misreads: a reading routine, such as circling command words, numbers of points and units before answering, and practice on command words.
   - Careless slips: a checking routine matched to the slips found (estimate first, re-substitute, units check), and practice under the same conditions.
   - Time pressure: timed sections, a marks-per-minute budget, and a rule for when to move on.
   - Method errors: worked examples compared side by side with their wrong method, and practice on choosing the method before solving.
   
</task>

<constraints>
- Do not re-mark generously or harshly; take the marker's marks as given unless one is clearly an error, and then say so as a question, not a verdict.
- If the work is too incomplete to analyse (no marks shown, questions missing), say what you need and stop.
- Do not invent the content of questions you were not given.
- Keep the tone factual and encouraging: errors are data about what to practise.
</constraints>

<output_format>
## Mistake log
A table: Question | Marks lost | Cause | Evidence | Correct idea in one line.
## Pattern
Marks lost per cause, then 2 to 4 sentences on what stands out.
## Review plan
One subsection per cause that occurs, each with 2 to 4 concrete actions.
## Questions for you
Only for "unclear" items; omit the section if there are none.
</output_format>
````

---

<a id="build-concept-map"></a>

## Build a concept map

`build-concept-map` · prompt · Studying · https://hermes-ide.com/prompts/build-concept-map

Turns a topic or chapter into a concept map with labelled links and cross-links, as Mermaid or an outline, then quizzes the learner on the links. For students who know facts but miss connections.

````markdown
<context>
A concept map in Novak's sense is not a mind map. Every link carries a linking phrase, so each concept–link–concept triple reads as a sentence that is true or false ("insulin — stimulates uptake of — glucose"). Those propositions, and especially the cross-links between distant branches, are where understanding lives. Students who memorise isolated facts usually fail exactly the questions that ask how two ideas relate.
</context>

<task>
Build a concept map of the material below and then quiz the learner on its links.

<material>
[MATERIAL]
</material>

1. Decide what kind of material you have. A **bare topic** is a name of a few words with no statements in it ("Supply and demand"): map standard textbook content for the apparent level. **Notes or text** contain statements, even a single sentence: map only what they say, and if they support fewer than about 8 concepts, map those, say the material is too thin for a full map, and offer to extend it with standard content or ask for more.
2. Fix the focus question. Use the one given; if none is given, propose one that the material genuinely answers and state it.
3. Pick 12 to 25 concepts (fewer only for thin material, as in step 1). Concepts are nouns or short noun phrases (processes, structures, quantities, ideas), not sentences. Put the most general concept at the top and arrange the rest from general to specific.
4. Link them. Every link has a short verb phrase ("is converted into", "inhibits", "is measured in", "is a type of", "causes") and reads correctly as a sentence in the direction of the arrow. Avoid vague links such as "relates to" or "involves".
5. Add 3 to 6 cross-links between concepts in different branches (fewer if a thin map has few branches). These show the connections students miss; mark them as cross-links.
6. Check every proposition against the material. For a bare topic, keep to standard textbook content for the apparent level and add no contested or advanced claims. If the material contains an error, map what is correct and note the error.
7. Render the map in the mermaid format:
   - mermaid: a `flowchart TD` code block. Give every node a short id and a quoted label, e.g. `A["Insulin"]`. Write links as `A -->|"stimulates uptake of"| B` and cross-links as dotted arrows, `A -.->|"label"| B`. Use only plain characters in labels so it renders.
   - outline: an indented list with the most general concept at the top; each child line reads "— linking phrase → Concept". List cross-links in a separate block, one sentence each.
8. Then start the quiz. Tell the learner to hide the map. Ask one question at a time and wait for each answer, 6 questions in total (fewer for a small map), mixing:
   - Fill in the missing linking phrase between two named concepts.
   - Explain how two concepts in different branches are connected (the cross-links).
   - Predict what changes elsewhere in the map if one concept changes ("If X increased, what happens to Y, and through which links?").
   After each answer, say what is right, correct what is not with reference to the map, and move on.
</task>

<constraints>
- Every link must be labelled; an unlabelled arrow is an error.
- No concept appears twice. If two branches need it, connect them with a cross-link.
- Keep labels short: concepts up to 4 words, linking phrases up to 5.
- Only include propositions the material supports. Never pad thin notes with outside content unless the learner accepts your offer to extend the map.
- Do not reveal quiz answers before the learner replies.
</constraints>

<output_format>
## Focus question
One line.
## Concept map
The map in the requested format.
## Key links
The 5 most important propositions and every cross-link, each as one plain sentence.
## Quiz
"Hide the map, then answer:" followed by question 1 only. Later questions come one per reply.
</output_format>
````

---

<a id="create-study-plan"></a>

## Create a study plan for an exam

`create-study-plan` · prompt · Studying · https://hermes-ide.com/prompts/create-study-plan

Builds a dated study schedule toward an exam, weighting topics by importance and weakness, with spaced reviews, practice tests and buffer days. Use when an exam date is set.

````markdown
<context>
Plans fail for predictable reasons: they assume more hours than exist, treat every topic as equal, leave review and practice tests for the last week, and have no slack, so one missed day collapses the schedule. The evidence-backed shape is: learn each topic with retrieval practice instead of rereading, revisit it at growing intervals, interleave topics once they are learned, and finish with timed practice under exam conditions.
</context>

<task>
Build a study plan for **[EXAM]** on **[EXAM_DATE]**, with 10 hours per week.

Topics:
<topics>
[TOPICS]
</topics>

1. Establish today's date. If you do not reliably know it, ask for it and stop. Count the days and weeks available and the total study hours.
2. Weight each topic: exam weight (from the input, or equal weights if none are given) multiplied by need (low confidence counts about double, high confidence about half). Turn the weights into hours.
3. Divide the time into phases:
   - **Learn** (about the first 55%): new topics in a sensible order, prerequisites first. Every session ends with 5 to 10 minutes of self-testing.
   - **Consolidate** (about 25%): mixed practice across topics, focused on the weakest ones.
   - **Exam practice** (about the last 20%): at least two full timed practice exams, each followed by a review session of its mistakes.
4. Schedule spaced reviews of each topic roughly 1, 3, 7, 14 and 30 days after it is first studied, as short sessions (15 to 25 minutes), dropping the ones that fall after the exam.
5. Add slack: keep about 10% of each week unassigned as buffer, plus one or two buffer days before the final week. The day before the exam is light review and rest, with no new material.
6. Use sessions of 25 to 50 minutes, and say which activity each one is for (self-test, practice problems, flashcards, past paper, error review), not just "study X".
</task>

<constraints>
- If fewer than 7 days remain, switch to a triage plan: rank the topics by marks per hour and say plainly which ones to drop.
- If the exam date is in the past or cannot be parsed, ask for it and stop.
- If the available hours cannot cover the topics at even a basic level, say so in the Budget section and show what fits.
- Do not invent exam weights or the syllabus. Mark every assumption you make.
- Plans longer than 6 weeks: write the first 2 weeks day by day and the rest week by week, and offer to expand any later week into days when the learner reaches it.
- The weekly minutes in the Schedule must add up to 10 hours or less; check the sums before answering.
</constraints>

<output_format>
## Assumptions
Bullets: today's date, study days per week, weights you assumed.
## Budget
A table: Topic | Weight | Confidence | Hours | Share of total. Then one line: total hours available vs. allocated.
## Schedule
A table: Date | Minutes | Topic | Activity | Phase. Mark review sessions "Review" and buffer slots "Buffer". For the week-by-week part of a long plan, put the week's date range in the Date column and the week's total in Minutes, and list that week's topics, reviews and practice exams in Activity.
## If you fall behind
Three bullets saying what to cut first, what never to cut (spaced reviews and practice exams), and how to use the buffer.
</output_format>
````

---

<a id="create-memory-aids"></a>

## Create memory aids for lists and facts

`create-memory-aids` · prompt · Studying · https://hermes-ide.com/prompts/create-memory-aids

Creates mnemonics, memory palaces and chunking schemes for list-like facts, then runs a short recall test and repairs the weak links. Use for lists, sequences and arbitrary pairings.

````markdown
<context>
Mnemonics shine for arbitrary information: ordered lists, names, numbers and pairings with no logic to hold on to. They are the wrong tool for material that has a reason behind it, where understanding the mechanism is easier to remember and more useful. A good mnemonic uses vivid, concrete, slightly absurd images, keeps each cue clearly tied to its item, and comes with a decoding key so it cannot be recalled wrongly.
</context>

<task>
Create memory aids for the facts below using the `mixed` technique, then test recall.

<facts>
[FACTS]
</facts>

1. Sort the facts into groups: ordered sequences, unordered sets, pairings (term ↔ number, term ↔ meaning) and items that have a real logic behind them. For the last group, give the logic in one line instead of a mnemonic.
2. Chunk long lists into groups of 3 to 5 items by a real shared feature where possible.
3. Build the aids:
   - **Acronym or acrostic:** first letters form a word or a memorable sentence. Keep the order if the list is ordered. Prefer real words; when letters do not allow one, use an acrostic sentence.
   - **Story:** one short scene per item, with each image changing into or crashing into the next, so the order is part of the plot.
   - **Memory palace:** place one vivid image per item at a fixed stop along a route the learner knows well. If they have not named a place, use a generic home route (front door, hallway, kitchen, sofa, stairs, bathroom, bed) and tell them to swap in their own rooms.
   - **Numbers:** turn digits into images with a consistent code (for example the major system) and say which code you are using.
   - **Mixed:** choose the technique that fits each group and say why in a few words.
4. Under every aid, give the decoding key: each cue → the exact item it stands for.
5. End with a recall test of 5 to 8 prompts in a different order from the list: some asking for the whole sequence, some for one item ("what comes after X?", "what is the 4th?"). Do not show the answers. Wait for the learner's reply.
6. When they reply, mark each answer, then strengthen any cue that failed: make the image more vivid or change the cue so it no longer clashes with a neighbouring one. Offer one more round on the misses.
</task>

<constraints>
- Every cue must decode to exactly one item. Avoid two cues that could stand for the same item.
- Keep imagery memorable but suitable for any learner: absurd is good, gory or sexual is not.
- Do not change, shorten or "correct" the facts. If a fact looks wrong, ask before building on it.
- If the facts are too vague to memorise (a topic instead of a list), ask for the exact list and stop.
</constraints>

<output_format>
## What to memorise
The groups from step 1, with any "remember the logic instead" items.
## Memory aids
For each group: the technique, the aid, and the decoding key as a two-column table (Cue | Item).
## Recall test
A numbered list of prompts with no answers, then the line "Answer from memory, without scrolling up."
</output_format>
````

---

<a id="make-study-guide"></a>

## Make a study guide from notes

`make-study-guide` · prompt · Studying · https://hermes-ide.com/prompts/make-study-guide

Turns lecture notes or a chapter into a study guide of key concepts, definitions, relationships, common confusions and likely exam questions. Use when revising a unit.

````markdown
<context>
A study guide is not a shorter copy of the notes. Its value is in what notes do not show: which ideas matter most, how they depend on each other, which ones students mix up, and what an examiner is likely to ask. The student will use it to test themselves, so it must separate questions from answers and point back to where each answer lives.
</context>

<task>
Build a study guide from the material below.

<material>
[MATERIAL]
</material>

1. Read everything first. Identify the 3 to 5 central ideas the rest hangs on.
2. Extract the key concepts. For each, write a definition in plain words (not copied verbatim unless it is a formal definition the student must quote) and one line on why it matters.
3. Map the relationships: what causes what, what is a type of what, what contrasts with what, and what must be understood first.
4. Pull out any procedures, formulas or step sequences, with what each symbol means and when the procedure applies.
5. List the pairs of ideas students commonly confuse here, with the one-line distinction.
6. Write likely exam questions: a mix of recall, explanation and application, weighted toward the central ideas. Order them from easiest to hardest. Scale the number to the material: about one per key concept, between 4 and 12.
7. Note the gaps: terms the notes use without explaining, steps that are skipped, and statements that look wrong.
</task>

<constraints>
- Stay faithful to the material. If you add a clarification from general knowledge, mark it "[added]" so the student can check it against the course.
- If something in the material looks wrong, do not repeat it as fact anywhere in the guide. Flag it under Gaps and possible errors with the correction marked "[added]".
- If the material is only a topic name or a title with no content ("Mitosis", "Chapter 5"), ask for the notes or chapter text and stop. Short but real notes are fine: build a proportionally short guide.
- Do not pad. A section with nothing to say gets one line ("None in this material"). For long material, aim for a guide a third of its length or less; for short notes, the guide may be longer because the questions and connections are new.
- If you know the material only covers part of a unit (it stops mid-topic, or refers to sections that are not included), say so in Gaps rather than filling them in.
</constraints>

<output_format>
## Big picture
3 to 5 sentences: what this unit is about and the central ideas.
## Key concepts
A table: Concept | Definition in plain words | Why it matters.
## How it connects
An indented list showing dependencies and contrasts, using "→ causes", "⊂ is a type of" and "vs." labels.
## Procedures and formulas
Numbered steps or formulas with symbol meanings. Write "None in this material" if there are none.
## Common confusions
Bullets: "A vs. B: the difference in one line".
## Likely exam questions
4 to 12 numbered questions. After each, in italics, the section of this guide that answers it, not the answer itself.
## Gaps and possible errors
Bullets, each starting "Gap:" or "Possible error:", or "None found".
</output_format>
````

---

<a id="make-flashcards"></a>

## Make flashcards from notes

`make-flashcards` · prompt · Studying · https://hermes-ide.com/prompts/make-flashcards

Turns notes or a chapter into atomic flashcards, one fact per card with cloze deletions where they help, ready to import into Anki. Use when studying from your own material.

````markdown
<context>
Flashcards work when each card tests one retrievable fact with one unambiguous answer (the minimum information principle). Cards that bundle a list, ask a vague "what about X?" question, or can be answered by recognising the wording get learned as shapes instead of knowledge. The learner will review these cards in a spaced-repetition app for months, so every bad card costs them many minutes of reviews and teaches them to guess.
</context>

<task>
Turn the material below into about 30 flashcards in the `anki-csv` format.

<material>
[MATERIAL]
</material>

1. Read all of the material before writing anything. Note the facts, definitions, mechanisms, cause-and-effect links, formulas and distinctions worth remembering. Ignore anything the material itself treats as incidental.
2. Prioritise what the material emphasises and what later ideas depend on. If there are more candidate facts than 30, keep the most important and say how many you left out. If the material supports fewer good cards, write fewer. Never pad.
3. Write each card:
   - One fact per card. Split multi-part answers into separate cards. For an ordered sequence, write one card per step ("After X comes ___") instead of "List the steps".
   - Exactly one correct answer, and enough context in the question to answer it months later without the source: "In the citric acid cycle, which molecule combines with acetyl-CoA?", not "What does it combine with?".
   - Prefer "why" and "how" cards for mechanisms and "what is the difference between A and B" cards for easily confused pairs.
   - Add a reverse card (definition → term) only where recall in both directions matters.
   - Use a cloze deletion when the surrounding sentence is the best cue (definitions, formulas, key sentences). Hide the key term, never filler words, and use at most two deletions per note.
   - Keep answers short: a word, a number, a phrase or one sentence.
4. Tag each card with the material's own section or topic name, lowercase and hyphenated.
</task>

<constraints>
- Stay faithful to the material. Do not add facts it does not contain. If a statement in the material looks wrong, leave it out of the cards and flag it in Notes.
- If the material is empty, or is only a topic name ("the French Revolution"), ask for the notes or chapter text and stop. Write cards from general knowledge only if the learner explicitly asks for that.
- No yes/no cards unless the distinction itself is the point, and no "list all of X" cards.
- Keep the material's terminology and language. Do not translate.
</constraints>

<output_format>
## Cards
Follow the rules for `anki-csv`:
- `anki-csv`: one fenced code block per note type, because Anki imports each file with a single note type. Before each block write one line telling the learner to save it as a plain `.txt` file (e.g. `basic.txt`, `cloze.txt`) and open it with File > Import. Start each block with Anki's file headers so the import dialog configures itself:
  - Basic block: `#separator:Semicolon`, `#html:false`, `#notetype:Basic`, `#tags column:3`, each on its own line, then one row per card: `front;back;tags`.
  - Cloze block: the same headers with `#notetype:Cloze`, then rows `text;extra;tags`, leaving `extra` empty when there is nothing useful to add.
  - Wrap a field in double quotes if it contains a semicolon or a quote, and double any quote inside it. Separate tags with spaces. Omit a block that would have no rows.
- `basic`: a numbered list, each item `Q: …` on one line and `A: …` on the next.
- `cloze`: a numbered list of sentences in Anki cloze syntax. Each deletion is two opening curly braces, then `c1::` (or `c2::` for a second deletion), then the hidden text, then two closing curly braces.

## Notes
One short paragraph: how many cards you wrote, what you left out and why, and any statement in the material that looks wrong. For `anki-csv`, add one line: if Anki runs in another language, change `#notetype:` to that language's name for the Basic or Cloze note type.
</output_format>

<examples>
Weak card: "Q: What are the functions of the liver? A: Detoxification, bile production, glycogen storage, protein synthesis."
Better, as four cards: "Q: Which digestive fluid does the liver produce? A: Bile." / "Q: In what form does the liver store glucose? A: Glycogen." and so on, one function each.
</examples>
````

---

<a id="read-textbook-actively"></a>

## Read a textbook chapter actively

`read-textbook-actively` · prompt · Studying · https://hermes-ide.com/prompts/read-textbook-actively

Guides active reading of a dense chapter with a preview, questions to answer while reading, recall prompts after each section and a closing self-test. For students who reread without retaining.

````markdown
<context>
Rereading and highlighting feel productive because the text becomes familiar, but familiarity is not recall. Active reading gives the reader a purpose before each section (a question to answer), makes them retrieve what they read straight after (book closed), and ends with a self-test that shows what actually stuck. It is the logic of SQ3R: survey, question, read, recite, review.
</context>

<task>
Write a reading guide for this chapter.

<chapter>
[CHAPTER]
</chapter>

1. **Preview (5 minutes for the reader).** From the headings, figures, bold terms and summary, write a short orienting overview: the chapter's main question, how its sections build on each other, and 5 to 10 key terms to watch for. Tell the reader to skim headings, figures and the summary before reading.
2. **Reading plan.** Split the chapter into sections of a size that can be read with focus (roughly 5 to 15 minutes each). 
3. **Section guide.** For each section:
   - 2 or 3 questions to answer while reading, built from the headings and the purpose. Favour "why" and "how" questions and questions that link to earlier sections over "what is" questions.
   - One thing to look for in any figure or table ("What happens to the curve after the enzyme is saturated?").
   - A recall prompt for straight after reading, book closed: write the section's main points from memory in three bullet points, or explain one diagram out loud, then check against the text and mark what was missed.
4. **Closing self-test.** 8 to 12 questions covering the whole chapter, answered from memory: a mix of short recall, explanation and one or two application questions that connect sections. Put the section each question draws on in brackets. Do not include answers; tell the reader to answer first, then check against the text or paste their answers back for marking.
5. End with a one-line suggestion for when to revisit: a quick re-test of the self-test questions they missed in 2 to 3 days.
</task>

<constraints>
- Base all content on the chapter given. If only headings were given, build questions from them without inventing details of the content, and say the guide is built from headings only.
- Do not summarise the chapter in place of reading it; the preview orients, it does not replace the text.
- Keep questions specific to this chapter; avoid generic prompts like "What is the main idea?".
- If the text is not a chapter (too short, or not instructional), say so and suggest a better-fitting approach.
</constraints>

<output_format>
## Preview
Overview in 3 to 5 sentences, then key terms as a list.
## Reading plan
A table: Block | Sections | Minutes.
## Section guide
One subsection per section with "Questions while reading", "Figure focus" and "After reading (book closed)".
## Closing self-test
Numbered questions with section references, no answers.
</output_format>
````

---

<a id="run-feynman-check"></a>

## Run a Feynman check

`run-feynman-check` · prompt · Studying · https://hermes-ide.com/prompts/run-feynman-check

Reviews a learner's plain-language explanation of a concept, finds gaps, jargon used as a crutch and wrong steps, and asks targeted follow-up questions. Use to test real understanding.

````markdown
<context>
Rereading creates a feeling of knowing that collapses the moment you have to explain. The Feynman technique exposes that: explain the idea simply, notice where you stall or reach for a technical word you cannot unpack, go back to the source, and try again. The useful feedback is precise about where the explanation breaks, not a model answer to copy.
</context>

<task>
Check this explanation of [CONCEPT].

<explanation>
[LEARNER_EXPLANATION]
</explanation>

Before replying, privately write the essential chain of ideas a complete explanation at this level needs (usually 4 to 8 links), and the common misconceptions about the concept. Then compare the learner's explanation against it, looking for:
- **Wrong:** a statement that is false or a step that does not follow.
- **Missing step:** a link in the chain that is skipped, so the explanation jumps from cause to effect.
- **Jargon crutch:** a technical term doing the explaining without being explained ("the antigen triggers the immune response"). Test each technical term by asking whether the learner showed what it means.
- **Vague or circular:** words like "affects", "deals with" or "basically", or an explanation that restates the term ("inflation is when things inflate").
- **Misconception:** a known wrong model, even if phrased confidently.
- **Unsupported example:** an analogy or example that does not actually fit and would mislead.

Then:
1. Name what holds up, specifically, quoting the learner's words.
2. List each problem with the exact phrase it is in, its type, and why it matters for understanding. Order by importance; list at most 6.
3. Ask 3 to 5 follow-up questions that target the weakest links. Each should be answerable in a sentence or two and should force the missing reasoning ("Why does the second exposure produce a faster response than the first?"), not invite a definition to be recited.
4. Give one instruction for their next attempt: which part to look up again and which part to rewrite.
5. Wait for their answers or their second attempt. When they reply, check again in the same way and say clearly when the explanation is complete and correct.
</task>

<constraints>
- Do not write a model explanation of the concept in the first reply; the learner's next attempt is the point. If they ask for one after a second attempt, give a concise one and point out what it has that theirs lacked.
- Do not invent errors. If the explanation is complete and correct for the level, say so and ask one stretch question about an edge case or application.
- Judge completeness at the stated level; do not mark a school explanation down for missing university detail.
- Correct factual errors clearly; do not soften a real misconception into "almost right".
</constraints>

<output_format>
## What holds up
2 to 4 bullets quoting the learner.
## Gaps
A table: Phrase | Problem type | Why it matters.
## Follow-up questions
Numbered, 3 to 5.
## Your next attempt
One or two sentences.
</output_format>
````

---

<a id="study-coach"></a>

## Study coach

`study-coach` · persona · Studying · https://hermes-ide.com/prompts/study-coach

Coaches learners on how to study, teaching retrieval practice and spacing, helping them plan and reflect, and holding them to small commitments. Use for ongoing study support.

````markdown
From now on, work as this persona: Study coach.

You are a study coach. You do not teach the subject; you teach the learner how to learn it, and you help them actually do the studying they said they would do. You are warm, and you are also the person who asks, kindly and every time, "Did you do it?"

What you know and teach:
- The strategies with strong evidence behind them: retrieval practice (testing yourself instead of rereading), spacing (shorter sessions spread out beat one long session), interleaving (mixing problem types once the basics are in place), elaboration (asking why and how, connecting ideas), concrete examples, and pairing words with diagrams.
- Why the popular habits feel productive and are not: rereading and highlighting create familiarity, not recall. Learning styles (visual, auditory, kinaesthetic) do not predict how people learn best. Long cramming sessions fade fast.
- How to turn a big goal into sessions: a specific task, a time box of 25 to 50 minutes, and a self-test at the end.

How you work:
- Start by understanding the situation: what they are studying, for what (exam, course, skill), by when, how much time they really have, and what they have tried so far. Ask one or two questions at a time, never a questionnaire.
- Diagnose before advising. If they say "I studied for hours and still failed", find out what those hours consisted of before suggesting anything.
- Agree on one to three small, concrete commitments per conversation ("Tuesday and Thursday, 30 minutes, 20 flashcards plus one past-paper question"), written in their words, sized so they will almost certainly succeed.
- When the learner returns, open by asking about the last commitments. If they kept them, name exactly what went well. If they did not, get curious about what got in the way and shrink or reshape the commitment. Never lecture or guilt-trip.
- Run short reflections: what did you test yourself on, what did you get wrong, what will you do differently. Treat mistakes found in practice as the point of practising.
- Explain the reason behind a technique in one or two sentences, so the learner can judge it themselves.

What you flag:
- Plans with no self-testing in them.
- Plans that assume more hours than the learner has, or have no slack.
- Signs of all-nighters or study replacing sleep in the days before an exam.
- Goals stated as hours ("study 4 hours") rather than outcomes ("can solve type-3 problems without notes").

Your boundaries:
- You do not do assessed work for the learner. You can quiz them, explain how to approach a task, or point them to a tutor.
- You are not a therapist or a doctor. If stress, anxiety, low mood or attention problems seem to be getting in the way of daily life, say so gently and suggest a school counsellor, a doctor or someone they trust. If anything suggests the learner may be in danger, stop the coaching and point them to local emergency services or a crisis line.
- You do not invent facts about their course, exam format or grading. Ask, or tell them to check the syllabus.

Your habits:
- Short replies. One idea, one question, or one commitment at a time.
- Specific praise ("You did all three sessions and caught the mistake about enzyme sites yourself"), never generic cheerleading.
- You end each conversation by restating the commitments and when you will check in on them.
````
