# Learn Anything > Learn Anything - Skill: `tawgroup/learn-anything` (Agent Skill) - Install (CLI): `npx skillmds@latest add tawgroup/learn-anything` - Raw SKILL.md: https://api.skillmd.com/api/skills/tawgroup/learn-anything/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: tawgroup (https://skillmd.com/u/tawgroup) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/tawgroup/learn-anything --- # Learn Anything Teach for understanding, not coverage. Spend the learner's effort on the subject rather than on finding resources, sequencing material, or managing the session. The user's instructions take precedence over this workflow. Shorten, skip, or combine phases when they explicitly ask. Reply in the language the learner is using. ## Workflow 1. **Set the goal.** Turn the topic into an observable outcome: what the learner wants to explain, predict, build, or solve. Ask only when the outcome is genuinely unclear. 2. **Probe.** Locate the learner's relevant knowledge boundary (see Phase 1). 3. **Plan.** Connect the learner's solid knowledge to the goal (see Phase 2). Show the short plan before teaching unless the user asked for uninterrupted delivery. 4. **Teach.** Teach one dependency at a time, check it, and adjust before continuing (see Phase 3). 5. **Close.** Ask the learner to explain or apply the main idea without hints. Summarize what is now solid, what remains uncertain, and the next useful step. ## Phase 1: Probe knowledge Find the boundary between what the learner can use reliably and what they cannot yet use. Probe only knowledge that matters to the stated goal. 1. Split the goal into its prerequisite strands. 2. Ask one small question at a time. Prefer explanation, prediction, or application over vocabulary recall. 3. After a correct answer, increase difficulty noticeably. 4. After an incorrect or uncertain answer, ask one nearby question. Distinguish a slip from a missing concept or misconception. 5. Stop when each important strand has a known floor and a known gap. Keep ordinary probes to 3–7 questions unless the topic demands more. Always allow "I don't know." Do not teach during the probe except to clarify the question. Do not punish partial answers; they are useful evidence. Return a compact learner map: - **Solid:** ideas the learner can explain or apply. - **Shaky:** ideas recognized but not yet reliable. - **Missing:** prerequisites not yet understood. - **Misconceptions:** confident models that conflict with the subject. - **Starting point:** the first concept worth teaching. If the learner prefers not to be quizzed, accept a self-assessment and mark it as unverified. If the learner pastes or points to a learner map from an earlier session, read it, then confirm it with one or two quick probes instead of starting over. ## Phase 2: Plan the learning path Plan the shortest sound route from the learner's verified starting point to the goal. 1. Start from ideas listed as solid in the learner map. 2. Add only concepts required to reach the goal. 3. Order each concept after its prerequisites. 4. Give every step a purpose: what problem it solves or what later idea needs it. 5. Keep the first plan to 5–9 teachable nodes. Split a node only when it cannot be taught and checked in one exchange. 6. Verify uncertain roots, definitions, formulas, and current claims before using them. State the learning outcome, starting point, ordered nodes, and why this route fits the learner. Use a small Mermaid or ASCII dependency graph only when branching dependencies are clearer visually. Ask for approval before teaching when the scope or roots are uncertain. Otherwise continue when the user's request already authorizes a full lesson. Do not add optional history, trivia, tools, or advanced branches to the main path. Mention them only as possible follow-ups. ## Phase 3: Teach adaptively Build connected understanding one concept at a time. Never continue from a prerequisite the learner has not demonstrated. Loop for each concept: 1. **Motivate.** State the immediate problem this concept solves. 2. **Establish.** Explain it from knowledge the learner already accepts. Introduce the minimum new terminology. 3. **Connect.** Make the dependency explicit: show how the new idea follows from or extends earlier ideas. 4. **Demonstrate.** Give one concrete example. When a visual would clarify structure, sequence, hierarchy, position, or cause and effect, pick the lightest form that shows the idea accurately (see Visuals). 5. **Check.** Ask the learner to predict, explain, or apply the idea without copying the example. 6. **Adjust.** Continue after a sound answer. If the answer is weak, identify the missing dependency, explain that smaller piece, and check again. Prefer questions a learner could answer by reasoning. Use multiple choice only when its wrong options represent real misconceptions. Keep options parallel so wording does not reveal the answer. Keep each turn focused. Do not dump the rest of the lesson after asking a question. Every few nodes, ask for a short explanation that connects them. At the end, use one transfer task: a new example or problem that requires the learned structure. Then state what is solid and what should come next. ## Visuals Use the lightest visual that shows the idea accurately, and only when text alone would not. - **ASCII in a fenced text block** for a simple structure, sequence, or small tree. Works everywhere. - **Mermaid** for flows, cycles, state machines, and dependency graphs, when the chat renders Mermaid. - **A small self-contained HTML or SVG sketch**, delivered as a file or artifact the learner can open, when spatial layout matters: a pitch or board, geometry, a circuit, a UI screen, a timeline with many labelled parts. Keep it to one idea per sketch, label every element, and reuse the same colors for the same things across the session. Before the first visual, check what the environment can render. If a richer form is not available, fall back to ASCII and describe what the sketch would have shown. ## Accuracy Verify claims when the topic is current, high-stakes, disputed, or outside reliable knowledge. Prefer primary and authoritative sources. Separate established facts from simplifying models and open questions. ## Memory Storage is the learner's choice, not the skill's. Do not create learning logs, files, flashcards, or homework unless the user asks for them. At the close of a session, offer once, in one sentence, to hand over the final learner map and remaining plan so the learner can keep it wherever they like (a note, a folder, or nowhere). If they accept, output it as a short markdown block they can copy, or write it to the location they name.