agentleFS
baney75baney75/myelinate/skills/memory-hooks/SKILL.md

memory-hooks

Skill4 starsChanged 2 days agoMITView on GitHub

Use when someone asks for a mnemonic or memory trick for a list, sequence, name, or vocabulary. Understanding a mechanism belongs to tutor or teach.

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Meaning before mnemonic: If the content has a mechanism, teach the mechanism first.

    line 16
  2. Distinctive:

    Every hook must encode true facts: Check each letter, image and word against the source before giving it.

    line 17
  3. Distinctive:

    Ask: *Could the learner reconstruct this from a principle?*

    line 24
  4. Distinctive:

    Run this in order. In chat, one step per turn.

    line 59
  5. Distinctive:

    Gate: Use §1. If the content is understandable, say so and teach it: "This one has a reason.

    line 61
  6. Distinctive:

    Card it: Put the target fact on the card, not the mnemonic.

    line 71
  7. Distinctive:

    Fade: When the learner answers fast without running the hook (twice, across a delay), drop the hint line.

    line 72
  8. Distinctive:

    Retire or repair: If the hook causes errors (letter maps to the wrong term, image gets confused), replace it, add a contrast card, and say …

    line 73

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Memory Hooks

A memory hook is a deliberately built retrieval cue for material that has little internal logic: names, codes, orders, lists, foreign words, drug stems. Hooks are scaffolding. The goal is fluent retrieval of the real fact, after which the hook fades. A hook that replaces understanding is a liability: it produces confident recall of a label with nothing behind it.

Paths like references/…, templates/… and examples/… are relative to this skill's folder, which carries its own copy of every file it names. Other Myelinate skills are siblings (../tutor/SKILL.md) and are installed as their own skills too. If a file is missing from your copy, fetch it from https://raw.githubusercontent.com/baney75/myelinate/main/ plus its path in the repository (references/…, skills/<name>/SKILL.md); if you can't fetch, work from this file and the router's core rules.

Evidence tags: Strong (replicated, meta-analytic or classroom-relevant), Moderate (consistent lab evidence, limited transfer or delay data), Weak (thin, mixed, or mostly short-delay lab lists), Craft (expert practice, not tested as such).

0. Rules that override everything else

  1. Meaning before mnemonic. If the content has a mechanism, teach the mechanism first. Hook only the arbitrary residue.
  2. Every hook must encode true facts. Check each letter, image and word against the source before giving it. A wrong hook is harder to remove than a wrong fact.
  3. Hooks get tested, not admired. A hook counts only after the learner retrieves the target from it, unaided, after a delay.
  4. Learner builds or co-builds where possible. Offer a draft; invite them to replace any part that does not feel like theirs.
  5. No crude, sexual, demeaning or slur-based mnemonics: even "classic" ones. Offer a clean version that works as well.

1. Decide whether a hook is warranted

Ask: Could the learner reconstruct this from a principle?

Material shapeHook?What to do instead or as well
Arbitrary names, codes, labels (cranial nerve names, one-letter amino acid codes, SI prefixes)YesHook + retrieval cards
Fixed sequences with no causal order (book order, carpal bones, taxonomic ranks)YesAcrostic, chunking, route
Sequences with causal logic (Krebs cycle, action potential, a mechanism)PartialTeach the logic; hook only the names; test the logic separately
Foreign or technical vocabularyYes, brieflyKeyword or etymology hook, then many spaced retrievals and use in context
Concepts, mechanisms, derivations, arguments (why competitive inhibition leaves Vmax unchanged, why SN2 inverts)NoExplanation, worked examples, self-explanation, practice problems
Procedures and problem-solving (ICE tables, free-body diagrams)RarelyA step checklist is fine; a cute acronym for a procedure the learner cannot yet perform is not

Signals of a crutch: the learner can recite the mnemonic but cannot answer "why"; they must run the whole acrostic to get item 7; they have stacked a mnemonic onto another mnemonic; or the exam asks for application and they only have the list.

2. What the evidence says, by hook type

TechniqueWhat it isEvidenceWhat the evidence supports, and its limits
Keyword methodLink a foreign or unfamiliar word to a similar-sounding familiar word, then make an interacting image with the meaning (Spanish pato = duck → a duck sitting in a pot)Moderate for immediate recall; Weak for long-termAtkinson & Raugh (1975, JEP: Human Learning and Memory 1:126 to 133) found large gains for Russian vocabulary. Wang, Thomas & Ouellette (1992, J Educ Psychol) found faster forgetting with keywords than with rote rehearsal across delays; Wang & Thomas (1995, J Educ Psychol 87:468 to 475) found keyword recall worse than a semantic-context strategy after two days, with the gap much reduced by more review. Dunlosky et al. (2013, Psychol Sci Public Interest 14:4 to 58) rated the keyword mnemonic low utility overall. Use it to get a word in; use spaced retrieval to keep it.
Method of loci (memory palace)Place vivid images of items along a familiar route; walk the route to recallModerate-Strong for serial recall of lists; low-quality trial baseTwomey & Kroneisen (2021, QJEP 74:1317 to 1326): meta-analysis of 13 studies, g = 0.65, with high risk of bias in most trials. Ondřej (2025, Br J Psychol, doi:10.1111/bjop.12799): d = 0.88 vs rehearsal for immediate serial recall in adults; evidence graded low to very low quality. Dresler et al. (2017, Neuron 93:1227 to 1235): ~6 weeks of 30-min daily loci training in novices (n = 17 per group) raised recall of 72-word lists from roughly 26 to 30 to ~62 words, gains persisted at 4 months, and brain connectivity shifted toward memory-athlete patterns; n-back and no-training controls did not show the same gains. Best for ordered lists the learner must produce in sequence. Training cost is real; it does not make understanding.
First-letter acronyms and acrosticsInitial letters as a cue (OIL RIG; "Some Say Marry Money…")WeakLab results are mixed (e.g., Carlson, Zimmer & Glover 1981, J Gen Psychol 104:287 to 292, titled "First-letter mnemonics: DAM (Don't Aid Memory)"; Morris & Cook 1978, Br J Educ Psychol 48:22 to 28, on when they help). They help mainly with order and completeness of items the learner already knows; they do not supply the item itself. A letter that could be three different terms is a weak cue.
Peg systemPre-memorized number-image pairs (1 = bun, 2 = shoe…); attach each item to its pegWeak-Moderate (older list-learning lab studies; not re-verified here)Useful when position-by-number matters (cranial nerve V, item 7). Requires learning the pegs first.
ChunkingGroup items into meaningful unitsStrong as a principleWorking memory holds a few chunks, roughly 4 in Cowan (2001, Behav Brain Sci 24:87 to 114). This revises Miller's (1956) "seven plus or minus two". Chunk by meaning when possible (12 prophets → 4 groups of 3).
Elaborative interrogation / elaborationAsk "why would this be true?" and connect to known factsModeratePressley et al. (1988, J Educ Psychol 80:268 to 278) improved learning of "confusing facts"; Dunlosky et al. (2013) rated it moderate utility. This is the meaning-based alternative to a mnemonic, and often the better one.
Self-referenceRelate the item to yourselfModerateSymons & Johnson (1997, Psychol Bull 121:371 to 394) meta-analysis: self-referenced encoding beat semantic encoding on average. Use it in hooks ("the muscle I use to…"), not as a personal-data request.
Concreteness, imagery, dual codingPair words with concrete images; use both verbal and visual codesModerateClark & Paivio (1991, Educ Psychol Rev 3:149 to 210) review the dual coding account. Concrete, interacting images beat abstract words or separate images. Dunlosky et al. rated imagery for text learning low utility. Imagery helps imageable items, not dense prose.
BizarrenessMake images strangeWeak / mixedMcDaniel & Einstein (1986, JEP:LMC 12:54 to 65): bizarre images helped only when mixed with common ones in the same list. This was a distinctiveness effect, not a general one. Prefer interacting and distinctive over merely weird.
Narrative chainingLink items into a storyModerate (lab)Bower & Clark (1969, Psychonomic Science 14:181 to 182) found large serial-recall gains for lists woven into stories. Good for 6 to 15 ordered items.
Songs and rhythmSet text to a melodyWeak-Moderate; conditionalWallace (1994, JEP:LMC 20:1471 to 1485) found better verbatim text recall when verses repeated a simple, easily learned melody. Racette & Peretz (2007, Mem Cognit 35) found fewer words recalled when singing than speaking for unfamiliar songs. Ludke, Ferreira & Overy (2014, Mem Cognit 42:41 to 52) found singing helped adults learn Hungarian phrases. Use only a melody the learner already knows; songs fit verbatim lists, not concepts.
Generation (learner-made hooks)The learner produces the cueModerate for generation in general; mixed for mnemonics specificallyBertsch et al. (2007, Mem Cognit 35:201 to 210): across 86 studies, generating beat reading by d ≈ 0.40. For mnemonics, results vary: Ley & Locascio (1977) found subject-generated cues beat experimenter-supplied cues; Lieberman et al. (1968) found no difference. Supplied keywords can win when learners make poor ones. Default: co-create. Draft a good hook; let the learner adapt it.
Commercial visual-story mnemonics (Sketchy, Pixorize, Picmonic)Illustrated scenes where each object encodes a factWeak (limited independent evidence)No peer-reviewed outcome study of Sketchy or Pixorize was found in a Europe PMC search (2026-10-06). Picmonic has a small single-center randomized pilot (Yang et al. 2014, Adv Med Educ Pract 5:125 to 132) and a non-randomized cohort comparison (Hoyt et al. 2025, JMIR Med Educ 11:e55206), both positive. Widely used for pharmacology and microbiology; useful as an image source. Not a replacement for mechanism learning or question practice.

Interference (Moderate-Strong as a principle): similar hooks compete. Two acrostics starting with "S" for two different nerve lists, or two "duck" keywords in one language, will blur. Wang & Thomas attribute part of keyword forgetting to competition from existing associations. Fix: one hook per confusable set, a contrast card for the pair, and distinct images.

3. Procedure

Run this in order. In chat, one step per turn; the learner does steps 3, 4 and 6.

  1. Gate. Use §1. If the content is understandable, say so and teach it: "This one has a reason. Let's get the reason, and you won't need a trick." Hook only what is left.
  2. Pick the hook by shape.
    • Unordered set of 3 to 7 items → acronym, or one interacting scene.
    • Ordered list of 5 to 15 → acrostic sentence, story chain, or memory palace (palace if they will reuse the list for years).
    • Numbered positions → peg, or a number embedded in the hook ("1973").
    • Word ↔ meaning (foreign, Latin, drug stems) → etymology first (real roots are meaning-based and transfer); keyword image only if no usable root.
    • Visual-spatial layout (carpal bones, brachial plexus) → draw it; hook the reading order.
    • Large related sets (bug-drug tables) → one scene per cluster, or a commercial visual set if they already use one.
  3. Build it right. Concrete, interacting (the items do something to each other), distinctive, short, and tied to meaning where possible: the hook should point at why, not just what. Learner-made or co-made. No more than one layer: a mnemonic for the mnemonic is a stop sign.
  4. Test it. Hide the hook. Ask for the target from the cue, then the target without the cue. Then a delayed check later in the session, mixed with other items.
  5. Card it. Put the target fact on the card, not the mnemonic. Optionally put the hook on the back as a hint line. Separate cards for meaning ("why does ↑CO₂ shift the curve right?"). See flashcards for card rules.
  6. Fade. When the learner answers fast without running the hook (twice, across a delay), drop the hint line. Keep the hook only for rare items or exam-day recovery.
  7. Retire or repair. If the hook causes errors (letter maps to the wrong term, image gets confused), replace it, add a contrast card, and say why.

Accuracy check before delivering any hook: list every letter/image and its target; verify each against the course source or a reference; check spelling of the real terms; check the hook does not encode a superseded or contested fact without saying so.

4. Worked examples

Each is accurate as written; each names what the hook does not teach. Use examples that fit the learner's subject and level; a child learning map directions and a professional reviewing anatomy need different cues.

1. Compass directions (elementary map reading). "Never Eat Soggy Waffles" cues north, east, south, west clockwise from the top of a north-up map. It does not teach how to orient a map or what to do when north is not at the top; practice with the map's north arrow.

2. Treble staff lines (beginning music). "Every Good Boy Deserves Fudge" cues E, G, B, D, F from the bottom line upward. It does not teach pitch, rhythm or bass clef; point to each line on a staff before relying on the phrase.

3. Cranial nerve names, in order (anatomy). "Oh Oh Oh To Touch And Feel Very Good Velvet, Ah Heaven" → olfactory, optic, oculomotor, trochlear, trigeminal, abducens, facial, vestibulocochlear, glossopharyngeal, vagus, accessory, hypoglossal. Weakness: three O's and two T's. Pair it with a peg or position check ("V is trigeminal, VI is abducens").

4. Cranial nerve modality. "Some Say Marry Money But My Brother Says Big Brains Matter More" → I S, II S, III M, IV M, V B, VI M, VII B, VIII S, IX B, X B, XI M, XII M (sensory/motor/both). Add a number hook: parasympathetic fibers ride on 10, 9, 7, 3: "1973." Not taught: what each nerve does; teach that from function and lesions.

5. Right shift of the O₂-hemoglobin curve (physiology). "CADET, face Right": CO₂ ↑, Acid (↑H⁺, ↓pH), DPG (2,3-BPG) ↑, Exercise, Temperature ↑ → lower affinity, more O₂ unloading. Meaning layer (teach first): every item is a signature of hard-working tissue, so the shift delivers O₂ where it is needed. "Exercise" is a context, not a molecule; it works through the other four. If the learner can derive the list from "what does active muscle look like?", drop the acronym.

6. Amino acid one-letter codes (biochemistry). Eleven match their first letter (A, C, G, H, I, L, M, P, S, T, V). Hook the other nine: R = "aRginine" (pirate "aRRR"), K = lysine (K sits next to L), D = "asparDic" acid, E = "glutEmate", N = "asparagiNe", Q = "Q-tamine" (sounds like "cute-amine") for glutamine, F = "Fenylalanine", Y = "tYrosine", W = tryptophan (the double ring of indole → "double-u"). Then card side-chain properties separately by structure, not by hook.

7. Essential amino acids, a hook that needs a footnote. The common "PVT TIM HALL" (Phe, Val, Thr, Trp, Ile, Met, His, Arg, Leu, Lys) lists ten. The standard adult list is nine; arginine is conditionally essential. Teach the nine, keep the hook only if the learner annotates "R = conditional." This is the pattern for every inherited mnemonic: audit before reuse.

8. Krebs cycle intermediates (biochemistry). "Citrate Is Krebs' Starting Substrate For Making Oxaloacetate" → citrate, isocitrate, α-ketoglutarate, succinyl-CoA, succinate, fumarate, malate, oxaloacetate. Hook = names and order only. The logic is taught and tested separately: CO₂ leaves at the two decarboxylations (isocitrate → α-KG, α-KG → succinyl-CoA), NADH at those two plus malate → OAA, FADH₂ at succinate → fumarate, GTP/ATP at succinyl-CoA → succinate. A memory-palace version (eight stations in the learner's kitchen) suits someone who will need it for years.

9. Carpal bones (anatomy), clean replacement. "So Long To Pinky, Here Comes The Thumb" → proximal row lateral to medial: scaphoid, lunate, triquetrum, pisiform; distal row medial to lateral: hamate, capitate, trapezoid, trapezium (anatomical position). Trapezium sits by the thumb. Draw the hand and trace the U-shaped path; the drawing does most of the work.

10. Formal charge and hydrogen bonding (general/organic chemistry). Formal charge = valence electrons − dots − sticks (lone-pair electrons, and one per bond). O in H₃O⁺: 6 − 2 − 3 = +1. N in NH₄⁺: 5 − 0 − 4 = +1. Hydrogen-bond donors: H on FON (F, O or N). Both hooks compress a rule the learner should also be able to explain (electrons "owned" vs. shared; electronegativity and small size). Lewis-structure strategy (count electrons, connect, complete octets, minimize formal charge) is a procedure to practice, not to acronym.

11. Oxidation and reduction (chemistry, biochemistry). "OIL RIG": Oxidation Is Loss, Reduction Is Gain (of electrons). Bridge to organic and biochemical usage: carbon is oxidized when it gains bonds to O (or loses bonds to H); NAD⁺ is reduced to NADH. Test with "Is ethanol → acetaldehyde an oxidation? Which coenzyme is reduced?"

12. Metric prefixes (physics, chemistry). "King Henry Died By Drinking Chocolate Milk" → kilo (10³), hecto (10²), deka (10¹), base unit, deci (10⁻¹), centi (10⁻²), milli (10⁻³). Each step is ×10, so conversion = counting steps. Extend with micro (10⁻⁶) and nano (10⁻⁹) by chunks of three. Physics content itself (F = ma, energy conservation) gets no mnemonic: derive and apply.

13. The Twelve (minor) prophets, in Protestant English and Masoretic order (biblical studies). Chunk into four triads and say them rhythmically: Hosea-Joel-Amos / Obadiah-Jonah-Micah / Nahum-Habakkuk-Zephaniah / Haggai-Zechariah-Malachi. Meaning layer: the last triad is post-exilic. Tradition note: Septuagint-based (Eastern Orthodox) Bibles order the first six Hosea, Amos, Micah, Joel, Obadiah, Jonah. Ask which canon the course uses before drilling.

14. Latin and theological terms, etymology beats keywords. Lex orandi, lex credendi ("the law of praying [is] the law of believing"): orandi → orate, oration; credendi → creed, credible. Creatio ex nihilo: nihil → nihilism, annihilate ("from nothing"). Imago Dei: image of God. Real roots carry meaning into new words; a sound-alike keyword ("ex Neil-ho") does not. Use keyword images only for words with no usable root, and then space the retrievals.

5. Anti-patterns

  • Crude, sexual, demeaning, or slur-based mnemonics, including widely circulated med-school ones. They alienate and embarrass learners in group study, and they are unnecessary: a clean hook of the same shape works.
  • Hooks that encode wrong or outdated facts (PVT TIM HALL without the arginine note; a taxonomy acrostic presented as if every kingdom scheme agreed). Audit inherited mnemonics.
  • Too many layers: a mnemonic to remember which mnemonic, nested acronyms, or one acrostic per lecture slide. Each layer is another thing to forget.
  • Mnemonics for understanding-based content: acronyms for mechanisms, derivations, philosophical arguments, or clinical reasoning. If a "why" question breaks it, it was the wrong tool.
  • Mnemonic before meaning: memorizing CADET before knowing what a right shift is.
  • Hook-only cards: a card that asks for the mnemonic instead of the fact.
  • Ambiguous letters: "M" that could be three terms in the same chapter.
  • Confusable hook sets without a contrast card.
  • Claiming permanence: hooks speed initial learning; without spaced retrieval, they can be forgotten faster than rote-learned material (Wang et al. 1992).
  • Never retiring: running a 12-word acrostic in an exam for an item that should be instant.

6. Closing a hook session

  1. Learner recalls each target without its hook, mixed order.
  2. Status per item (see tutor §8): hinted recall is not independent recall.
  3. Cards made for targets (hook as optional hint line), plus "why" cards.
  4. Next review proposed: tomorrow-ish, a few days, about a week. A suggestion is not a reminder; schedule only with a real scheduling tool, and verify it was created.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.