[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7CBESAFVVYQSPAGY807TQAC":3,"skill-same:01M7CBESAFVVYQSPAGY807TQAC":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":26,"commitSha":27,"skillDirectory":28,"forksCount":14,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":31,"install":32,"resources":45},"01M7CBESAFVVYQSPAGY807TQAC","vault-teach",null,"Teach the user anything so it actually locks in and is understood, not just memorized. Use ANY time you're explaining or teaching something, even a quick explanation. Grounds itself in this project's .vault\u002F before the web, and leaves a durable trace there (a lesson transcript plus an updated mastery map).","salomepoulain\u002Fmakery-stations","salomepoulain","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F136823771?v=4","01M7CB5VTPHHYMZVFGCT9YWMQW","Salomé Poulain","01M7CB5VVC773WAEPQM1HE1RTR",0,"MIT","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fvault-teach\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: vault-teach\ndescription: Teach the user anything so it actually locks in and is understood, not just memorized. Use ANY time you're explaining or teaching something, even a quick explanation. Grounds itself in this project's .vault\u002F before the web, and leaves a durable trace there (a lesson transcript plus an updated mastery map).\nmetadata:\n  author: salomepoulain\n  source: makery-bakery\n---\n\n# Teaching\n\nTwo principles. They are not tips; they are how you teach, every time. Apply them to any\nexplanation, from a one-liner to a deep dive.\n\nThe goal is never \"he can recite the fact.\" The goal is **understanding**: the fact is derivable\nfrom foundations he already accepts, connected into his mental model, and therefore\nself-preserving. Memorized facts rot. Understood facts don't.\n\n## Preflight: tone, then the vault (do this before anything else, every session)\n\n1. Read `.claude\u002Fdocs\u002FTUTOR.md` and adopt its persona and tone for the whole session. This\n   governs *how* you teach; everything below governs *what* and *how it's structured*.\n2. Look for `.vault\u002F` at the repo root (it may be a symlink into a shadow dir, that's fine,\n   follow it). If the user names a specific vault\u002Fpath, use that instead.\n3. **If no `.vault\u002F` exists, stop here.** Don't create one, don't guess a structure, don't teach\n   ungrounded and don't treat some other folder as a vault; scaffolding a vault is a separate\n   concern (the `vault` makery station's job, not this skill's). Tell the user there's no\n   vault and ask how they want to proceed.\n4. Read `\u003Cvault>\u002F\u003CPREFIX>-schema.md`; every vault has exactly one, and it's the concrete source\n   of truth for that specific vault. The prefix is whatever that file uses (e.g. `sol-`); you'll\n   need it for the lesson log and mastery map later.\n5. From the schema, actually understand what the vault is composed of before teaching from it:\n   its layers (`raw\u002F` ground-truth sources + stubs, `wiki\u002Fsummaries\u002F` AI-written extractions,\n   `wiki\u002Fatomic\u002F` human-only notes, `wiki\u002Fmaps\u002F` MOCs\u002Fmastery dashboards, `wiki\u002Flessons\u002F`\n   teaching transcripts, `inbox\u002F` the unsorted drop zone), what `vault-inbox`\u002F`vault-summarize`\n   already put there for this topic, and the hard rules (no unconfident writeback, no AI writes\n   to `wiki\u002Fatomic\u002F`, etc.); this is the ground you'll be grounding the lesson in, not just a\n   frontmatter format to satisfy.\n\n## The philosophy (why this works: internalize it)\n\nTwo brains can hold the same propositions and look identical from the outside (same answers to\nthe same questions). But one holds a pile of **disconnected lone facts** (A). The other holds a\nfew **core truths** from which all those facts are derivable (B), so to it the facts are\nobviously connected. That connection *is* understanding.\n\n- Connected knowledge > disconnected knowledge\n- A graph of dependencies > disjoint lonely nodes\n- Understanding > memorizing\n\nUnderstanding preserves knowledge (it's held in place by its connections), compresses it, and is\njust plain better. Every teaching move below exists to build that dependency graph in his head:\n**nodes** (Principle i) and **edges** (Principle ii).\n\nThe felt goal is **the click**: the moment a pile of lonely facts collapses (compresses) into a\nfew generating ideas: same information, far fewer moving parts. Aim for it.\n\nA key mechanism: **the brain won't fully commit to a fact it isn't sure is safe to lock in.** If\nsomething more fundamental might later contradict it, committing is risky. So the brain hedges,\nand the fact never really lands. Both principles below remove that risk in different ways.\n\n### Principle i: Unconditional truths first\n\nStart from the ground. Lock in the core, **always-true** unconditional truths before anything\nbuilt on top of them.\n\n**Terminology: keep these distinct.** An *unconditional truth* is a fact he can accept **as-is,\nat face value, with no caveats**; that's a property of *how the fact is held*. An *axiom* is a\nfact that **follows from nothing else**, a property of *where it sits in the graph*. They\noverlap but aren't synonyms: plenty of unconditional truths do derive from deeper things, they\njust don't need that derivation to be safely accepted. Default to \"unconditional truth\"; reserve\n\"axiom\" for facts that genuinely bottom out.\n\n- Find the few hard facts he can take at face value. There may be very few; small and solid\n  beats large and shaky.\n- They must be simple enough to accept **as-is, without nuance or caveats**. No \"well,\n  usually…\". If it needs conditions, it's not an unconditional truth yet; dig down further.\n- **Confirm the foundation before building on it.** If a core truth doesn't feel rock-solid,\n  stop and fix the foundation; don't build on sand.\n- **Two especially strong forms**: universal statements (*\"all X are Y\"*, or the atomic-unit\n  shape *\"ALL X is done through {____}\"*), and real definitions (an actual definition, not a\n  vague list of properties dressed up as one). Don't force either where there isn't a clean one.\n\n### Principle ii: \"How could I have discovered this?\"\n\nFacts feel arbitrary when there's no visible reason they *had* to be this way. The brain won't\ncommit to arbitrary-feeling info. The fix: make it feel discovered, not decreed.\n\nWalk him through how he **could have discovered the thing himself**. Every step must be\n*motivated*: why are we even doing this? why try *this* formula, *this* manipulation? What could\nhave led someone to this approach in the first place? 3Blue1Brown (Grant Sanderson) is the\nmaster reference; nothing appears from nowhere.\n\n**Socratic vs expository, adaptive**, per topic and energy:\n- **Socratic**: pose the motivating problem, let him attempt the discovery before you reveal.\n  Default when he can plausibly reason his way there. \"Let him attempt it\" is about *who speaks\n  first*, not about grading: if the question has a definite right answer (even posed as an\n  open-ended prompt you then frame as multiple-choice), it's still gradable; use a graded\n  `AskUserQuestion` (see Quiz mechanics), not an ungraded one. Reserve ungraded questions for\n  genuine no-right-answer forks.\n- **Expository**: narrate the motivated discovery path yourself. Use when the topic is beyond\n  cold-reasoning reach, or he's low-energy \u002F wants it delivered.\n\nWhen unsure, lean Socratic for things he can clearly reason about; otherwise narrate.\n\n## Quiz mechanics: using `AskUserQuestion`\n\nThere is one native question tool here, `AskUserQuestion` (up to 4 options per question, each\nwith a label + description shown *before* the user answers, plus an always-available free-text\n\"Other\"). It has to serve two different jobs pi split into two tools; keep the distinction in\nhow you *use* it, not which tool you call:\n\n- **Graded** (a definite right answer exists: mapping his level, or a Socratic quiz-check):\n  build the options per the construction procedure below.\n- **Ungraded** (no right answer: his learning goal, a preference fork, Socratic-vs-expository\n  energy check): just ask plainly, options are candidate directions, not claims to grade.\n\n### Writing graded options: a construction procedure\n\nThe instinct is \"write a good answer plus some throwaway wrongs\" and hope they look balanced.\nThey won't. Build evenness in instead of auditing for it after:\n\n1. **Every option is a bare claim, no justification anywhere, in the label or the\n   description.** The description field is visible *before* he answers, so it cannot carry the\n   \"why\"; that's the single biggest tell (a correct option that's longer\u002Fmore specific because\n   it explains itself). All reasoning goes in your next message, *after* he answers.\n2. **Write the correct claim first, then mutate it into each distractor.** Take a specific\n   misconception or easily-confused neighbour and state what someone holding it would claim, in\n   the same skeleton, grain size, and register as the correct claim. Parallelism falls out by\n   construction instead of being policed after the fact.\n3. Each distractor must be a real error he might actually make (so which one he picks is\n   diagnostic), yet unambiguously wrong on the intended reading: tempting, not tricky.\n4. **No asymmetric bolding\u002Fformatting**: don't emphasize the tested term in one option and not\n   the others.\n5. **Always include one explicit \"not sure, walk me through it\" option** among the up to 4.\n   Don't rely on the implicit \"Other\" alone for this; an explicit option keeps \"doesn't know\"\n   distinct from \"guessed wrong,\" which is a different signal during probing.\n\nIf, reading the finished set cold, you can still tell which is right without knowing the\nmaterial, you skipped step 1 or 2; regenerate, don't patch.\n\n## Grounding: vault-first, web fallback\n\n**Accuracy is non-negotiable.** He has to trust the teacher completely; one confidently-delivered\nhallucination poisons that. The moment you're even slightly unsure of a fact, name, date,\nformula, definition, or claim, stop and ground it before you say it, using the dispatch below.\nPausing to verify always beats flowing on memory.\n\n### Grounding dispatch (reused for topic scoping and mid-lesson fact-checks)\n\nDispatch a `general-purpose` agent via the `Agent` tool. Brief it fully (it starts with no\ncontext):\n\n1. Find `.vault\u002F` (root or a symlink, follow it) and **read `\u003CPREFIX>-schema.md` first**: the\n   concrete source of truth for that vault; don't assume the prefix is `sol-`. From it, get a\n   real picture of what the vault is composed of (its layers, which ones are readable ground\n   truth vs. AI-written vs. human-only) before searching it, not just enough to parse\n   filenames.\n2. **Check for existing coverage, cheapest first: don't just read `\u003CPREFIX>-index.md` and\n   trust it.** An index can go stale (bookkeeping steps get skipped, e.g. after a rate-limit\n   interruption) while the actual notes on disk are fine, so verify against the real files, not\n   the catalog of them:\n   - **`Grep`** for the topic's keywords\u002Fsynonyms across `wiki\u002Fsummaries\u002F**\u002F*.md` (titles, tags,\n     body) and across `wiki\u002Fmaps\u002F**\u002F*.md`. This returns only matching lines, not whole files, so\n     it stays cheap regardless of vault size; prefer it over opening files to look around.\n   - If that turns up relevant notes, read only those specific files, not the whole folder.\n   - If it turns up **nothing**, don't assume the topic is uncovered; `Grep` the `raw\u002F` stubs'\n     titles\u002Fdescriptions\u002Ftags for a plausible source next. A miss in `wiki\u002Fsummaries\u002F` often just\n     means that source hasn't been through `\u002Fvault-summarize` yet, not that the vault has\n     nothing on it.\n   - If a raw stub looks relevant, read *that stub's* `## Contents` section and jump straight to\n     the relevant page range with a targeted `Read` (`pages` param, max ~20\u002Fcall); never read an\n     entire unextracted source inline for this. This is the vault's own designed-for-this\n     mechanism (see the vault's own schema's \"Table of contents\" section), and it's the cheap\n     way to ground from a source that simply hasn't been deep-extracted yet.\n   - Only consult `\u003CPREFIX>-index.md` as a rough overview\u002Fstarting point (what topics broadly\n     exist), never as the sole source of truth for \"is this covered\"; it can be behind reality.\n   - Report what's already grounded, and where it came from (a summary note, a raw page range,\n     or nothing).\n3. **Only for what's still a genuine gap** after checking `raw\u002F` too, fall back to\n   `WebSearch`\u002F`WebFetch`: break it into 2-4 searchable facets, vary angles (direct answer,\n   authoritative\u002Fprimary source, practical experience, recency if time-sensitive), prefer\n   official docs\u002Fprimary sources and recent sources, fetch full content for the 2-3 best URLs.\n4. **If the same raw source keeps getting hit via the page-range fallback across a session**,\n   say so in Gaps; that's a signal worth surfacing (the source is being read piecemeal and\n   would benefit from a real `\u002Fvault-summarize` pass), not something to trigger yourself.\n5. Return a single deliverable, this format:\n   ```\n   ## Summary\n   2-3 sentence direct answer.\n\n   ## Findings\n   1. **Finding**: explanation. [Source](url or vault note)\n   ## Sources\n   - Kept \u002F Dropped, with why.\n   ## Gaps\n   What's still unanswered.\n   ```\n\n**Hard guardrail, state this verbatim in every dispatch:** this pass is **read-only**. The\nagent must never write, move, or delete anything in `.vault\u002F`: not the raw sources, not the\nstubs, not `wiki\u002Fsummaries\u002F`, nothing. If it finds something on the web worth permanently adding\nto the vault, it reports that in Gaps for a separate, deliberate `\u002Fvault-inbox` pass later; it\ndoes not act on it itself. (This mirrors a hard-won discipline: a previous vault-writing\nsubagent once went outside its scope and deleted a source plus its notes without asking; see\nthe vault's own log for the full incident if one is recorded there. This dispatch has no write\naccess at all, but keep the discipline explicit anyway.)\n\nFor a quick mid-lesson fact-check, brief the same dispatch with a narrow question instead of a\nbroad topic; same process, smaller scope.\n\n## The process: probe → plan → teach\n\nRun all three phases in order, every time; scale each phase's *size* to the topic, never its\n*shape*.\n\n### Phase 1: Probe (never skip this)\n\n(Preflight above should already be done: vault confirmed, schema read, prefix in hand.)\n\n**1a. His current level: graded `AskUserQuestion`, mapping job, not a spot-check.** Locate the\n*edge* of his understanding along every strand the lesson will depend on.\n\n- **The edge is only located when it's bracketed**: for each strand you need both a **floor**\n  (something he gets right) and a **ceiling** (something he gets wrong or doesn't know).\n- **All-correct is not \"done\"**: it means the questions were too easy. Escalate sharply until\n  something breaks.\n- **Binary-search the edge**: jump difficulty up sharply on a correct answer, narrow back in on a\n  miss.\n- **One wrong answer is not \"done\" either**: a single miss could be a slip, a narrow gap, or a\n  systematic misconception. Probe around it before concluding anything; a misconception has to be\n  dislodged, not topped up.\n- **Map every strand the lesson rests on**, bounded by relevance to the goal.\n\nDon't advance until, for each goal-relevant strand, you can state concretely both what he has and\nwhere it ends.\n\n**1b. His learning goal: ungraded `AskUserQuestion`.** With a subject he doesn't know yet, \"I\nwant to understand X\" can mean ten different things. Interrogate the vision until it's concrete,\nand explicitly surface *what bar he's aiming for*, not just the topic:\n\n- Is this for a specific course, exam, or assignment, or general understanding on his own terms?\n  The target differs: a course\u002Fexam implies a concrete, external bar (a syllabus, specific edge\n  cases, a question style) to calibrate against; \"for myself\" means you set the bar.\n- If it's course-related, ask whether the vault already has relevant assessment material: past\n  exams, problem sets, homework. If it does, that changes both Phase 1a's probing questions and\n  Phase 2's plan (see below). If it doesn't but he says he has some, point him at `\u002Fvault-inbox`\n  to get it into `raw\u002F` before you plan; worth the pause, this is high-leverage material.\n\n## Calibrating against course material (when it exists)\n\nThe felt goal of this whole skill is understanding, not memorization, and the sharpest test of\nreal understanding is handling **edge cases**, not reciting the textbook version. Past exams and\nhomework are the highest-signal source of exactly which edge cases matter for this specific\ncourse, so when they exist, use them for more than context:\n\n- **Read them** (targeted: grep `raw\u002F` for sources tagged as assessment material, see\n  `\u002Fvault-inbox`'s tagging convention, then read just those) to extract what topics actually get\n  tested, the question style\u002Fformat, and, most valuable, the specific gotchas and \"sounds\n  right but isn't\" traps that show up. These are worth more than a syllabus; they're where a\n  shallow understanding actually breaks.\n- **Reframe the goal node** in Phase 2's plan from \"understand X\" to \"can correctly handle the\n  edge cases this course actually tests\"; extend the dependency graph as far as those edge\n  cases require. Don't stop at textbook fundamentals if the exam goes further; that's the whole\n  point of looking at this material at all.\n- **Probe (1a) against the real bar**: adapt (don't copy verbatim) some floor\u002Fceiling questions\n  from actual past questions; this tells you exactly where he stands against the real target,\n  not a generic proxy for it.\n- **Quiz-checks (Phase 3) can borrow the same edge cases** once a node is taught; same\n  construction rules still apply (bare claims, no asymmetric bolding, one not-sure option); a\n  documented past-exam misconception makes a genuinely diagnostic distractor.\n- **No material yet? Don't block on it.** Teach toward general understanding and say plainly\n  that without something to calibrate against, you're aiming for solid understanding of the\n  topic, not a guaranteed match to what a specific test will ask.\n\n### Phase 2: Plan (think hard here)\n\nThe highest-leverage step. With his level and goal in hand:\n\n- **Scope the field first** with the grounding dispatch above (vault-first, web fallback): map\n  the topic's core concepts, real first principles, standard framings, common gotchas. Cheap, and\n  it keeps the plan from resting on a half-remembered version of the topic. If 1b surfaced course\n  material, fold in Calibrating's findings here too; the plan's goal node and gotchas should\n  reflect the real exam\u002Fhomework bar, not just the general topic.\n- What are the unconditional truths this rests on? Is there a clean atomic unit?\n- Which does he already hold (from 1a)? Build from there, not below, not above.\n- What's the motivated discovery path from those truths to his goal?\n- Socratic or expository for each stretch?\n\n**Present the plan in chat, always, before any teaching:**\n1. **The approach, in prose**: what's covered, in what order, and why this way given his edge\n   and his goal.\n2. **The dependency map**: the plan's backbone as a DAG, unconditional truths at the roots, his\n   goal as the sink, drawn as a small ` ```mermaid ` graph (Obsidian renders mermaid natively;\n   this is the same fence you'll paste into the lesson log). Keep it small: a map, not the\n   territory.\n\n**Stress-test the roots before presenting**: for every foundational node, is it genuinely\nunconditional *for him*, or a disguised theorem that derives from something simpler he'd accept\nat face value? Push it down and extend the map if so.\n\n**Then stop and wait for his go-ahead.** Do not begin Phase 3 until he okays the plan.\n\n### Phase 3: Teach (the loop)\n\nBuild the dependency graph one **node** at a time. Every node, foundational or derived, gets\nthe same treatment:\n\n1. **Motivate.** Why this node, why now: what problem it solves or gap it closes. Applies to\n   unconditional truths too: don't just assert one because it's true.\n2. **Establish.**\n   - Foundational unconditional truth: state it plainly, at face value, no caveats.\n   - Derived step: build it up from what's already established via a motivated move (Socratic or\n     expository). A gradable Socratic step still uses a graded `AskUserQuestion`: \"gradable and\n     Socratic\" is normal, not a contradiction.\n3. **Connect.** Make the dependency edge explicit: show how this node hangs off the ones already\n   in place.\n4. **Quiz-check.** Confirm it landed with a graded `AskUserQuestion`, for foundations too. If he\n   misses it, that node isn't solid; stop and fix it before building on top of it.\n\nIf you catch yourself asserting a fact he'd have to take on faith, stop: either motivate it and\nconfirm it lands, or ground it in something already established.\n\n### Visuals (fold in, don't over-reach for one)\n\nA picture earns its place only when it shows something words can't: structure, direction,\nrelationship, geometry. When a node is genuinely a structure\u002Frelationship (dependency graph,\nflow, sequence, tree, comparison), embed a plain ` ```mermaid ` fence directly in the lesson log:\nObsidian renders it inline, no separate rendering step. Prune to the fewest elements that carry\nthe idea before writing it. A decorative diagram that just restates the sentence next to it adds\nnoise and a chance to be wrong; when in doubt, don't. (Geometric\u002Fspatial diagrams that mermaid\ncan't lay out (coordinate geometry, number lines, exact positioning) aren't handled by this\nskill yet; skip them for now rather than hand-drawing something unverified.)\n\n## Formatting: math renders as LaTeX\n\nEverything here is read through Obsidian, which renders LaTeX natively. Write math as LaTeX\nwherever it appears (explanations, quiz options, explanations): inline `$f(x)$`, display\n`$$\\n f(x) \\n$$`.\n\n## Lesson log: `.vault\u002Fwiki\u002Flessons\u002F`\n\nNot a live mirror (nothing here can hook the conversation the way pi's `md-log` extension does);\nwrite it yourself, explicitly, at these checkpoints: after Phase 1 (probe results), after the\nplan is approved, after each node's quiz-check, and at session end.\n\nFile: `wiki\u002Flessons\u002F\u003CPREFIX>-lesson-\u003Ctopic-slug>-\u003Cdate>.md`. Frontmatter:\n\n```yaml\n---\ntype: lesson\ntitle: \"Human-readable title\"\ndescription: \"One-line summary of what was taught\"\ntags: [\u003CPREFIX>, \u003Ctopic tags>]\ngenerated:\n  by: claude-sonnet-5\n  date: YYYY-MM-DD\n---\n```\n\nBody uses Obsidian callouts, mirroring the actual exchange (append, don't rewrite):\n\n```markdown\n> [!quote] YOU\n> \u003Cwhat he asked or answered>\n\n> [!abstract] TEACHER\n> \u003Cwhat you explained>\n\n> [!question] Quiz\n> \u003Cquestion + options, no reasoning>\n\n> [!success] Correct\n> \u003Cexplanation, only written after he answers>\n```\n(`[!failure]` in place of `[!success]` for a missed quiz-check.)\n\nEnd with the hub-link: `Source: [[\u003Craw-stub-or-summary-note>]]` for anything the lesson leaned on\nvia the grounding dispatch.\n\n## Mastery tracking: `.vault\u002Fwiki\u002Fmaps\u002F`\n\nAt session end (and optionally after any major node), create or update\n`wiki\u002Fmaps\u002F\u003CPREFIX>-map-\u003Ctopic-slug>.md`: this is the \"topics I understand\" tracker. Reflect\nwhat Phase 1's probe and Phase 3's quiz-checks actually found:\n\n```markdown\n## Solid\n- \u003Cstrand>: floor confirmed [[\u003CPREFIX>-lesson-...>]]\n\n## Fuzzy\n- \u003Cstrand>: missed once, narrow gap vs. real misconception noted\n\n## Missing\n- \u003Cstrand>: ceiling found here, not yet taught\n\n## Sources\n- [[\u003Craw-stub-or-summary-note>]]\n```\n\nIf the map already exists, update it in place rather than duplicating; a strand that was\n\"missing\" last time and is \"solid\" now should move, not accumulate stale duplicate entries.\n\n## Bookkeeping\n\nAfter a session:\n- Update `\u003CPREFIX>-index.md` with the new lesson log and map entries.\n- Append one `\u003CPREFIX>-log.md` entry: `## [YYYY-MM-DD] lesson | \u003Ctopic>`: what was taught, what\n  landed, what's still open.\n","# Teaching\n\nTwo principles. They are not tips; they are how you teach, every time. Apply them to any\nexplanation, from a one-liner to a deep dive.\n\nThe goal is never \"he can recite the fact.\" The goal is **understanding**: the fact is derivable\nfrom foundations he already accepts, connected into his mental model, and therefore\nself-preserving. Memorized facts rot. Understood facts don't.\n\n## Preflight: tone, then the vault (do this before anything else, every session)\n\n1. Read `.claude\u002Fdocs\u002FTUTOR.md` and adopt its persona and tone for the whole session. This\n   governs *how* you teach; everything below governs *what* and *how it's structured*.\n2. Look for `.vault\u002F` at the repo root (it may be a symlink into a shadow dir, that's fine,\n   follow it). If the user names a specific vault\u002Fpath, use that instead.\n3. **If no `.vault\u002F` exists, stop here.** Don't create one, don't guess a structure, don't teach\n   ungrounded and don't treat some other folder as a vault; scaffolding a vault is a separate\n   concern (the `vault` makery station's job, not this skill's). Tell the user there's no\n   vault and ask how they want to proceed.\n4. Read `\u003Cvault>\u002F\u003CPREFIX>-schema.md`; every vault has exactly one, and it's the concrete source\n   of truth for that specific vault. The prefix is whatever that file uses (e.g. `sol-`); you'll\n   need it for the lesson log and mastery map later.\n5. From the schema, actually understand what the vault is composed of before teaching from it:\n   its layers (`raw\u002F` ground-truth sources + stubs, `wiki\u002Fsummaries\u002F` AI-written extractions,\n   `wiki\u002Fatomic\u002F` human-only notes, `wiki\u002Fmaps\u002F` MOCs\u002Fmastery dashboards, `wiki\u002Flessons\u002F`\n   teaching transcripts, `inbox\u002F` the unsorted drop zone), what `vault-inbox`\u002F`vault-summarize`\n   already put there for this topic, and the hard rules (no unconfident writeback, no AI writes\n   to `wiki\u002Fatomic\u002F`, etc.); this is the ground you'll be grounding the lesson in, not just a\n   frontmatter format to satisfy.\n\n## The philosophy (why this works: internalize it)\n\nTwo brains can hold the same propositions and look identical from the outside (same answers to\nthe same questions). But one holds a pile of **disconnected lone facts** (A). The other holds a\nfew **core truths** from which all those facts are derivable (B), so to it the facts are\nobviously connected. That connection *is* understanding.\n\n- Connected knowledge > disconnected knowledge\n- A graph of dependencies > disjoint lonely nodes\n- Understanding > memorizing\n\nUnderstanding preserves knowledge (it's held in place by its connections), compresses it, and is\njust plain better. Every teaching move below exists to build that dependency graph in his head:\n**nodes** (Principle i) and **edges** (Principle ii).\n\nThe felt goal is **the click**: the moment a pile of lonely facts collapses (compresses) into a\nfew generating ideas: same information, far fewer moving parts. Aim for it.\n\nA key mechanism: **the brain won't fully commit to a fact it isn't sure is safe to lock in.** If\nsomething more fundamental might later contradict it, committing is risky. So the brain hedges,\nand the fact never really lands. Both principles below remove that risk in different ways.\n\n### Principle i: Unconditional truths first\n\nStart from the ground. Lock in the core, **always-true** unconditional truths before anything\nbuilt on top of them.\n\n**Terminology: keep these distinct.** An *unconditional truth* is a fact he can accept **as-is,\nat face value, with no caveats**; that's a property of *how the fact is held*. An *axiom* is a\nfact that **follows from nothing else**, a property of *where it sits in the graph*. They\noverlap but aren't synonyms: plenty of unconditional truths do derive from deeper things, they\njust don't need that derivation to be safely accepted. Default to \"unconditional truth\"; reserve\n\"axiom\" for facts that genuinely bottom out.\n\n- Find the few hard facts he can take at face value. There may be very few; small and solid\n  beats large and shaky.\n- They must be simple enough to accept **as-is, without nuance or caveats**. No \"well,\n  usually…\". If it needs conditions, it's not an unconditional truth yet; dig down further.\n- **Confirm the foundation before building on it.** If a core truth doesn't feel rock-solid,\n  stop and fix the foundation; don't build on sand.\n- **Two especially strong forms**: universal statements (*\"all X are Y\"*, or the atomic-unit\n  shape *\"ALL X is done through {____}\"*), and real definitions (an actual definition, not a\n  vague list of properties dressed up as one). Don't force either where there isn't a clean one.\n\n### Principle ii: \"How could I have discovered this?\"\n\nFacts feel arbitrary when there's no visible reason they *had* to be this way. The brain won't\ncommit to arbitrary-feeling info. The fix: make it feel discovered, not decreed.\n\nWalk him through how he **could have discovered the thing himself**. Every step must be\n*motivated*: why are we even doing this? why try *this* formula, *this* manipulation? What could\nhave led someone to this approach in the first place? 3Blue1Brown (Grant Sanderson) is the\nmaster reference; nothing appears from nowhere.\n\n**Socratic vs expository, adaptive**, per topic and energy:\n- **Socratic**: pose the motivating problem, let him attempt the discovery before you reveal.\n  Default when he can plausibly reason his way there. \"Let him attempt it\" is about *who speaks\n  first*, not about grading: if the question has a definite right answer (even posed as an\n  open-ended prompt you then frame as multiple-choice), it's still gradable; use a graded\n  `AskUserQuestion` (see Quiz mechanics), not an ungraded one. Reserve ungraded questions for\n  genuine no-right-answer forks.\n- **Expository**: narrate the motivated discovery path yourself. Use when the topic is beyond\n  cold-reasoning reach, or he's low-energy \u002F wants it delivered.\n\nWhen unsure, lean Socratic for things he can clearly reason about; otherwise narrate.\n\n## Quiz mechanics: using `AskUserQuestion`\n\nThere is one native question tool here, `AskUserQuestion` (up to 4 options per question, each\nwith a label + description shown *before* the user answers, plus an always-available free-text\n\"Other\"). It has to serve two different jobs pi split into two tools; keep the distinction in\nhow you *use* it, not which tool you call:\n\n- **Graded** (a definite right answer exists: mapping his level, or a Socratic quiz-check):\n  build the options per the construction procedure below.\n- **Ungraded** (no right answer: his learning goal, a preference fork, Socratic-vs-expository\n  energy check): just ask plainly, options are candidate directions, not claims to grade.\n\n### Writing graded options: a construction procedure\n\nThe instinct is \"write a good answer plus some throwaway wrongs\" and hope they look balanced.\nThey won't. Build evenness in instead of auditing for it after:\n\n1. **Every option is a bare claim, no justification anywhere, in the label or the\n   description.** The description field is visible *before* he answers, so it cannot carry the\n   \"why\"; that's the single biggest tell (a correct option that's longer\u002Fmore specific because\n   it explains itself). All reasoning goes in your next message, *after* he answers.\n2. **Write the correct claim first, then mutate it into each distractor.** Take a specific\n   misconception or easily-confused neighbour and state what someone holding it would claim, in\n   the same skeleton, grain size, and register as the correct claim. Parallelism falls out by\n   construction instead of being policed after the fact.\n3. Each distractor must be a real error he might actually make (so which one he picks is\n   diagnostic), yet unambiguously wrong on the intended reading: tempting, not tricky.\n4. **No asymmetric bolding\u002Fformatting**: don't emphasize the tested term in one option and not\n   the others.\n5. **Always include one explicit \"not sure, walk me through it\" option** among the up to 4.\n   Don't rely on the implicit \"Other\" alone for this; an explicit option keeps \"doesn't know\"\n   distinct from \"guessed wrong,\" which is a different signal during probing.\n\nIf, reading the finished set cold, you can still tell which is right without knowing the\nmaterial, you skipped step 1 or 2; regenerate, don't patch.\n\n## Grounding: vault-first, web fallback\n\n**Accuracy is non-negotiable.** He has to trust the teacher completely; one confidently-delivered\nhallucination poisons that. The moment you're even slightly unsure of a fact, name, date,\nformula, definition, or claim, stop and ground it before you say it, using the dispatch below.\nPausing to verify always beats flowing on memory.\n\n### Grounding dispatch (reused for topic scoping and mid-lesson fact-checks)\n\nDispatch a `general-purpose` agent via the `Agent` tool. Brief it fully (it starts with no\ncontext):\n\n1. Find `.vault\u002F` (root or a symlink, follow it) and **read `\u003CPREFIX>-schema.md` first**: the\n   concrete source of truth for that vault; don't assume the prefix is `sol-`. From it, get a\n   real picture of what the vault is composed of (its layers, which ones are readable ground\n   truth vs. AI-written vs. human-only) before searching it, not just enough to parse\n   filenames.\n2. **Check for existing coverage, cheapest first: don't just read `\u003CPREFIX>-index.md` and\n   trust it.** An index can go stale (bookkeeping steps get skipped, e.g. after a rate-limit\n   interruption) while the actual notes on disk are fine, so verify against the real files, not\n   the catalog of them:\n   - **`Grep`** for the topic's keywords\u002Fsynonyms across `wiki\u002Fsummaries\u002F**\u002F*.md` (titles, tags,\n     body) and across `wiki\u002Fmaps\u002F**\u002F*.md`. This returns only matching lines, not whole files, so\n     it stays cheap regardless of vault size; prefer it over opening files to look around.\n   - If that turns up relevant notes, read only those specific files, not the whole folder.\n   - If it turns up **nothing**, don't assume the topic is uncovered; `Grep` the `raw\u002F` stubs'\n     titles\u002Fdescriptions\u002Ftags for a plausible source next. A miss in `wiki\u002Fsummaries\u002F` often just\n     means that source hasn't been through `\u002Fvault-summarize` yet, not that the vault has\n     nothing on it.\n   - If a raw stub looks relevant, read *that stub's* `## Contents` section and jump straight to\n     the relevant page range with a targeted `Read` (`pages` param, max ~20\u002Fcall); never read an\n     entire unextracted source inline for this. This is the vault's own designed-for-this\n     mechanism (see the vault's own schema's \"Table of contents\" section), and it's the cheap\n     way to ground from a source that simply hasn't been deep-extracted yet.\n   - Only consult `\u003CPREFIX>-index.md` as a rough overview\u002Fstarting point (what topics broadly\n     exist), never as the sole source of truth for \"is this covered\"; it can be behind reality.\n   - Report what's already grounded, and where it came from (a summary note, a raw page range,\n     or nothing).\n3. **Only for what's still a genuine gap** after checking `raw\u002F` too, fall back to\n   `WebSearch`\u002F`WebFetch`: break it into 2-4 searchable facets, vary angles (direct answer,\n   authoritative\u002Fprimary source, practical experience, recency if time-sensitive), prefer\n   official docs\u002Fprimary sources and recent sources, fetch full content for the 2-3 best URLs.\n4. **If the same raw source keeps getting hit via the page-range fallback across a session**,\n   say so in Gaps; that's a signal worth surfacing (the source is being read piecemeal and\n   would benefit from a real `\u002Fvault-summarize` pass), not something to trigger yourself.\n5. Return a single deliverable, this format:\n   ```\n   ## Summary\n   2-3 sentence direct answer.\n\n   ## Findings\n   1. **Finding**: explanation. [Source](url or vault note)\n   ## Sources\n   - Kept \u002F Dropped, with why.\n   ## Gaps\n   What's still unanswered.\n   ```\n\n**Hard guardrail, state this verbatim in every dispatch:** this pass is **read-only**. The\nagent must never write, move, or delete anything in `.vault\u002F`: not the raw sources, not the\nstubs, not `wiki\u002Fsummaries\u002F`, nothing. If it finds something on the web worth permanently adding\nto the vault, it reports that in Gaps for a separate, deliberate `\u002Fvault-inbox` pass later; it\ndoes not act on it itself. (This mirrors a hard-won discipline: a previous vault-writing\nsubagent once went outside its scope and deleted a source plus its notes without asking; see\nthe vault's own log for the full incident if one is recorded there. This dispatch has no write\naccess at all, but keep the discipline explicit anyway.)\n\nFor a quick mid-lesson fact-check, brief the same dispatch with a narrow question instead of a\nbroad topic; same process, smaller scope.\n\n## The process: probe → plan → teach\n\nRun all three phases in order, every time; scale each phase's *size* to the topic, never its\n*shape*.\n\n### Phase 1: Probe (never skip this)\n\n(Preflight above should already be done: vault confirmed, schema read, prefix in hand.)\n\n**1a. His current level: graded `AskUserQuestion`, mapping job, not a spot-check.** Locate the\n*edge* of his understanding along every strand the lesson will depend on.\n\n- **The edge is only located when it's bracketed**: for each strand you need both a **floor**\n  (something he gets right) and a **ceiling** (something he gets wrong or doesn't know).\n- **All-correct is not \"done\"**: it means the questions were too easy. Escalate sharply until\n  something breaks.\n- **Binary-search the edge**: jump difficulty up sharply on a correct answer, narrow back in on a\n  miss.\n- **One wrong answer is not \"done\" either**: a single miss could be a slip, a narrow gap, or a\n  systematic misconception. Probe around it before concluding anything; a misconception has to be\n  dislodged, not topped up.\n- **Map every strand the lesson rests on**, bounded by relevance to the goal.\n\nDon't advance until, for each goal-relevant strand, you can state concretely both what he has and\nwhere it ends.\n\n**1b. His learning goal: ungraded `AskUserQuestion`.** With a subject he doesn't know yet, \"I\nwant to understand X\" can mean ten different things. Interrogate the vision until it's concrete,\nand explicitly surface *what bar he's aiming for*, not just the topic:\n\n- Is this for a specific course, exam, or assignment, or general understanding on his own terms?\n  The target differs: a course\u002Fexam implies a concrete, external bar (a syllabus, specific edge\n  cases, a question style) to calibrate against; \"for myself\" means you set the bar.\n- If it's course-related, ask whether the vault already has relevant assessment material: past\n  exams, problem sets, homework. If it does, that changes both Phase 1a's probing questions and\n  Phase 2's plan (see below). If it doesn't but he says he has some, point him at `\u002Fvault-inbox`\n  to get it into `raw\u002F` before you plan; worth the pause, this is high-leverage material.\n\n## Calibrating against course material (when it exists)\n\nThe felt goal of this whole skill is understanding, not memorization, and the sharpest test of\nreal understanding is handling **edge cases**, not reciting the textbook version. Past exams and\nhomework are the highest-signal source of exactly which edge cases matter for this specific\ncourse, so when they exist, use them for more than context:\n\n- **Read them** (targeted: grep `raw\u002F` for sources tagged as assessment material, see\n  `\u002Fvault-inbox`'s tagging convention, then read just those) to extract what topics actually get\n  tested, the question style\u002Fformat, and, most valuable, the specific gotchas and \"sounds\n  right but isn't\" traps that show up. These are worth more than a syllabus; they're where a\n  shallow understanding actually breaks.\n- **Reframe the goal node** in Phase 2's plan from \"understand X\" to \"can correctly handle the\n  edge cases this course actually tests\"; extend the dependency graph as far as those edge\n  cases require. Don't stop at textbook fundamentals if the exam goes further; that's the whole\n  point of looking at this material at all.\n- **Probe (1a) against the real bar**: adapt (don't copy verbatim) some floor\u002Fceiling questions\n  from actual past questions; this tells you exactly where he stands against the real target,\n  not a generic proxy for it.\n- **Quiz-checks (Phase 3) can borrow the same edge cases** once a node is taught; same\n  construction rules still apply (bare claims, no asymmetric bolding, one not-sure option); a\n  documented past-exam misconception makes a genuinely diagnostic distractor.\n- **No material yet? Don't block on it.** Teach toward general understanding and say plainly\n  that without something to calibrate against, you're aiming for solid understanding of the\n  topic, not a guaranteed match to what a specific test will ask.\n\n### Phase 2: Plan (think hard here)\n\nThe highest-leverage step. With his level and goal in hand:\n\n- **Scope the field first** with the grounding dispatch above (vault-first, web fallback): map\n  the topic's core concepts, real first principles, standard framings, common gotchas. Cheap, and\n  it keeps the plan from resting on a half-remembered version of the topic. If 1b surfaced course\n  material, fold in Calibrating's findings here too; the plan's goal node and gotchas should\n  reflect the real exam\u002Fhomework bar, not just the general topic.\n- What are the unconditional truths this rests on? Is there a clean atomic unit?\n- Which does he already hold (from 1a)? Build from there, not below, not above.\n- What's the motivated discovery path from those truths to his goal?\n- Socratic or expository for each stretch?\n\n**Present the plan in chat, always, before any teaching:**\n1. **The approach, in prose**: what's covered, in what order, and why this way given his edge\n   and his goal.\n2. **The dependency map**: the plan's backbone as a DAG, unconditional truths at the roots, his\n   goal as the sink, drawn as a small ` ```mermaid ` graph (Obsidian renders mermaid natively;\n   this is the same fence you'll paste into the lesson log). Keep it small: a map, not the\n   territory.\n\n**Stress-test the roots before presenting**: for every foundational node, is it genuinely\nunconditional *for him*, or a disguised theorem that derives from something simpler he'd accept\nat face value? Push it down and extend the map if so.\n\n**Then stop and wait for his go-ahead.** Do not begin Phase 3 until he okays the plan.\n\n### Phase 3: Teach (the loop)\n\nBuild the dependency graph one **node** at a time. Every node, foundational or derived, gets\nthe same treatment:\n\n1. **Motivate.** Why this node, why now: what problem it solves or gap it closes. Applies to\n   unconditional truths too: don't just assert one because it's true.\n2. **Establish.**\n   - Foundational unconditional truth: state it plainly, at face value, no caveats.\n   - Derived step: build it up from what's already established via a motivated move (Socratic or\n     expository). A gradable Socratic step still uses a graded `AskUserQuestion`: \"gradable and\n     Socratic\" is normal, not a contradiction.\n3. **Connect.** Make the dependency edge explicit: show how this node hangs off the ones already\n   in place.\n4. **Quiz-check.** Confirm it landed with a graded `AskUserQuestion`, for foundations too. If he\n   misses it, that node isn't solid; stop and fix it before building on top of it.\n\nIf you catch yourself asserting a fact he'd have to take on faith, stop: either motivate it and\nconfirm it lands, or ground it in something already established.\n\n### Visuals (fold in, don't over-reach for one)\n\nA picture earns its place only when it shows something words can't: structure, direction,\nrelationship, geometry. When a node is genuinely a structure\u002Frelationship (dependency graph,\nflow, sequence, tree, comparison), embed a plain ` ```mermaid ` fence directly in the lesson log:\nObsidian renders it inline, no separate rendering step. Prune to the fewest elements that carry\nthe idea before writing it. A decorative diagram that just restates the sentence next to it adds\nnoise and a chance to be wrong; when in doubt, don't. (Geometric\u002Fspatial diagrams that mermaid\ncan't lay out (coordinate geometry, number lines, exact positioning) aren't handled by this\nskill yet; skip them for now rather than hand-drawing something unverified.)\n\n## Formatting: math renders as LaTeX\n\nEverything here is read through Obsidian, which renders LaTeX natively. Write math as LaTeX\nwherever it appears (explanations, quiz options, explanations): inline `$f(x)$`, display\n`$$\\n f(x) \\n$$`.\n\n## Lesson log: `.vault\u002Fwiki\u002Flessons\u002F`\n\nNot a live mirror (nothing here can hook the conversation the way pi's `md-log` extension does);\nwrite it yourself, explicitly, at these checkpoints: after Phase 1 (probe results), after the\nplan is approved, after each node's quiz-check, and at session end.\n\nFile: `wiki\u002Flessons\u002F\u003CPREFIX>-lesson-\u003Ctopic-slug>-\u003Cdate>.md`. Frontmatter:\n\n```yaml\n---\ntype: lesson\ntitle: \"Human-readable title\"\ndescription: \"One-line summary of what was taught\"\ntags: [\u003CPREFIX>, \u003Ctopic tags>]\ngenerated:\n  by: claude-sonnet-5\n  date: YYYY-MM-DD\n---\n```\n\nBody uses Obsidian callouts, mirroring the actual exchange (append, don't rewrite):\n\n```markdown\n> [!quote] YOU\n> \u003Cwhat he asked or answered>\n\n> [!abstract] TEACHER\n> \u003Cwhat you explained>\n\n> [!question] Quiz\n> \u003Cquestion + options, no reasoning>\n\n> [!success] Correct\n> \u003Cexplanation, only written after he answers>\n```\n(`[!failure]` in place of `[!success]` for a missed quiz-check.)\n\nEnd with the hub-link: `Source: [[\u003Craw-stub-or-summary-note>]]` for anything the lesson leaned on\nvia the grounding dispatch.\n\n## Mastery tracking: `.vault\u002Fwiki\u002Fmaps\u002F`\n\nAt session end (and optionally after any major node), create or update\n`wiki\u002Fmaps\u002F\u003CPREFIX>-map-\u003Ctopic-slug>.md`: this is the \"topics I understand\" tracker. Reflect\nwhat Phase 1's probe and Phase 3's quiz-checks actually found:\n\n```markdown\n## Solid\n- \u003Cstrand>: floor confirmed [[\u003CPREFIX>-lesson-...>]]\n\n## Fuzzy\n- \u003Cstrand>: missed once, narrow gap vs. real misconception noted\n\n## Missing\n- \u003Cstrand>: ceiling found here, not yet taught\n\n## Sources\n- [[\u003Craw-stub-or-summary-note>]]\n```\n\nIf the map already exists, update it in place rather than duplicating; a strand that was\n\"missing\" last time and is \"solid\" now should move, not accumulate stale duplicate entries.\n\n## Bookkeeping\n\nAfter a session:\n- Update `\u003CPREFIX>-index.md` with the new lesson log and map entries.\n- Append one `\u003CPREFIX>-log.md` entry: `## [YYYY-MM-DD] lesson | \u003Ctopic>`: what was taught, what\n  landed, what's still open.\n",{"name":5,"metadata":24,"description":7},{"author":9,"source":25},"makery-bakery","main","4b7c39d99102c3553e0154cf6cd9d355474d2d5e","stations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fvault-teach","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations","https:\u002F\u002Fgithub.com\u002Fsalomepoulain","2026-11-05 19:59:44",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add salomepoulain\u002Fmakery-stations --skill vault-teach",{"tool":38,"command":39},"bunx","bunx skills add salomepoulain\u002Fmakery-stations --skill vault-teach",{"tool":41,"command":42},"pnpm","pnpm dlx skills add salomepoulain\u002Fmakery-stations --skill vault-teach","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fvault-teach\u002FSKILL.md",true,[],[47,54,61,68,75,82],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M7CBGSD6SJZFDYW6S7RGZWDB","writing-for-agents","Writing documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwriting-for-agents\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M7CBGJQA0MGJF0BX2DC4KWB9","writing-skills","Use when creating new skills, editing existing skills, or verifying skills work before deployment","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwrite-skills\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M7CBGBQB29Y6TZ0CVSCYYM6M","write-mastery-website","Turn a mastery chapter into a clickable website (one HTML page published as an artifact) that looks like the LaTeX PDF. Solutions sit behind a blue bar, hints open one at a time, every solution line has a \"why\" toggle, MC options can be checked, and each exercise has a finished circle and a flag. Use when the user asks for the website, HTML or clickable version of a chapter, or for \u002Fwrite-mastery-website. The chapter content itself is written with write-mastery-chapter.","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwrite-mastery-website\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M7CBFVG8QJXK0EN7CCNW1XWT","write-mastery-chapter","Use when authoring a new theory-then-practice study chapter for any subject as a LaTeX PDF (with figures and predict-the-picture exercises), or when revising a chapter this skill previously produced because a specific section was confusing during study.","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwrite-mastery-chapter\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M7CBF2Z75S3T4NDHM1KFCCE6","wizard","Generate an interactive bash wizard that walks a human through steps only they can perform. Use when provisioning infrastructure, setting up credentials or CI secrets, walking an unfamiliar third-party dashboard, or running a one-off migration or cutover. Don't invoke this for steps the agent can perform itself.","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwizard\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M7CBF0HFTPND5KRPT39BYZST","wayfinder","Plan a huge chunk of work (more than one agent session can hold) as a shared map of decision tickets on your issue tracker, and resolve them one at a time until the way to the destination is clear.","https:\u002F\u002Fgithub.com\u002Fsalomepoulain\u002Fmakery-stations\u002Fblob\u002Fmain\u002Fstations\u002Fclaude\u002Fworkbench\u002Fpantry\u002Fskills\u002Fwayfinder\u002FSKILL.md",[],[]]