I fed the 40k-star "How To Live Better" into my AI and it started doing the math with me
Skills used in this guide (6)
以证据为先的现况调研:要新事实、对比、补全信息或基于当下公开证据给出建议时使用。
npx skills add affaan-m/ECC -g就计划、决策或想法 relentless 地追问用户。当用户想压力测试自己的思考,或使用任何追问触发短语时使用。
npx skills add mattpocock/skills -g1-3-1 决策简报格式:一个问题、三个选项、一个选择。
npx skills add NousResearch/hermes-agent -g在 Obsidian 库里读取、搜索、创建与编辑笔记。
npx skills add NousResearch/hermes-agent -g把签证申请材料(图片)翻译成英文,并生成原文与译文对照的双语 PDF。
npx skills add affaan-m/ECC -g(占位)文本人性化/润色能力。
npx skills add blader/humanizer -g
The book is static, the AI is not. The book answers "which advice is worth doing"; a skill answers "does this one hold for the mess I'm in right now". Pack it, and the next time you're stuck you just ask — but don't expect it to decide for you.
1. Before you bookmark it: advice was never the thing you were missing
Late September, a Chinese repo appeared on GitHub called 《How To Live Better》. One month later: forty thousand stars.
My first reaction was the same as yours — another book of platitudes. Ten minutes in, it turned out not to be. It doesn't moralise, it does arithmetic. Thirty-four topics, 600+ entries, and every one of them states what you spend (money, time, effort), what you get back (years, money, freedom), how strong the evidence is, and where it came from. It even grades the evidence, and for the weak ones it says out loud: this is the author's judgment, not evidence.
That matters. Most life advice fails because it's only half an answer — it tells you what to do and never what it costs.
But a book has one built-in flaw: it's static. You bookmark it. Then what? It doesn't know you're 35, just laid off, three months behind on social insurance, rent due next week. It can only give you the general answer, and what you need is "does this hold for me, right now".
So this isn't a review of how good the book is. It's how to turn it into an advisor that knows your situation. Half an hour. Every prompt is copy-paste.
2. Find the original first: the one with ~40k stars
Once it went viral, five or six same-named repos appeared overnight. We hit the GitHub API on each while writing this (2026-10-05) — stars and last-commit dates fetched live, not recalled.
| Repo | Stars | Last push | What it is |
|---|---|---|---|
| eternity4719/HowToLiveBetter | 39,989 | 2026-10-04 | The original. 34 topics, 600+ entries, each with cost, benefit, evidence grade and source. CC-BY-4.0 |
| cdyforever/how-to-live-better | 7,290 | 2026-10-04 | Single-page reader for the whole book. Searchable, offline |
| HuiTurn/HowToLiveBetter | 20 | 2026-10-04 | WeChat mini-program build, 630 entries, offline |
| madkoding/HowToLiveBetter-CL | 8 | 2026-09-21 | Chilean Spanish adaptation, 568 entries |
Fastest tell: the original sits under an organization account (eternity4719), not a personal one. It was created 2026-09-07 and crossed 40k stars in a month.
One correction: the "37,000 stars" going around in video scripts was a few days old. It's 40k now. Repos like this move daily — don't anchor on a stale number.
If digging through GitHub is tiring, search it on Skills (skills.guanliai.com). It indexes 8,800+ agent skills with a page each, including every one we use below.
3. Step-by-step: turn it into an advisor you can call (prompts ready to copy)
1. Installing a skill is copying a folder
| Client | Directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Codex | ~/.codex/skills/ |
| Cursor | ~/.cursor/skills/ |
| Trae | ~/.trae/skills/ |
| CodeBuddy | ~/.codebuddy/skills/ |
| Lingma | ~/.lingma/skills/ |
2. Don't make it re-read the book every time — pack it
Most people get this first step wrong. They hand the book to the AI and say "read this and answer me". That's asking an intern to re-read six hundred pages from page one, every single time. Slow, and by the end they've forgotten the beginning.
Pack it instead, so only the relevant chapter gets opened:
Pack everything at https://github.com/eternity4719/HowToLiveBetter into an agent skill
under ~/.claude/skills/life-guide/. Requirements:
1. In SKILL.md, state when to invoke it (life decisions, health, money, legal,
moving abroad, career) and when not to (purely technical questions);
2. Split the 34 topics into 34 md files under references/, keeping every entry's
cost, benefit, evidence grade and original source. Drop none of the four;
3. Cite as "topic + entry number", e.g. "Medical 12", so I can check the original;
4. When an entry is graded C, you must say out loud: this is the author's
judgment, not evidence.
When done, tell me how many entries you indexed and which topics are largest.3. Make it go read the source for you
Every entry carries a citation, but papers get updated and policies move. Pick the one you're actually about to act on and make it check:
Use research-ops to verify "Medical 12" from life-guide:
- Find the original paper or official document it cites, with a link;
- Check whether anything published after 2025 overturns or narrows the finding;
- Separate clearly what the evidence supports from what is the author's
cost-benefit judgment;
- End with one line: should I act on this now, and why.
Primary sources only. No aggregators, no blog commentary.4. Get someone who argues back
What you're short of isn't advice. It's someone willing to say "you didn't count that". That's grilling — it comes at you one round at a time:
I'm going to keep paying social insurance myself, 1,800 a month, for 15 years.
Use grilling on me: one point at a time, wait for my answer before the next,
aim at the math I haven't done and the outcome I'm most afraid of. Stop after 8 rounds.Honestly, this part is often uncomfortable. Good. Left to yourself, you steer neatly around the question you're most afraid of.
5. Compress it into something you could send to family
Don't stop at "I'll think about it". What you can't think through, you can usually write through:
Write the social insurance question above as a 1-3-1 brief:
one question (what we're actually solving), three options (each with cost,
risk, worst case), one recommendation plus the reason. Under 400 words,
readable by a family member.6. Don't let the conclusion die in a chat log
The step everyone skips, and the one you'll regret in three months. A conclusion that isn't filed is a calculation you'll redo:
Write this conclusion into my Obsidian note "decisions/social-insurance":
the situation at the time, what I chose, the options I dropped and why,
and which life-guide entries it rested on. Remind me to revisit in a year.7. The going-abroad bit
When you have to submit documents, visa-doc-translate saves the back-and-forth:
Use visa-doc-translate on these 6 documents. Output a bilingual PDF with
original and translation side by side. Keep proper nouns (names, institutions,
ID numbers) untranslated.4. What the book can't cover, these can
One general book can't go deep on the specialised directions. Same procedure — API called on each, 2026-10-05.
| Area | Project | Stars | Last push | What only it does | When it's wrong |
|---|---|---|---|---|---|
| Health | DrCatHicks/informed-patient | 153 | 2026-10-04 | Produces an evidence review for one specific health question, not wellness tips | Emergencies — go to a hospital |
| Health | jmandel/health-skillz | 82 | 2026-06-08 | Connects to your own health record over SMART on FHIR and analyses it | Most hospitals outside the US don't expose FHIR |
| Mental health | zxc7563598/mental-health-companion | 9 | 2026-06-09 | Chinese-language emotional support with East Asian context, not a translated Western bot | Diagnosis or crisis — see a clinician |
| Legal | zubair-trabzada/ai-legal-claude | 1,766 | 2026-03-27 | Contract review, risk analysis, NDA drafting, outputs a red-line list you can act on | Unmaintained 6+ months; real disputes need a lawyer |
| Money | zubair-trabzada/ai-finance-claude | 75 | 2026-05-12 | Runs cash flow, debt and retirement together instead of staring at returns | Anything about a specific instrument — not investment advice |
| Decisions | itshussainsprojects/Claude-Council-Skill | 93 | 2026-07-16 | Convenes 7 expert personas to argue, surfacing objections you didn't have | Small calls — seven rounds is slow |
| Career | lauradfarbo/claude-career-skills | 10 | 2026-08-31 | Treats CV, positioning and portfolio as one asset set | You only wanted a one-page template |
| General | byoungd/up | 67,120 | 2026-10-04 | Han Xiankai's life-advancement guide, career and self-improvement leaning | You want evidence grades and citations |
If that's too many: health goes to informed-patient, contracts to ai-legal-claude, money math to ai-finance-claude, and anything you're unsure about gets a round of grilling first.
5. Straight talk: who this is for, who it isn't, and where you'll trip
For: people whose "read later" folder is a graveyard; people who keep re-deciding the same things about work, money, law and health. You don't have to be rigorous — willing to open the citation once is enough.
Not for: anyone who wants the AI to make the life decision. Every entry lists a cost and a benefit, which means the book assumes you're the decider. The AI can lay out the accounts and the evidence. It can't sign for you.
The good part's over. Now the ugly:
| Trap | Symptom | Way around |
|---|---|---|
| Wrong repo | You installed a fan build, entries missing, no sources | Look for the org account eternity4719 |
| Fields dropped in packing | Advice kept, evidence grade gone | Name each field in the prompt — step 2 does |
| Treating AI as the answer | You get a conclusion that merely sounds right | Make it state the evidence grade; C must be flagged |
| Stale citations | Papers and insurance rules both move | Run key entries through research-ops |
| Pack once, never touch | Redoing the same math in three months | File it in Obsidian with a yearly review |
| AI voice creeping in | The conclusion reads like a memo your family won't read | Finish with blader/humanizer, numbers locked |
6. What you're probably wondering (FAQ)
Which repo is the original?
eternity4719/HowToLiveBetter — organization account, CC-BY-4.0, verified 39,989 stars on 2026-10-05. The same-named ones are readers, mini-programs or regional adaptations built on top of it.
I can't code. Can I still do this?
Yes. Installing a skill is copying a folder. The only thing you type is the packing prompt in step 2, and that's copy-paste.
600+ entries — won't that be slow?
No. You're loading files, not context; only the matching topic gets read. What's slow is "read the whole thing, then answer" — which is exactly why step 2 splits it into 34 files.
Are there French or Spanish versions?
There are community adaptations: madkoding/HowToLiveBetter-CL is the Chilean Spanish one (568 entries). French, Russian and others circulate as community reading sites rather than branches of the original. I haven't checked those translations entry by entry — look at who translated it and which version before you cite it.
Can I act on this advice directly?
For health, check the evidence grade and whether your situation matches. Legal, tax and social-insurance entries are jurisdiction-bound; most assume mainland China rules, so re-check local rules before or after moving.
Why run humanizer at the end?
AI conclusions drift into "in conclusion" phrasing that you stop noticing and your family can't read. humanizer changes the voice, not the facts.
7. Sources and where to go next
- eternity4719/HowToLiveBetter (original)
- cdyforever/how-to-live-better (single-page reader)
- DrCatHicks/informed-patient
- zubair-trabzada/ai-legal-claude
- blader/humanizer
- Agent Skill open format spec
One small thing you can do right now: open Skills, search grilling, install it, and type in the thing you've been putting off deciding — then let it interrogate you for eight rounds. You don't need to read the book first. Getting stuck is how you find out what you were missing.
Bottom line
The book tells you which advice is worth doing; the skill tells you whether it holds for you. Pack the original, verify with research-ops, push back with grilling, compress with 1-3-1, file it in Obsidian — five steps, and your bookmarks folder finally does something.
References
Related Scenes
Ray Dalio built an AI clone of himself — it still cannot write your principles
The clone is fluent in Dalio. That is the product, and that is the ceiling. His book argues the opposite of what the clone does: a principle only fires if you wrote it. So don't ask an AI to recite his — make it interview you until the borrowed ones fall out.
Build a personal second brain with AI (notes/consolidation/planning)
Personal efficiency is not "busier" — it's offloading fragments to a reliable system: capture → consolidate → plan → review. Let AI do note cleanup, knowledge merge, planning breakdown; you only judge "keep or not, what's next", leaving your head for what truly needs thinking.