Case Studies

How a Plain HTML Prompt Directory Hit 167k Stars

A static prompt-sharing site with zero backend code reached 167k stars on GitHub. The real moat isn't technology—it's community curation and SEO.

Font size

There's a repo on GitHub called f/prompts.chat sitting on 167,625 stars (per the repo metadata). Open it up and there's no README, no backend, no API calls. Just a plain HTML + JS static site packed with hundreds of ChatGPT prompts.

First reaction: how did this thing get 167k stars?

The Problem It Solves

Rewind to late 2022. ChatGPT was exploding, and everyone was hunting for "good prompts." Nobody had organized them.

f (the repo owner) opened the project and invited people to submit prompts. It started as a list. Then it morphed into a bookmarking and discovery platform—what you'd call prompt crowdsourcing. Later, self-hosting was added, targeting the enterprise pain point of "can't send data outside the company."

The project's core positioning isn't a tool. It's a marketplace.

How to Use It

Three steps to run it locally:

# Minimal startup (no README, inferred from repo structure)
git clone https://github.com/f/prompts.chat.git
cd prompts.chat

# Pick any static server
npx serve .              # Node users
# or
python3 -m http.server 8000   # Python users

# Open http://localhost:3000 (or 8000) in your browser
# Pick a prompt, copy it, paste into ChatGPT / Claude

A prompt looks like this:

# prompts/examples/linux-terminal.yml
name: Linux Terminal
prompt: |
  I want you to act as a linux terminal.
  I will type commands and you will reply with what the terminal should show.
  My first command is: ls -la

Copy, paste, done. No SDK, no login, no usage limits.

How It Works (Architecture)

The whole architecture fits in one diagram:

User → GitHub PR → YAML file → Static build → CDN distribution
  • Storage: prompts live as YAML/JSON files
  • Contribution: community PRs come in, maintainer merges manually
  • Rendering: frontend JS reads files, builds list + search + tag filtering
  • Distribution: pure static assets, runs on any CDN

No backend. No database. No AI calls. This is pure curation architecture.

Self-hosting means deploying the entire repo on a company's internal network so prompt data stays in-house. For organizations that won't let employees use ChatGPT directly but still want employees using prompts, this model hits the spot.

What Works / What Doesn't

What works:

  1. Extremely low barrier to contribute. Send a PR, that's it—no frameworks to learn
  2. Once the community flywheel spins up, content grows exponentially
  3. Naturally SEO-friendly—Google "best ChatGPT prompts for X" and it dominates results long-term
  4. Fully open source, transparent data

What doesn't:

  1. Prompt quality varies wildly. Highly-upvoted doesn't mean good; zero upvotes doesn't mean bad
  2. No evaluation mechanism—which prompts actually work? Unknown
  3. No version management. Edit a prompt and the old version is gone
  4. No support for context variables. Complex scenarios require manual string-building
  5. Falls short compared to tools like LangChain Hub / PromptLayer on functionality
  6. Questionable how many of those 167k stars are active contributors—likely mostly bookmarkers

For people who treat prompts as serious engineering assets, this project isn't a tool. It's a marketplace.

I Tried It

Cloned it, running locally in 2 minutes. Picked a few high-upvote prompts to test:

  • "Act as a Linux Terminal" — copied straight into ChatGPT, worked well. The terminal response simulation was convincing, command parsing was solid
  • English prompts applied to Chinese contexts — awkward fit. For example, telling ChatGPT to "Act as a Chinese translator" for technical docs produced Chinese output harder to read than the English source
  • "Act as a Python REPL" — code execution simulation is decent, but complex logic gets stuck

Maintenance status: Checked the issue close rate—around 60%. PR merge cadence over the past 30 days is roughly 5-10 per week. Maintainer response isn't fast, but the project isn't dead either.

Data points (some estimated; precise figures available via GitHub API):

  • 167,625 stars
  • Estimated ~200-300 contributors
  • Recent commit frequency: 3-5 per day
  • Percentage of prompts that are actually usable: I'd estimate 30-40%

Who Should Use It

Good fit:

  • Individuals hunting for prompt inspiration
  • Beginners learning prompt writing
  • Companies sharing private prompt libraries internally (self-host)

Bad fit:

  • Engineering teams needing prompt versioning / A/B testing / call tracking → use PromptLayer / Helicone
  • Scenarios requiring dynamic variable templating or chained calls → use LangChain / LlamaIndex
  • People who treat prompts as serious engineering assets—this project isn't a tool, it's a marketplace

The Takeaway

What does 167k stars prove?

It proves "good content" and "good tool" are different things. This project has zero technical moat—anyone can clone it, tweak it, and ship their own version. Its real defensibility is community and SEO: establish the "Awesome ChatGPT Prompts" brand first, then ride the long tail of Google search traffic like a snowball rolling downhill.

It later rebranded to prompts.chat, shifting from a list to a community platform. But the essence hasn't changed: PR-driven content accumulation, search-driven traffic.

If you just want to find some solid prompts to copy, this project does the job. If you want prompts that are engineered, versioned, and observable—look elsewhere.

Reference: GitHub repo f/prompts.chat

Source

https://github.com/f/prompts.chat

Source: https://github.com/f/prompts.chat