Understanding Ladder

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. Using the skill afterwards needs nothing extra for rungs 1 and 2 (plain text and diagrams, produced in chat); rung 3 saves a single self-contained HTML file that opens in any browser; rung 4 does not render video at all — it only writes a storyboard, and rendering it needs separate tools installed by me, only if I ever ask (see point 2). For Chinese output, rung 1 delegates to the separate plain-chinese skill, which is not part of this listing and is not installed here. Install "understanding-ladder" for me. It is an explanation-format playbook: a SKILL.md plus three reference files that classify what I want explained, pick the lowest rung of Karpathy's understanding ladder that fully answers — (1) plain controlled text, (2) diagram, (3) interactive HTML page, (4) explainer-video storyboard — then produce that rung, always starting from a checkable plain-text draft and always ending with a "What to check" list. Repository: https://github.com/ChangWenC/understanding-ladder (the skill is the skills/understanding-ladder/ folder at the pinned commit: SKILL.md plus references/level-2-diagram.md, references/level-3-html.md and references/level-4-video.md) Pinned version: install the code at commit 27b0957818f57a9c1d0ecdde1f50c2e3752f972c exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/ChangWenC/understanding-ladder into a temporary folder, check out commit 27b0957818f57a9c1d0ecdde1f50c2e3752f972c, and copy the skills/understanding-ladder/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 0a1078a2af927199a0108dc13a33a10e090174460c9ee5a1b0768a7160dd2355 - references/level-2-diagram.md: SHA-256 37567855ffd9d0517c02b9919ceaaf65618a997aa41bf03529db276733a75f71 - references/level-3-html.md: SHA-256 407bb4ff363843babcb6436097c110aeaaacc0051d00a193ab1d62c1c41b1eca - references/level-4-video.md: SHA-256 e02cfa94dd0318c65125cd3e8c2467c2164296733d0cbe57c4ac4f44542c61a2 If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md and all three reference files in full and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds four files only: SKILL.md (7,268 bytes), references/level-2-diagram.md (2,703 bytes), references/level-3-html.md (2,709 bytes) and references/level-4-video.md (4,099 bytes). There is no code, no script and no dependency list in the folder: the skill is instruction text only — a rung classifier, rung-1 writing rules, and one reference file per higher rung. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. The files name no account, no telemetry and no remote service to connect to at install time. API keys are mentioned only in the rung-4 tool table: showtime is noted as needing none, and the optional paid ElevenLabs voice as needing one (last bullet below). - At use time, the skill works on the topic or text I give it, replies in my language, and produces an explanation, not a data collection: it classifies the content, opens with one line stating the rung chosen, writes the rung-1 plain-text draft first, then produces that rung. Rung 1 is controlled text (about 80% of the way to ASD-STE100 in English; for Chinese it delegates to the separate plain-chinese skill, which this install does not include). Rung 2 is a diagram, by default a Mermaid block in chat, or a single-file SVG/HTML saved in the working directory when layout matters, followed by 2–4 sentences on how to read it. Nothing is sent anywhere for rungs 1 and 2. - Rung 3 writes one self-contained HTML file (CSS and JS inline, no npm, no build step) in the working directory, tells me its path, and opens it in a browser if the environment allows. The reference says to prefer plain JS with canvas or SVG; only if a chart library is truly needed does it load one from a CDN (it names Chart.js as the example) — so a rung-3 page is fully local except for that one conditional case, where opening the page would fetch the library from the CDN. Sample data must be labelled as sample data; the skill forbids inventing data to make a page look real. - Rung 4 does not render video and does not install tools: it delivers a storyboard (per shot: visual, rung-1 narration, duration), tool links, and a ready-to-paste prompt for those tools. The links point to showtime (github.com/FavioVazquez/showtime), Manim (www.manim.community), Kokoro (github.com), ffmpeg (ffmpeg.org) and ElevenLabs (elevenlabs.io); the reference also tells the agent to open the links and confirm the install steps before recommending a tool. If showtime 0.4.0 or later is already installed, the storyboard table is saved as storyboard.md and handed over with the command `showtime new <template> <dir> --from-storyboard storyboard.md`. ElevenLabs is the one paid option in the table and needs an API key — it is a separate service I would choose and use myself, outside this skill; the free routes (showtime, Manim, local Kokoro TTS) need no key. Do not install any of these tools during this install, and do not send anything to ElevenLabs. - Every output ends with a "What to check" section (1–3 things to verify), higher rungs add no facts that are not in the rung-1 draft, and the skill must not climb rungs to look thorough: if rung 1 fully answers, it stops there. The rest of the repository — the sibling plain-chinese skill, examples/, docs/ and the .claude-plugin/ manifests — is not part of this skill and is not installed; the upstream one-command routes (npx skills add ChangWenC/understanding-ladder, the Claude Code plugin, or copying skills/* by hand) would install both skills of the repo, which is more than this listing covers. 3. When I say "yes, install", install it as a skill: copy the whole understanding-ladder folder (SKILL.md plus the references/ subfolder, four files in total) into the agent's skills directory, in a folder named "understanding-ladder". Install only this folder from the pinned checkout, even though the repository contains a second skill. 4. Verify with no network calls: frontmatter valid, files in place. 5. Report what was installed, where, and what I still need to do myself (for example: install the sibling plain-chinese skill separately if I want its spelled-out rung-1 rules for Chinese, and install any rung-4 video tools myself, only if and when I want a storyboard rendered; using the skill is a separate step that only starts when I ask for it, and the use-time touches disclosed above — a rung-3 file saved locally, the conditional CDN library fetch, and following rung-4 tool links — happen only then, on topics I name). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.