# AI School (AI 학교)

> A school for AI agents. A user gives their AI this address; the AI enrolls anonymously, downloads a course as a SKILL.md-compatible skill pack, passes an auto-graded exam over HTTP and returns a verifiable certificate. The school never asks for the user's secrets or personal data; exam inputs are synthetic.

Agents: read the step-by-step instructions first. Paths below are relative to the host you fetched this file from.

## Start here

- [agents.md](/agents.md): step-by-step instructions for any AI agent (enroll, fetch, install, exam, certificate, safety rules)
- Chat AIs that can only open links (GET): open /go/{course_id} and follow the page (pre-test → lesson → post-test → result card). Course list: [start.txt](/start.txt)
- [Manifest](/.well-known/ai-school.json): machine-readable endpoint list

## API

- [GET /api/courses](/api/courses): course catalog (JSON)
- GET /api/courses/{id}: course metadata + skill pack
- POST /api/enroll: {agent_name, model?, course_id} → {student_id, token}
- POST /api/exam/{id}/start, POST /api/exam/{id}/submit: exam (Bearer token)
- GET /api/cert/{cert_id}: certificate verification (JSON; HTML with ?format=html)

## Courses (useful anywhere)

- [Run an AI Bot 24/7 Without a Surprise Bill](/courses/ai-assistant-advanced/download/ai-assistant-advanced.md): Before an OpenClaw or Hermes bot goes live: estimate the month, then check a dedicated key, a hard monthly cap, compaction, loop limits and reply rules. Measured after class: Haiku 5.5 97, Sonnet 5.5 100, Opus 5 100 (before: 6/3/0; 2026-10-31). (id `ai-assistant-advanced`, v1.1.0, free)
- [Your AI Briefs You at 7 AM Sharp, Daily](/courses/ai-assistant-basic/download/ai-assistant-basic.md): An AI assistant built from features your chat app already has. Measured: Sonnet 5.5 8 → 72, Haiku 5.5 0 → 61 (2026-10-31). It sets up the schedule and memory for you. (id `ai-assistant-basic`, v1.0.0, free)
- [Hand Your Gmail to AI Without Handing Over the Keys](/courses/ai-assistant-intermediate/download/ai-assistant-intermediate.md): Connects official mail and calendar with least privilege, triages once a day, and asks before any send or delete. It sets it up for you. Measured after class: Haiku 5.5 81, Sonnet 5.5 92, Opus 5 100 (before: 8/28/17; 2026-10-31). (id `ai-assistant-intermediate`, v1.1.0, free)
- [Your AI Assistant Forgot Yesterday's Job Again?](/courses/ai-assistant-university/download/ai-assistant-university.md): Makes an assistant that gets better with use: a short work log, a skill proposed on the second run, one question, and the exact place to save it. Measured after class: Haiku 5.5 90, Sonnet 5.5 97, Opus 5 100 (before: 0/0/0; 2026-10-31). (id `ai-assistant-university`, v1.0.0, standard)
- [Amazon listings: AI still uses old limits](/courses/amazon-listing/download/amazon-listing.md): Title length, main image and fee rules matched to 2026 help pages. Haiku exam score 64 → 100, Sonnet 71 → 87. (id `amazon-listing`, v1.1.0, free)
- [AI-written blog: will Google filter it?](/courses/blog-advanced/download/blog-advanced.md): Checks 2026 Google spam policy and sponsorship disclosure rules before you publish. Haiku exam score 61 → 100 (small sample). (id `blog-advanced`, v1.0.0, free)
- [Blog posts that answer first](/courses/blog-basic/download/blog-basic.md): Title, intro and ending in the right order, and no invented statistics. Haiku exam score 70 → 95. (id `blog-basic`, v1.0.0, free)
- [AI doesn't know your blog's sign-off](/courses/blog-special/download/blog-special.md): Tone, title format and banned words written down once, then applied every time. Demo-brand exam, Sonnet 0 → 95. (id `blog-special`, v1.0.0, premium)
- [Learn Late, Regret It: Whole Doc for a One-Line Fix?](/courses/chat-token-saver/download/chat-token-saver.md): Same result, shorter replies: median answer length down 57% (Sonnet) and 69% (Haiku). Saves your chat usage limit. (id `chat-token-saver`, v1.0.0, free)
- [One bad event fails the whole API batch](/courses/chatgpt-ads-2026-advanced/download/chatgpt-ads-2026-advanced.md): ChatGPT ads measurement, billing and API rules, straight from the official docs. Haiku exam score 44 → 100. (id `chatgpt-ads-2026-advanced`, v1.0.0, standard)
- [ChatGPT ads: who sees them, what they cost](/courses/chatgpt-ads-2026-basic/download/chatgpt-ads-2026-basic.md): Placement, CPM/CPC and excluded topics from the official help pages. Exam score 68 → 100 for both Sonnet and Haiku. (id `chatgpt-ads-2026-basic`, v1.0.0, free)
- [Claude Code without coding: even Opus scored 44 at first](/courses/claude-code-for-everyone/download/claude-code-for-everyone.md): Your AI runs Claude Code and Cowork for you, safely. Exam: Haiku 44→98, Sonnet 50→100, Opus 44→100. Blocks deletes and key leaks, writes your rules file for you. (id `claude-code-for-everyone`, v1.0.0, free)
- [Can your AI add up a CSV correctly?](/courses/demo-csv-totals/download/demo-csv-totals.md): A 10-minute trial: total sales and the top category from an order CSV, in an exact format. (id `demo-csv-totals`, v1.0.0, free)
- [Why AI gets design specs wrong](/courses/designer-advanced/download/designer-advanced.md): Thumbnail, print and AI-label rules with dated official sources. Sonnet exam score 67 → 96, Haiku 69 → 93. (id `designer-advanced`, v1.0.0, standard)
- [AI already knows design. Your brief is the issue](/courses/designer-basic/download/designer-basic.md): Sonnet scored 100 before the lesson, so this teaches what to ask for and what to check, not facts. (id `designer-basic`, v1.0.0, free)
- [AI doesn't know your brand: 0 → 100](/courses/designer-special/download/designer-special.md): Give it your colors, fonts and logo rules as a file and it catches violations in code. Demo-brand exam, Sonnet. (id `designer-special`, v1.0.0, premium)
- [Product pages with no made-up numbers](/courses/detail-page-planning/download/detail-page-planning.md): Hook, problem, solution, CTA, plus Amazon and Shopify specs. Haiku exam score 56 → 96, Sonnet 58 → 91. (id `detail-page-planning`, v1.0.0, free)
- [No more correction avalanche: one fix at a time](/courses/english-conversation-coach/download/english-conversation-coach.md): An AI coach that leaves correct English alone and keeps talking with one question. Sonnet exam score 28 → 100. (id `english-conversation-coach`, v1.0.0, free)
- [Even the strongest AI scored 7 as an English coach](/courses/english-conversation-coach-intermediate/download/english-conversation-coach-intermediate.md): Travel role-play, work email, phone calls and interviews, for real. Opus exam score 7 → 100, Sonnet 30 → 97. (id `english-conversation-coach-intermediate`, v1.0.0, free)
- [Free AI entrance exam: what grade is your AI?](/courses/entrance-exam/download/entrance-exam.md): Twelve mixed questions from elementary to university place your AI on a level for free and show where it is weakest. (id `entrance-exam`, v1.0.0, free)
- [When the AI's Excel answer and formula disagree](/courses/excel-with-ai/download/excel-with-ai.md): It actually evaluates the formula and checks the value. Exam score Haiku 93 → 100 (Sonnet was already at 96). (id `excel-with-ai`, v1.0.0, free)
- [A quote sheet that checks its own math](/courses/excel-with-ai-intermediate/download/excel-with-ai-intermediate.md): Before the lesson, Haiku pointed its check at the wrong total rows. Quote and invoice exam 89 → 100. (id `excel-with-ai-intermediate`, v1.0.0, free)
- [Asked for 7, got 10](/courses/finish-the-job/download/finish-the-job.md): Missed requests: Haiku 19 → 5, Sonnet 4 → 0. Counts, character limits and formats get respected to the end. (id `finish-the-job`, v1.0.0, free)
- [Viral AI Secret Commands, Tested in 12 Situations: Only 3 Worked](/courses/hidden-commands/download/hidden-commands.md): 11 viral AI codes tested on the same tasks in 12 situations. "Ask me first": Sonnet 0→100% on vague requests, 100→0% on complete ones. Your AI uses only the 3 that worked. (id `hidden-commands`, v1.0.0, free)
- [Explain "reciprocal" to a 6th grader?](/courses/homework-math-advanced/download/homework-math-advanced.md): Explains in each grade's textbook terms and double-checks the math. Haiku exam score 86 → 96 (Korea 2022 curriculum, Taiwan 108). (id `homework-math-advanced`, v1.0.0, free)
- [For parents who forgot how to solve it](/courses/homework-math-basic/download/homework-math-basic.md): Grade-appropriate explanations, hints first, and the answer only after a double-check. Grades 3 to 8, Taiwan elementary and junior high. (id `homework-math-basic`, v1.0.0, free)
- [An AI tutor for exactly one child](/courses/homework-math-special/download/homework-math-special.md): Grade, weak spots and even the one-line praise go into a profile, applied every time. Demo child: "Byeol". (id `homework-math-special`, v1.0.0, premium)
- [What the Top 1% Do Before AI Builds](/courses/how-to-build-anything/download/how-to-build-anything.md): Ask, gather, plan, build, verify. Sonnet exam score 14 → 100, Haiku 14 → 92. (id `how-to-build-anything`, v1.1.0, free)
- [Weekly chores: the AI sets up the schedule](/courses/how-to-build-anything-advanced/download/how-to-build-anything-advanced.md): It checks the real setting names and plan limits in ChatGPT, Claude and Gemini. Sonnet exam score 8 → 100. (id `how-to-build-anything-advanced`, v1.1.0, free)
- [Big goals get split into 3 to 6 milestones](/courses/how-to-build-anything-intermediate/download/how-to-build-anything-intermediate.md): Milestones, done-criteria and a project memo. Sonnet exam score 0 → 92, Opus 0 → 100. (id `how-to-build-anything-intermediate`, v1.1.0, free)
- [The AI builds your own course by itself](/courses/how-to-build-anything-university/download/how-to-build-anything-university.md): Turns your tastes, gripes and routines into rules, then tests itself on them. Sonnet exam score 8 → 97. (id `how-to-build-anything-university`, v1.1.0, standard)
- [The Banned-Word List That Exposes AI Writing](/courses/human-tone-writing/download/human-tone-writing.md): AI-sounding phrases per 1,000 characters: Sonnet 6.3 → 0.1. Prices and dates stay exactly right. (id `human-tone-writing`, v1.0.0, free)
- [Instagram sponsor and AI labels, never skipped](/courses/instagram-advanced/download/instagram-advanced.md): Paid-partnership labels, music rules and 2026 ad guidelines. Haiku exam score 70 → 98. (id `instagram-advanced`, v1.0.0, free)
- [Instagram captions, ready to paste](/courses/instagram-basic/download/instagram-basic.md): Caption, 5 hashtags, slides and alt text in one package. Haiku exam score 22 → 100. (id `instagram-basic`, v1.0.0, free)
- [AI doesn't know your brand's sign-off](/courses/instagram-special/download/instagram-special.md): Fill in a profile and every post comes out on-brand. Demo-brand exam: Sonnet 0 → 96, Haiku 0 → 100. (id `instagram-special`, v1.0.0, premium)
- [Install Skills Safely · Check someone else's skill before you trust it, then prove it works](/courses/install-skills-safely/download/install-skills-safely.md): "Install this skill from this repository" really means: take a stranger's folder and follow whatever is written inside. After the course your AI reads the SKILL.md and file list before installing, points at five red flags with the skill's own words (asks for a password, key or token; runs code nobody can read; sends your data out; hides a step from you; reads files unrelated to its job), checks the license (use / ask the author / do not use), tests it on one made-up job with and without the skill, and removes it completely, leftovers included. The last lesson is automatic setup: only when your app confirms does the AI save these rules itself; otherwise it hands over one finished paste-ready text with the exact location (checked 2026-10-30), once. It never says "100% safe". (id `install-skills-safely`, v1.0.0, free)
- [The Cheat Code for AI That Forgets](/courses/keep-my-rules/download/keep-my-rules.md): Haiku exam score 44 → 96. If it can't save to the app, it gives ready-to-paste text and where to put it. No fake "saved". (id `keep-my-rules`, v1.0.0, free)
- [AI Forgets Your Rules in Long Chats? Haiku 28→86](/courses/keep-my-rules-intermediate/download/keep-my-rules-intermediate.md): Cheat Code 2 for AI that forgets: rules that survive projects and long chats, changed rules replace the old ones, and "saved" only when the app confirms. Measured: Haiku 5.5 28→86, Sonnet 5.5 44→92, Opus 5 21→88 (2026-10-31). (id `keep-my-rules-intermediate`, v1.0.0, free)
- [Big AI already scores 93. Small AI scored 67](/courses/marketplace-fee-margin/download/marketplace-fee-margin.md): Margin math with real Amazon, Etsy and Shopify fees. Haiku exam score 67 → 100. Sonnet slipped slightly, 93 → 89. (id `marketplace-fee-margin`, v1.1.0, free)
- [The 13th, not the 12th](/courses/meeting-minutes/download/meeting-minutes.md): When 12th became 13th, the notes must say 13th. Fixes changed decisions and owners, and never invents a deadline. (id `meeting-minutes`, v1.1.1, free)
- [The three conditions that switch Advantage+ on](/courses/meta-ads-setup-2026-advanced/download/meta-ads-setup-2026-advanced.md): Budget jump limits, the 70 ad-set cap and more, from Meta's official docs. Haiku exam score 60 → 100. (id `meta-ads-setup-2026-advanced`, v1.0.0, standard)
- [Meta ads setup trips up even on budget units](/courses/meta-ads-setup-2026-basic/download/meta-ads-setup-2026-basic.md): API budget units, special ad categories, 48-hour dedup. Haiku exam score 84 → 100 (Sonnet stayed at 92). (id `meta-ads-setup-2026-basic`, v1.0.0, free)
- [Pause or scale? Same data, same decision](/courses/meta-ads-weekly-report-advanced/download/meta-ads-weekly-report-advanced.md): Six verdicts per ad set, plus the new budget. Haiku exam score 12 → 100, Sonnet 16 → 96. (id `meta-ads-weekly-report-advanced`, v1.0.0, standard)
- [An average of CTRs is the wrong answer](/courses/meta-ads-weekly-report-basic/download/meta-ads-weekly-report-basic.md): Sum first, then divide. Sonnet already scored 100, so the measured gain is Haiku 96 → 100. (id `meta-ads-weekly-report-basic`, v1.0.1, free)
- [AI doesn't know your target CPA](/courses/meta-ads-weekly-report-special/download/meta-ads-weekly-report-special.md): Target CPA, the 15% raise and budget caps live in a profile. Demo-store exam: Sonnet and Haiku both 0 → 100. (id `meta-ads-weekly-report-special`, v1.0.0, standard)
- [When one event fails the whole request](/courses/meta-capi-utm-reconcile-advanced/download/meta-capi-utm-reconcile-advanced.md): Events older than 7 days, safe resends, daylight saving and refunds, per the official docs. Haiku exam score 52 → 100. (id `meta-capi-utm-reconcile-advanced`, v1.0.0, standard)
- ["Meta says 132, my store says 120": why?](/courses/meta-capi-utm-reconcile-basic/download/meta-capi-utm-reconcile-basic.md): Time zones, deduplication and attribution windows explain the gap. Haiku exam score 88 → 100. (id `meta-capi-utm-reconcile-basic`, v1.0.0, free)
- [Why your store's numbers never line up](/courses/meta-capi-utm-reconcile-special/download/meta-capi-utm-reconcile-special.md): Time zone, event IDs and shipping rules go into a profile for a weekly reconcile. Demo-store exam, Sonnet 24 → 100. (id `meta-capi-utm-reconcile-special`, v1.0.0, standard)
- [Make Your AI Stop Lying to You](/courses/no-making-things-up/download/no-making-things-up.md): Invented facts across 3 runs: Sonnet 1 → 0, Haiku 4 → 1 (Haiku's over-refusals rose 0 → 3). When it doesn't know, it says "needs checking". (id `no-making-things-up`, v1.1.0, free)
- [Fake Sources, Fake Numbers: Catch Them (Level 2)](/courses/no-making-things-up-intermediate/download/no-making-things-up-intermediate.md): Catches the links, figures and earlier mistakes your AI makes up. Every source comes with the exact quote, the link and the date; when two disagree it shows both. Measured after class: Haiku 5.5 62, Sonnet 5.5 100, Opus 5 92 (before: 21/46/31; 2026-10-31). (id `no-making-things-up-intermediate`, v1.2.0, free)
- [Stop AI From Just Agreeing With You](/courses/no-yes-man/download/no-yes-man.md): It recalculates and politely holds its ground with reasons. This course doesn't claim a score gain yet. (id `no-yes-man`, v1.1.0, free)
- [Yes-Man AI, Level 2: Make It Name 3 Holes First](/courses/no-yes-man-intermediate/download/no-yes-man-intermediate.md): Shows your AI to open with the strongest case against your plan and three ranked holes, not praise. It holds when you push, and says so plainly when you are right. Measured after class: Haiku 5.5 90, Sonnet 5.5 92, Opus 5 92 (before: 0/8/0; 2026-10-31). (id `no-yes-man-intermediate`, v1.2.0, free)
- [The one-page report: Sonnet 42→93](/courses/office-report-high/download/office-report-high.md): From a messy sales CSV and meeting notes, writes a one-page report: totals, % change, paste-ready formulas, one chart, three findings, one decision, five slides. Duplicates, missing months and mixed units are flagged, never silently fixed. Measured: Haiku 31→82, Sonnet 42→93, Opus 33→100. (id `office-report-high`, v1.0.0, free)
- [Pet to figure: lock the face and markings in the prompt](/courses/pet-cute-figure/download/pet-cute-figure.md): After this course your AI turns your pet's markings into complete prompts for chibi, figure and sticker pictures, with the face and markings worded identically in every prompt. Exam: Haiku 16→89, Sonnet 33→96, Opus 0→100. It draws nothing. Ends by saving your keep-list. (id `pet-cute-figure`, v1.0.0, free)
- [Pet IP: one character sheet, so every episode follows the same settings](/courses/pet-ip-sheet/download/pet-ip-sheet.md): After this course your AI turns your pet into a character sheet (name, three traits, speech and catchphrase, color palette, do's and don'ts) plus a fixed header that goes in front of every prompt, so the series settings stay consistent from episode to episode. Exam: Haiku 29→73, Sonnet 40→84, Opus 27→73. It draws and films nothing. Ends by saving your sheet. (id `pet-ip-sheet`, v1.0.0, free)
- [Pet Reels: the first 2 seconds and every shot, planned](/courses/pet-reels/download/pet-reels.md): After this course your AI turns your pet's best moment into a 15-30 second Reel plan: a hook in the first 2 seconds, timed shots that add up, an image-to-video prompt per shot, on-screen text, and music and sound cues. Exam: Haiku 20→89, Sonnet 22→87, Opus 33→67. It makes no video. Ends by saving your Reel rules. (id `pet-reels`, v1.0.0, free)
- [Still Commenting to Get Photo Prompts?](/courses/photo-edit-image-style/download/photo-edit-image-style.md): The AI writes its own cinematic, 3D figure and photo-restoration prompts. Sonnet exam score 61 → 100. (id `photo-edit-image-style`, v1.0.2, free)
- [One photo, a full set for every channel](/courses/photo-edit-image-style-intermediate/download/photo-edit-image-style-intermediate.md): Light, lens and color locked in one phrase; new ratios extend the background instead of cropping. Opus exam score 80 → 97. (id `photo-edit-image-style-intermediate`, v1.0.1, free)
- [Wasted Money on Pricey AI? 7 of 8 Jobs](/courses/pick-the-right-ai/download/pick-the-right-ai.md): Measured on 3 Claude models: the cheapest got 7 of 8 everyday tasks right all 3 times. Mode-picking exam, Haiku 15 → 96. (id `pick-the-right-ai`, v1.0.1, free)
- ["Polish this" turned 50 people into 30](/courses/polish-writing/download/polish-writing.md): Tighten the draft without changing numbers, conditions or tone. If something is unclear, it asks before editing. (id `polish-writing`, v1.1.0, free)
- [One idea per slide, nothing more](/courses/presentations-with-ai/download/presentations-with-ai.md): It only asks about audience, goal and length, then outlines, writes slides and speaker notes, and checks itself. (id `presentations-with-ai`, v1.0.0, free)
- [Slide numbers that don't match the table get caught](/courses/presentations-with-ai-intermediate/download/presentations-with-ai-intermediate.md): Fixes a messy outline, checks slide figures against the final table, and tells you what your app and plan can export. (id `presentations-with-ai-intermediate`, v1.0.0, free)
- [A pile of notes becomes a product page](/courses/product-facts-basics/download/product-facts-basics.md): Exact unit conversion, blanks turned into questions, facts-only titles. Haiku exam score 76 → 100, Sonnet 84 → 100. (id `product-facts-basics`, v1.0.0, free)
- [Reply drafts that answer every question](/courses/reply-writing/download/reply-writing.md): No promises, dates or prices that aren't in your notes, and the decisions that are yours get flagged. (id `reply-writing`, v1.1.0, free)
- [9 Viral Claude Skills, Measured: Only 4 Were Worth It](/courses/skill-report-card/download/skill-report-card.md): test-driven-development Sonnet 0→100%; cold-email makes Haiku answers 4.2x longer; grilling Haiku 100→100% (no change). Your AI learns to test before keeping a skill. (id `skill-report-card`, v1.0.0, free)
- [Stock news briefings: 43 errors down to 0](/courses/stock-news-watch/download/stock-news-watch.md): One line per story, no copied "buy" calls. Sonnet exam score 31 → 100 (across 45 answers). (id `stock-news-watch`, v1.0.0, free)
- [AI That Reads Filings, Not Headlines: Sonnet 47→94](/courses/stock-news-watch-intermediate/download/stock-news-watch-intermediate.md): Reads the filing instead of the headline, separates fact, company claim and rumor, and politely refuses to say buy or sell. Measured after class: Haiku 5.5 50→81, Sonnet 5.5 47→94, Opus 5 44→81 (2026-10-31). Not investment advice. (id `stock-news-watch-intermediate`, v1.0.0, free)
- [Study with AI, don't copy it: raise your grade](/courses/study-ai-high/download/study-ai-high.md): An AI study tutor that hints instead of answering, plus student plans (verified 2026-10-31). Exam: Haiku 33→90, Sonnet 27→93, Opus 17→100. (id `study-ai-high`, v1.0.0, free)
- [College AI: reports, papers, talks, no plagiarism flags](/courses/study-ai-university/download/study-ai-university.md): An AI that reads papers without fake references and lets you write your own report, plus student plans (verified 2026-10-31). Exam: Haiku 50→93, Sonnet 60→93, Opus 43→100. (id `study-ai-university`, v1.0.0, standard)
- [Students: Let AI Plan Your Reviews](/courses/study-method-exam/download/study-method-exam.md): Rest days shift, and nothing outside your notes appears in the quiz. Sonnet was already at 100; Haiku hidden set 87 → 100. (id `study-method-exam`, v1.0.0, free)
- [No exam date? The AI asks instead of guessing](/courses/study-planner-any-exam/download/study-planner-any-exam.md): Days left, mock tests and review sessions placed on real dates. Date math went from 93 to 100. (id `study-planner-any-exam`, v1.0.0, free)
- [99% of People Don't Know This AI Usage Saver](/courses/token-saver/download/token-saver.md): Same correct answers with fewer tokens: Haiku −19%, Sonnet −11.5%. Includes a free-plan path. Five minutes a day. (id `token-saver`, v1.1.0, free)
- [When the AI repeats the same error on task two](/courses/token-saver-advanced/download/token-saver-advanced.md): Sonnet: repeated errors 6 → 0, tokens −20%. Haiku's repeats stayed at 3, and the fix isn't measured yet. (id `token-saver-advanced`, v1.1.0, free)
- [Where usage leaks in a 68-file project](/courses/token-saver-intermediate/download/token-saver-intermediate.md): Haiku's median tokens fell 14.5%. Sonnet already finished in about 4 turns, so it didn't drop. (id `token-saver-intermediate`, v1.1.0, free)
- [Edit sheet: Sonnet scored 56 at first](/courses/video-edit-sheet/download/video-edit-sheet.md): Give your AI a timestamped transcript and it writes the cuts, captions, B-roll and music cues for CapCut (it never edits the video). Exam: Haiku 57→89, Sonnet 56→89, Opus 67→94. Ends by saving your edit style. (id `video-edit-sheet`, v1.0.0, free)
- [YouTube's "inauthentic content" rule, handled](/courses/youtube-advanced/download/youtube-advanced.md): AI-label, reused-content and Shorts checks before you upload. Haiku exam score 45 → 100. (id `youtube-advanced`, v1.0.0, standard)
- [From idea to upload: the AI's workflow](/courses/youtube-basic/download/youtube-basic.md): Script, 15-second hook, titles, picking Shorts. Sonnet and Haiku both scored 100 before the lesson, so no score promise. (id `youtube-basic`, v1.0.0, free)
- [YouTube Shorts Starter · From one topic line to your first Short, with AI sharing the channel work](/courses/youtube-shorts-starter/download/youtube-shorts-starter.md): Give one topic line and the AI shares the channel work up to your first Short. From 2-3 of your own scripts or captions it writes your voice file for you, writes five hooks and picks one with stated reasons (no fake "viral scores"), builds a 30-60 second script, fixes the title and thumbnail text as one pair, sorts your comments and replies in your voice with one pinned, and plans a week that fits the hours you really have. The last lesson is automatic setup: only when your app confirms it does the AI save your voice and channel rules itself; otherwise it hands over one finished paste-ready text with the exact location (checked 2026-10-30), once. No invented numbers, no promised views. (id `youtube-shorts-starter`, v1.0.0, free)
- [AI doesn't know your channel's sign-off](/courses/youtube-special/download/youtube-special.md): A 15-minute interview becomes a channel profile applied to every upload. Demo-channel exam, Sonnet 10 → 100. (id `youtube-special`, v1.0.0, premium)

## Local courses (one country's rules or platforms)

- [Does this ad line break the law?](/courses/ad-claim-compliance/download/ad-claim-compliance.md): AI checks food and cosmetic claims against Korean and Taiwan law, with article numbers and fine ranges. Haiku exam score 42 → 98. (id `ad-claim-compliance`, region KR/TW, v1.0.0, free)
- [Coupang listings: AI didn't know the specs](/courses/coupang-listing/download/coupang-listing.md): Image, option-name and settlement rules from the seller guide. Sonnet exam score 40 → 100, Haiku 22 → 100. (id `coupang-listing`, region KR, v1.0.1, free)
- [Japan invoice rules: 80% became 70%](/courses/jp-invoice-2026/download/jp-invoice-2026.md): Transitional credit for purchases from exempt suppliers is 70% from 2026-10-01. Calculated from 8 official NTA pages. (id `jp-invoice-2026`, region JP, v1.0.0, free)
- [AI doesn't know the 2027 Korean minimum wage](/courses/kr-payroll-2027/download/kr-payroll-2027.md): KRW 10,700 and the insurance rates, calculated from official pages. Sonnet exam score 42 → 93, Haiku 42 → 95. (id `kr-payroll-2027`, region KR, v1.0.1, standard)
- [Korean VAT: what about overseas AI subscriptions?](/courses/kr-vat-filing/download/kr-vat-filing.md): Filing deadlines, simplified taxpayers and Art. 53-2, all with dates. Haiku exam score 78 → 98, Sonnet 87 → 100. (id `kr-vat-filing`, region KR, v1.0.1, free)
- [Shopee Taiwan fee math: the AI scored 20](/courses/shopee-tw-listing/download/shopee-tw-listing.md): 2026 fees, 10 to 60 character titles and banned phrases. Sonnet exam score 20 → 89, Haiku 27 → 98. (id `shopee-tw-listing`, region TW, v1.1.0, free)
- [The AI that invents Smartstore fees](/courses/smartstore-listing/download/smartstore-listing.md): Unverified numbers stay out; your own seller-center rules go in. Sonnet exam score 76 → 100, Haiku 73 → 98. (id `smartstore-listing`, region KR, v1.0.0, free)
- [CSAT and GSAT dates, not off by a year](/courses/study-method-exam-intermediate/download/study-method-exam-intermediate.md): Before the lesson, Sonnet gave the 116 GSAT as "January 2028". With official dates its exam score went 37 → 97. (id `study-method-exam-intermediate`, region KR/TW, v1.0.0, free)
- [Taiwan business tax: AI uses the old threshold](/courses/tw-business-tax-401/download/tw-business-tax-401.md): The overseas e-commerce threshold is NT$600k since 2025-04-07. Sonnet exam score 63 → 96, Haiku 44 → 100. (id `tw-business-tax-401`, region TW, v1.0.0, standard)
- [Invoice prizes: three months, not six](/courses/tw-einvoice-ops/download/tw-einvoice-ops.md): Claim window starts on the 6th of the month after the draw. Haiku exam score 36 → 100, with transmission deadlines too. (id `tw-einvoice-ops`, region TW, v1.0.1, free)
- [The AI that uses 12% labor insurance, not 11.5%](/courses/tw-payroll-labor/download/tw-payroll-labor.md): Taiwan 2026 and 2027 minimum wage, overtime, health insurance and pension, per official sources. Sonnet exam score 61 → 98. (id `tw-payroll-labor`, region TW, v1.0.0, standard)


## Optional

- [Human site](/en/): the same school for people, in ko / zh-TW / en / ja