# An aviation maintenance manual's English standard, and how a Claude skill got 6000 stars · GitHub open source project

> Source: https://stringers.io/en/github/gh_danyuchn__asd-ste100-skill.html (Stringers / 什么值得做, generated 2026-11-08)
> Stringers answers one question every day: what small product can one or two people build in a few months and actually make money from. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these verdicts and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools today / hot / products / directions / direction / data. Setup: https://github.com/komako-workshop/stringers (the user logs in at stringers.io → "配置 agent" to get a personal URL). Site index: https://stringers.io/llms.txt

> levelsio (850k views) and Karpathy both posted telling people to make models talk in ASD-STE100. It's the earliest ready-made version, and its repo name matches the search term exactly, so the people relaying it and the directory sites all link to it.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-6

danyuchn/asd-ste100-skill · danyuchn

# An aviation maintenance manual's English standard, and how a Claude skill got 6000 stars

**What it is** : a Claude Code skill that uses the simplified English rules from aviation maintenance manuals (ASD-STE100) to rewrite convoluted English into sentences that say one thing and can't be misread. It also includes a checker script.

**Why it took off** : levelsio (850k views) and Karpathy each posted tweets urging people to have models talk in ASD-STE100. This was the earliest ready-made version, with a repo name that exactly matched the search term, so people relaying the trick and directory sites linked to it by default.

**C** **How you could ride it** : don't build another STE skill. Watch big accounts' prompt tricks, and within 24 hours ship a skill whose repo name and keywords match word for word, with a checker script, or port the freshly viral English trick into Chinese or Japanese versions.

C grade: the STE angle has already been pushed by two top accounts, copies are piling up, and growth has moved into the long tail. The name-grabbing approach is learnable, but the real force was a viral tweet nobody could schedule. An author who rebuilt it with a different standard would only get 157 stars. Publicly known / widely discussed

Data (updated daily) · Repo created 2026-07-20 · 6,182 stars · 7 days +298

GitHub · stars

Data 2026-07-30 → 2026-11-08 (102 days, 24 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

In 2026 Jul 20, a Taiwanese indie developer living in Bangkok, fed up with Claude's increasingly convoluted writing, turned ASD-STE100, the simplified technical English standard from aviation, into a Claude Code skill. 19 days later levelsio tweeted "Keep talking to me in ASD-STE100 with Claude," and two months after that Karpathy recommended the same trick. By Nov 6 the repo had 6099 stars, the most in its category. But what it won wasn't a feature. It was being a step ahead of everyone, with a name that happened to match what people were searching for. This piece lays out how it got going and whether you can still use the same approach.

### The line "Claude hasn't talked like a human since 4.7"

The author, Dustin Yuchen Teng (@danyuchn on Threads, 4200 followers, normally builds GMAT prep tools), wrote on his traditional Chinese blog in Jul 24: "Claude hasn't talked like a person since 4.7. I thought my English was just bad until I went to Reddit and found foreigners complaining too."

His fix came from aviation maintenance manuals. ASD-STE100 is a controlled English standard the European aerospace industry began using 1979 years ago: each word has one meaning, the active voice is used, simple tenses, each sentence states one thing, instructions are kept under 20 words, and semicolons are avoided. The goal is that mechanics for whom English is a second language won't misread a maintenance manual. Misreading one sentence could cost lives. The blog's example was:

> The valve should be rotated and tested before continuing.

Rewritten as:

> Turn the valve. Do a test of the valve. Continue the procedure.

He wrote these rules into a skill. The README's opening line framed it well: English meant for agents (tool descriptions, error messages, instructions agents pass to each other) is in the same situation as a maintenance manual. The reader can't go back and ask "do you mean X or Y?" The story was later retold repeatedly on LinkedIn and in blogs.

### How the stars grew

| Timeline | star | What happened back then |
|---|---|---|
| 7-20 | Repo created | The author himself using it |
| 7-24 to 7-30 | 77 | The author's blog, and Taiwanese users reposting on Threads |
| 8-08 | — | levelsio tweeted, but the tweet didn't link the repo |
| Early September | About 1780 (back-calculated from Trendshift) | A string of "tip of the week" LinkedIn posts and newsletter follow-ups |
| September | 10–30 per day | Plateau |
| 10-02, 10-03 | +439 and +483 in single days | Karpathy tweeted, and the repo hit #14 and #13 on Trendshift's site-wide daily chart that day |
| 10-12 | 4625 | Reviews rolled out one after another |
| 11-06 | 6099,357 fork | 30–60 per day; +7 over the last 291 days |

Neither of the two big pushes was something the author went out and found.

**levelsio, Aug 8.** The original tweet read "Save in memory to always use ASD-STE100 Simplified Technical English when you talk to me," followed by "Thank god!" According to his blog archive, the tweet got 850k views and 7009 likes. He'd picked up the trick from server-ops blogger @DanielLockyer. The tweet was just a one-line prompt and didn't mention any GitHub repo, but it turned "ASD-STE100 + Claude" into a search term. From Aug 10 to 12, a stream of "hack of the week: tell your agent Use ASD-STE100" posts appeared on LinkedIn. From late July to early September, stars rose from 77 to about 1780. There are no day-by-day figures for the middle stretch, so this wave is inferred.

**Karpathy, Oct 2.** He posted a short piece on how to read LLM output. It listed four formats, and the first tier was "have the model explain it in ASD-STE100." He added that the spec was too strict and that he sometimes just wanted "80% ASD-STE100." According to a screenshot from a summary posted in Oct 4, the tweet had 6.4M views at the time. The repo gained over four hundred stars that day and the next, the clearest causal jump on the whole curve. I didn't confirm whether Karpathy himself linked the repo, but people relaying his tweet on LinkedIn (such as Simar Singh) attached the repo directly.

Then came the long tail: skills.sh, tessl, skillsdirectory, claudemarketplaces, and dozens of other skill directory sites listed it. stashbase and andrew.ooo wrote reviews, and there are introductory articles in Japanese and Chinese too. It now gains thirty to fifty stars a day, mostly from these sources.

### Why this one, out of the same thing

Getting a model to write in STE takes one line of prompt. So from late July on, a large batch of similar projects appeared:

| Project | Launched | star(11-06) | Notes |
|---|---|---|---|
| danyuchn/asd-ste100-skill | 7-20 | 6099 | The subject of this piece |
| AminBlg/SimpleEnglish | 7-21 | 4246 | 7-30 made the HN front page, 363 points; skills.sh installs 14.5k, more than the subject's 9100 |
| 0xpili/simplified-technical-english | 8-16 | 1023 | Ships with a glossary and checking tools, almost a month later |
| dualface/ste-zh | 10-05 | 874 | Using STE principles to get agents to report in Chinese |
| samirsawarkar, MDsniper, DotDebian and other same-named repos | since late July | Dozens to hundreds | knockoffs |

There's also a zero-cost rival: on HN most people think a single line in CLAUDE.md is enough and a skill is redundant. PostHog and similar companies also keep a copy in their own repos.

What made it win over these, ranked by weight:

**First, the name is the search term.** It was created a day before AminBlg, who hit the HN front page, and 19 days before levelsio's tweet. The repo is named `asd-ste100-skill`, so when someone searches "ASD-STE100 skill" it ranks first; AminBlg's repo is called SimpleEnglish and holds no spot on that term. The one that hit the HN front page was later overtaken on stars, largely because of this (inference).

**It's more than a single prompt line.** It ships a deterministic check script, `ste-lint.py`, that flags passive voice, present perfect, phrasal verbs like spin up / reach out, overlong sentences, and semicolons. It has two tiers: Strict (for error messages and tool descriptions) and STE-flavored (for READMEs and PR descriptions). It also says outright that it does not copy the 900 words from ASD's official dictionary, which is copyrighted. A review article recommends it for exactly that reason: "you get a real check, not only an instruction" (stashbase).

**Actively maintained.** Trendshift shows 61 days with 29 merged PRs, and the README links to Hebrew and Indonesian derivatives, which looks like a genuine member of this category.

The best counterexample is the author: in Aug 13 they made `iso-24495-skill` with the same packaging, turning ISO's plain-language standard into a skill, and it has 157 stars so far. Without a big account behind a standard, the same craft doesn't get traction.

### Is there still room?

The STE angle is saturated. The gag was pushed by two top accounts, the subject was named in a dozen-plus blog posts and LinkedIn threads, dozens of directories list it, same-named knockoffs are everywhere, and growth has moved to a long tail of a few dozen stars a day. Another English STE skill would most likely land at the few-dozen level that samirsawarkar and the others reached.

The exception is changing audience: dualface/ste-zh was created in Oct 5 and does "having agents use STE principles to speak Chinese," reaching 874 stars in a month. It rides Chinese-language demand that followed the Karpathy wave, which suggests porting a freshly viral English trick to another language still leaves a gap, but the window is a few weeks.

### If it were you

**C.** Don't build another STE skill, that slot is gone. What's worth learning is two things it happened to get right: the name equals the search term, and it adds a runnable check script to the prompt. But its real push came from levelsio and Karpathy, which nobody can schedule. If the author switched to another standard and tried again, they'd get only 157 stars.

**Who it suits** : developers already writing skills with Claude Code or Codex who can produce, in a day or two, a skill with a check script and a well-written README, and treat it as a small bet on building a name, not expecting a hit every time.

**Who it doesn't suit** : anyone counting on one skill for thousands of stars as the main result; anyone without time to watch prompt-tip tweets on X every day; anyone making the Nth English STE version.

**How to do it** : watch posts from people like levelsio, Karpathy and Simon Willison where they say "this one prompt fixed X," and within 24 hours build a skill version, with the repo name matching that keyword letter for letter. Don't just submit a prompt paragraph: include a locally runnable check script and a before/after comparison table. The other path is language porting: when a trick goes viral in English, ship Chinese, Japanese or Korean versions within a week or two, as ste-zh did.

**How to get users** : don't post on HN yourself; let the name catch search traffic, then submit the repo to directories like skills.sh, tessl and claudemarketplaces. By the time LinkedIn and blog re-posters look for "the ready-to-install one," you need to already be listed.

**Deadline** : from a big account's tweet to the topic being flooded with knockoffs is roughly one to two weeks. This repo was created before the tweet, which is why it took the spot.

**Signs you misjudged** : ① three skills in a row that grab a name fail to pass 500 stars after a big account tweets about the topic, meaning this repo was luck, not a method; ② skill directories start ranking by install count instead of stars and search stops favoring repo names, so grabbing the name stops being valuable (AminBlg already has more installs); ③ models learn to stop the "convoluted speech" habit on their own and the whole category of demand disappears.

### Data and sources

- GitHub: danyuchn/asd-ste100-skill's README and daily stars (recorded from 9-26, 10-09 onward, through 11-06); rival star counts come from GitHub, 2026-11-06.

- Trendshift (trendshift.io/repositories/103505): daily star growth from 9-07 to 11-06, 10-02/10-03 site-wide daily chart rank, and merged PR count.

- levelsio tweet: views and likes on levels.io's archived page "This prompt finally stopped Claude from getting more and more unintelligible every day" (8-08).

- Karpathy tweet: max.nardit.com (10-04 screenshot, 6.4M views) and the explainx.ai write-up.

- Author's blog: Dustin's AI Lab 〈讓 Claude 說人話〉, 7-24.

- skills.sh install count (11-06), CLI installs only, excluding git clone.

- Hacker News: the AminBlg/SimpleEnglish post (7-30) and related comments.

- Reviews and re-posts: stashbase.ai, andrew.ooo, LinkedIn posts from Simar Singh, Huseyin Ugur Yildiz and others.

[GitHub repo ↗](https://github.com/danyuchn/asd-ste100-skill)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| AminBlg/SimpleEnglish | Open source | 2026-07-21 | 4,246 stars; skills.sh 14.5K installs; 7/30 HN front page, 363 points | Aimed at documentation writing, with Plain/Strict tiers, benchmarks and a Codex plugin; hit the HN front page but the repo name lacks ASD-STE100, so stars were overtaken, yet installs are higher |
| 0xpili/simplified-technical-english | Open source | 2026-08-16 | 1,023 stars; skills.sh 112 installs | Ships an approximate word list and check tool, nearly a month late |
| dualface/ste-zh | Open source | 2026-10-05 | 874 stars | Using STE principles to get agents to report in Chinese |
| blagoySimandov/asd-ste100-writer-skill and a batch of other same-named small repos | Open source | From ¥2026-07-30 | A few dozen to a few hundred stars each. | Mostly machine-generated word lists or pure prompts, no linter, and they didn't get the name |
| One-line prompt Use ASD-STE100 (written into CLAUDE.md / memory) | Other | 2026-08 (levelsio tweet) | levelsio tweet 850k views | Zero cost, and most on HN think it's enough; this repo adds patterns, a linter and rules for agent-facing text |
| Ruben Hassid's /ste skill | Other | 2026-08-09 | Distributed with a newsletter; the author says 800k readers | Zip on a netdisk, aimed at ordinary Claude.ai users, not competing for GitHub stars |
| PostHog writing-simplified-technical-english / humanlayer asd-ste100 | Big-company features | 2026 | Not verified | Used internally in a company repo, not promoted separately |
| De-AI-ifying skills such as blader/humanizer, pbakaus/impeccable, petergyang/no-ai-slop | Open source | 2025-11 to 2026-07 | 13k to 100k stars | Aims to make text read as human-written; this repo aims for unambiguous text that machines can read |

Where users come from

- Inference **Big X account: levelsio tweet** : 2026-08-08 'always use ASD-STE100 when you talk to me', 852,922 views 7,009 likes (levels.io archive); a string of follow-ups on LinkedIn afterward; this repo 7/30 77 stars → about 1,780 stars by early September. The tweet itself didn't link this repo.

- **Verified** **Big X account: Karpathy tweet** : 2026-10-02 listed 'explain in ASD-STE100' as the first tier for understanding LLM output, 469k views within hours (explainx.ai); this repo 10/2, 10/3 with a single-day +439 and +483, ranked #14/13 site-wide on trendshift

- **Verified** **LinkedIn re-posts** : Simar Singh's repost of Karpathy linked this repo directly; Huseyin Ugur Yildiz named this repo and AminBlg as the two with the most stars; posts from Anton Lobach, coach-kenlin and others link this repo directly

- **Verified** **Blog reviews/SEO** : stashbase.ai (10/4), andrew.ooo (10/12), max.nardit.com, tameshita.net (Japanese), aiproducthub.cn (Chinese), enterprisedna (8/10) all recommend or review this repo

- **Verified** **Skill-directory long tail** : skills.sh 9.1K installs; listed on dozens of sites including tessl, skillsdirectory, claudemarketplaces, typingmind and 360 skill hubs; currently gaining 30–60 stars a day

- **Verified** **Author's own channels (cold start)** : 7/24 the author's Traditional Chinese blog Dustin's AI Lab publishes posts; Threads @danyuchn has 4.2K followers, and Taiwanese users 8/6 reposted it

- Inference **Hacker News** : 7/30 the post that hit the HN front page was AminBlg/SimpleEnglish (363 points); this repo has no HN post

## danyuchn/asd-ste100-skill: why it took off

> Research date 2026-11-06. Channel: github. Sources: our GitHub database (repo_profile / repo_snap, including wayback backfill points), trendshift.io daily star gains, web search (Jina / Exa), HN Algolia. The GitHub API was rate-limited that day, so the README was read via r.jina.ai.

### One line

It turns ASD-STE100 (Simplified Technical English, STE), the controlled-English standard from aviation, into a Claude Code skill that rewrites English meant for agents to read (tool descriptions, error messages, instructions between agents) so it is unambiguous. It took off less because of the product than because it caught a **prompt meme pushed by big accounts** : 2026-08-08 Pieter Levels (@levelsio) tweeted "always make Claude speak in ASD-STE100," and 2026-10-02 Karpathy put "explain in ASD-STE100" among the four formats he uses to understand LLM output. When people search "ASD-STE100 skill," this repo is **the earliest one created (7/20), with a name that matches the keyword exactly** , so LinkedIn, blogs and skill directories link to it on the way past.

### 1. Data: how the stars grew

| Time point | Stars | Source |
|---|---|---|
| 2026-07-20 | Repo created | GitHub |
| 2026-07-30 | 77 | GitHub repo, wayback backfill points |
| Around 2026-09-07 | About 1,780 (inferred from trendshift 9/7–11/6 totaling +4,317) | trendshift.io |
| 2026-09-26 | 2,240 | GitHub repo, wayback |
| 2026-10-02 / 10-03 | Single-day +439 / +483, site-wide trendshift rank #14 / #13 | trendshift.io |
| 2026-10-12 | 4,625 (our daily peak: +317; trendshift logged +230) | github repo / trendshift |
| 2026-11-06 | 6,099 stars, 357 forks; last 7 days +291, 30–60 per day | GitHub repo |

- August was the first growth phase: 7/30 had only 77 stars, reaching about 1,780 stars by early September. No day-by-day data exists for August, so we can only infer it came mainly from search traffic after levelsio's tweet (see below).

- In September it gained only 10–30 per day, a plateau.

- 10/2 Karpathy tweeted about it; on the day and the next, it gained 400+, then 100–200 per day for the following week, slowly tapering off. This is the clearest causal spike.

- Now it gains 30–60 per day, a long tail driven by directory sites and search. Fork/star ratio is 5.9%, suspect is empty, and it doesn't look like star farming.

- skills.sh install counts: this repo's asd-ste100 skill has 9.1K installs (9.3K across the author's account); rival AminBlg/SimpleEnglish has 14.5K; 0xpili has 112. ([skills.sh](https://www.skills.sh/danyuchn/asd-ste100-skill), 2026-11-06) So **by actual installs it ranks second, by stars it ranks first** .

### II. Competitors (doing the same thing)

"Have the LLM write in STE" can be done with a single prompt, so since 2026-07 a wave of similar projects has appeared, mostly skills / prompts:

| Name | Form | Launched | Scale (2026-11-06) | How it differs from this repo |
|---|---|---|---|---|
| AminBlg/SimpleEnglish | Open-source skills / plugins | 2026-07-21 | 4,246 stars, 14.5K installs on skills.sh; 7/30 hit the HN front page with 363 points and 121 comments | Targets "writing docs", with Plain / Strict modes, a built-in benchmark (78% fewer violations), and Claude Code + Codex plugins; it reached the HN front page but was overtaken on stars by this repo |
| 0xpili/simplified-technical-english | Open-source skills | 2026-08-16 | 1,023 stars | Ships an approximate word list and a checker tool |
| dualface/ste-zh | Open-source skills | 2026-10-05 | 874 stars | Uses STE principles to have agents report **in Chinese** |
| blagoySimandov/asd-ste100-writer-skill | Open-source skills | 2026-07-30 (Show HN 3 points) | Not in the top tier | Has a machine-generated word list; an issue pointed out 12 errors |
| NikolaRHristov/STE-Code | Open source | 2026-07-31 (Show HN 4 points) | Small | Aimed at code |
| menonpg/ste-prompt, samirsawarkar/asd-ste100-skill, MDsniper/asd-ste100-skill, DotDebian/asd-ste100-skill, nuelcyoung/asd-ste100, JAICHANGPARK/ASD-STE100, Ivapo/asd-ste-100-output-style | Open source | From ¥2026-07 | Small | Most are same-name or near-name copies |
| PostHog "writing-simplified-technical-english", the asd-ste100 skill in the humanlayer repo | Skills inside company repos | — | — | Large companies built one in themselves |
| Ruben Hassid's /ste skill | Cloud drive zip (distributed via a newsletter) | 2026-08-09 | Claimed to be written for 800k readers | Aimed at ordinary Claude.ai users, not on GitHub |
| One-line prompt "Use ASD-STE100" | Users write it into their own CLAUDE.md / memory | From ¥2026-08 | levelsio tweet 850k views | Zero cost; on HN most people think one line is enough |
| Broad "de-AI-ify" skills: blader/humanizer (65K stars), pbakaus/impeccable (100k stars), petergyang/no-ai-slop (13k stars), miqdadbadjuber/anti-slop (8.3K) | Open-source skills | 2025-11 to 2026-08 | Far larger than this repo | Goal is "sounds human-written"; this repo's goal is "unambiguous, for machines to read" |

Takeaway: lots of competitors, this isn't a blue ocean. It ranks first on stars within the STE niche.

### 3. Where users come from (growth levers)

1. **levelsio's tweet (2026-08-08, confirmed as the trigger, but it doesn't directly link this repo)** : "Save in memory to always use ASD-STE100 Simplified Technical English when you talk to me", 852,922 views, 7,009 likes (per [levels.io archive](https://levels.io/stop-claude-from-getting-unintelligible)). Then on LinkedIn 8/10–8/12 a string of "hack of the week: tell your agent Use ASD-STE100" posts appeared (Jeremy Caine, Oliver Benns, Michal Niec, etc.), and Ruben Hassid's newsletter 8/9 followed up. This wave turned "ASD-STE100 + Claude" into a search term.

2. **Karpathy's tweet (2026-10-02, confirmed)** : four formats for understanding LLM output, the first tier being "explain in ASD-STE100 / 80% of the way". Per [explainx.ai](https://explainx.ai/blog/karpathy-understand-llm-outputs-ste100-diagrams-html-video-2026) it got 469k views within hours. This repo gained 400+ on the day and the next, and ranked 13–14 on trendshift's site-wide chart. On LinkedIn, Simar Singh relayed Karpathy and attached this repo's link directly ([post](https://www.linkedin.com/posts/simar-singh-4b7617211_github-danyuchnasd-ste100-skill-asd-ste100-activity-7511925077710626816-iFjm)); a post by Huseyin Ugur Yildiz listed this repo and AminBlg as "the two with the most stars". Whether Karpathy himself linked this repo is unverified (none of the relays mention him linking GitHub).

3. **Follow-on content and reviews (confirmed)** : stashbase.ai (10/4), andrew.ooo review (10/12, which says "on GitHub weekly trending today"; we only captured daily charts, so unverified), max.nardit.com, tameshita.net (Japanese, 10/29), aiproducthub.cn (Chinese, 8/10), enterprisedna (8/10), and several YouTube overview videos. Most list this repo and AminBlg as "the two representatives".

4. **Long tail from skill directories (confirmed)** : skills.sh, tessl, skillsdirectory, claudemarketplaces, typingmind, hotskills, 360 security skill hub, and dozens of other aggregator sites list it. The current 30–60 stars per day most likely come from these sites and search.

5. **Author's own channels (confirmed, small volume)** : author Dustin Yuchen Teng (Threads @danyuchn, 4.2K followers, a Taiwanese indie dev based in Bangkok who builds GMAT prep tools) 7/24 published a post on his Traditional Chinese blog [Dustin's AI Lab](https://www.agentcrew.cc/blog/posts/zh/asd-ste100-claude-plain-english-skill/), and Taiwanese users shared it on Threads (Eric Lee 8/6). This is the cold-start stage, 0→77 stars, not the volume that came later.

6. **HN (negative)** : On 7/30, the HN front-page item was AminBlg/SimpleEnglish (363 points), not from this database; there were no posts from this database on HN.

### 4. Why it took off

- **Timing** : Mid-2026, a lot of people started complaining that Claude (after Opus 5 / 4.7) talks in more and more convoluted ways, with more and more jargon. An HN comment: "since Opus 5 I'm much more often staring at the wall of text". levelsio and Karpathy both pushed STE as the fix, and this repo is a plug-and-play version of it.

- **Earliest, with a name that exactly matches the search term** : 7/20 created the repo one day before AminBlg's HN hit, and 19 days before levelsio's tweet. The repo is called `asd-ste100-skill`, so anyone searching "ASD-STE100 skill" finds it first; AminBlg is named "SimpleEnglish" and doesn't own that term in search. This is probably the main reason it overtook AminBlg, which had the HN front page, on stars (inference).

- **The positioning fits in one sentence** : "the English used in aircraft maintenance manuals, applied to agent-to-agent communication". It has a story (a misread sentence can kill people, so an agent misreading gets no one to ask), which suits LinkedIn posts and blog relays.

- **There's some real substance here** : it ships a deterministic linter (`scripts/ste-lint.py`) that checks for passive voice, present perfect, phrasal verbs, long sentences, semicolons and more, in two modes, Strict and STE-flavored. It also says outright that it does not copy the official ASD 900 word dictionary (copyright), which is why reviewers are willing to recommend it ("you get a real check, not only an instruction" — stashbase).

- **Ongoing maintenance** : trendshift logged 61 days with 29 merged PRs, and the README cross-links Hebrew and Indonesian derivatives, which gives the impression that this is the authentic version of the category.

### 5. Differentiation: is it why it broke out?

- Compared with similar projects, it differs in: ① the target is text **meant for agents to read** (tool descriptions, error messages, system prompts), not replies meant for humans; ② it has a deterministic linter; ③ two modes; ④ it's the earliest, with the most direct name.

- But the mainstream view on HN is "a one-line prompt is enough, the skill is redundant". On installs, AminBlg (14.5K) also beats this repo (9.1K). So **product differentiation is not the main driver** ; the main drivers are **big accounts creating demand + it ranking first on that search term + relayers linking it in passing** . The linter and similar features are secondary reasons that made reviewers willing to recommend it and helped it hold first place.

- The author tried to replicate it: 8/13 built a similar-approach `iso-24495-skill` (ISO 24495 plain-language standard), which has only 157 stars. Without a standard backed by a big account, the same packaging gets no traction. This is the best counterexample to "it's timing and big accounts".

### 6. How many people see it

- It doesn't make money; there is no public revenue.

- Already widely discussed: the STE meme was pushed by levelsio (850k views) and Karpathy (hundreds of thousands of views), and this repo was named in a dozen-plus blog posts / LinkedIn posts and dozens of directory sites. It's already crowded.

- Copycats: many. Same-name clones (samirsawarkar, MDsniper, DotDebian's asd-ste100-skill), language ports (Hebrew g4lb, Indonesian jshsinaga, Chinese dualface/ste-zh with 874 stars), company-built versions (PostHog, humanlayer).

### VII. What others can borrow

- **"Packaging an existing industry standard as an agent skill"** is a replicable pattern (enterprisedna calls it "domain standards as a skill is turning into a repeatable pattern"), but it depends on the standard **happening to be pushed by a big account** ; standards without one (ISO 24495) get built and nobody looks.

- **Claim the name early** : when a meme first appears, occupy the position with a repo name **identical to the search term** ; most of the later traffic lands on you, which lasts longer than a single HN front-page spike.

- **Add a deterministic check script** so a "just a prompt" skill has something verifiable, which gives review writers a reason to recommend you.

- Watch the prompt-tip tweets from big accounts (levelsio, Karpathy) and ship a skill version within 24 hours; that is a steady source of stars.

### VIII. Unknowns

- No day-by-day data exists for the 7/30–9/6 stars (the 1,700 star period); that levelsio's tweet was the main cause is inference; whether anyone in its reply section directly posted this repo's link is unverified.

- Whether Karpathy's tweet itself linked this repo is unverified (none of the relays mention it).

- andrew.ooo says 10/12 was on GitHub weekly trending; we didn't capture the weekly Python chart for those days, so unverified.

- The 9.1K installs come only from skills.sh CLI anonymous telemetry and exclude git clone installs.

### Sources

- This repo's README: [github.com/danyuchn/asd-ste100-skill](https://github.com/danyuchn/asd-ste100-skill)

- Daily trendshift data: [trendshift.io/repositories/103505](https://trendshift.io/repositories/103505)

- HN (AminBlg post): [news.ycombinator.com/item?id=49114639](https://news.ycombinator.com/item?id=49114639); HN Algolia search ASD-STE100

- levelsio:[levels.io/stop-claude-from-getting-unintelligible](https://levels.io/stop-claude-from-getting-unintelligible)

- Karpathy post, as relayed by: [explainx.ai/blog/karpathy-understand-llm-outputs-ste100-diag…](https://explainx.ai/blog/karpathy-understand-llm-outputs-ste100-diagrams-html-video-2026) and [max.nardit.com/articles/karpathy-understanding-llm-outputs](https://max.nardit.com/articles/karpathy-understanding-llm-outputs)

- LinkedIn: Simar Singh, Huseyin Ugur Yildiz, Michal Niec, Jeremy Caine and others (see links above)

- stashbase:[stashbase.ai/blog/asd-ste100-claude/](https://stashbase.ai/blog/asd-ste100-claude/) ;andrew.ooo:[andrew.ooo/posts/asd-ste100-skill-review-simplified-technica…](https://andrew.ooo/posts/asd-ste100-skill-review-simplified-technical-english-for-agents/)

- Author's blog: [www.agentcrew.cc/blog/posts/zh/asd-ste100-claude-plain-engli…](https://www.agentcrew.cc/blog/posts/zh/asd-ste100-claude-plain-english-skill/)

- skills.sh: [www.skills.sh/danyuchn/asd-ste100-skill](https://www.skills.sh/danyuchn/asd-ste100-skill) and [www.skills.sh/AminBlg/SimpleEnglish](https://www.skills.sh/AminBlg/SimpleEnglish)

- Ruben Hassid:[ruben.substack.com/p/how-to-deslop-claude-in-2-words](https://ruben.substack.com/p/how-to-deslop-claude-in-2-words)

- Our GitHub repo: repo_profile / repo_snap (2026-11-06)

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.