Forem: Sarvesh

OpenClaw's Skills System: The npm Moment for Personal AI

Sarvesh — Sun, 26 Apr 2026 02:35:07 +0000

This is a submission for the OpenClaw Writing Challenge

OpenClaw's Skills System: The npm Moment for Personal AI

In 2010, npm didn't just give JavaScript a package manager. It gave
developers a shared language for building on top of each other's work.
Suddenly, you didn't reinvent authentication, date formatting, or HTTP
clients — you just installed them. The ecosystem exploded overnight.

I think OpenClaw's Skills system is having that exact same moment.
Right now. And most people are sleeping on it.

What Are OpenClaw Skills, Really?

On the surface, Skills sound simple: a directory with a SKILL.md file
containing metadata and instructions. That's it. No compiled binaries.
No complex plugin APIs. Just Markdown that tells the agent what it can
do and how to do it.

But that simplicity is the point.

Because Skills are plain text, anyone can write one. A developer, a
healthcare professional, a founder who's never touched a terminal — if
you can describe a workflow clearly, you can build a Skill. OpenClaw
handles the execution.

Here's what that looks like in practice:

my-skill/
└── SKILL.md   ← instructions, metadata, tool usage

That's the entire surface area. Compare that to building a VS Code
extension, an Obsidian plugin, or an npm package with proper TypeScript
types, tests, and a publish pipeline. The barrier to contribution just
dropped through the floor.

The npm Parallel

When npm launched, it solved two problems at once:

Discovery — find packages others built
Composability — chain them together into something bigger

OpenClaw Skills solve the same two problems for AI agents.

ClawHub already hosts over 13,000 community Skills. There are Skills
for web scraping, CRM integration, calendar management, health tracking,
GitHub automation, and dozens of domains in between. You don't build
from zero — you install, compose, and extend.

And critically, just like npm packages, Skills are composable. A
well-designed Skill has clean inputs, predictable outputs, and defined
failure modes. Chain them together and you get a workflow that would
have taken weeks to build manually.

One commenter in the OpenClaw community put it well:

"Think hard about skill composability. A skill that chains cleanly
into other skills tends to be far more useful in real agent workflows
than a monolithic 'do everything' skill."

That's npm wisdom applied to AI agents. It took the JavaScript
ecosystem years to learn that lesson. The OpenClaw community is
learning it in months.

Why This Is Different From Every Other "Plugin System"

Every AI product eventually builds a plugin system. ChatGPT had
plugins. Claude has tools. Dozens of agent frameworks have their own
extension APIs.

What makes OpenClaw Skills different is ownership and portability.

Your Skills live on your machine. They're not locked inside a vendor's
marketplace, subject to approval processes or deprecation notices. You
can write a Skill today, share it as a GitHub Gist, and someone on the
other side of the world can install it tomorrow — no platform
intermediary required.

This is the open-source ethos applied to AI capabilities. It's what
made npm work. It's what made the Linux ecosystem work. And it's
exactly what the personal AI space has been missing.

The Inflection Point

npm crossed 1,000 packages in 2011. It crossed 1,000,000 in 2016. The
growth wasn't linear — it was exponential, because each new package
made the ecosystem more useful, which attracted more developers, which
produced more packages.

OpenClaw Skills are at the 2011 moment. 13,000+ Skills sounds like a
lot, but it's nothing compared to where this is heading. The primitives
are right. The community is energized. The barrier to contribution is
genuinely low.

If you've been watching OpenClaw from the sidelines, the Skills system
is the reason to stop watching and start building. Not because the
agent runtime is impressive — it is — but because the ecosystem you
contribute to today will compound in value for years.

That's the npm lesson. And OpenClaw is living it in real time.

EcoChat — AI Sustainability Advisor for Earth Day 🌿

Sarvesh — Sun, 19 Apr 2026 06:52:08 +0000

This is a submission for Weekend Challenge: Earth Day Edition

What I Built

EcoChat is an AI-powered sustainability chatbot that gives practical,
actionable eco-friendly advice tailored to students and residents of
Coimbatore, Tamil Nadu, India.

Most sustainability apps are built for Western audiences — tips like
"drive an electric car" don't apply here. EcoChat knows about local
Coimbatore markets, Tamil Nadu government schemes, and campus life at KPRIET.

🌍 Live Demo: https://ecochat-rho.vercel.app

💻 GitHub: https://github.com/sarveshm555/EcoChat

Demo

Try it live at https://ecochat-rho.vercel.app

Get a free Gemini API key at aistudio.google.com
Paste it in the bar at the top
Ask "How do I reduce plastic at home?"
Upload a photo of any trash item or product label 📷
Click "I'll commit to this!" and watch the counter go up ✅

Code

GitHub: https://github.com/sarveshm555/EcoChat

Tech Stack:

React + Vite — frontend
Google Gemini 2.5 Flash — AI responses + image analysis
Vercel — deployment

How I Built It

Google Gemini 2.5 Flash — used in 3 ways

1. Text chat with hyper-local context

The secret sauce is the system prompt:

"You are EcoChat, a sustainability advisor for students in Coimbatore,
Tamil Nadu. When relevant, mention local initiatives, Tamil Nadu
government schemes, and campus-specific suggestions."

This makes every response feel genuinely useful rather than generic.

2. Multimodal image analysis

Users can upload a photo of trash or a product label. Gemini identifies
whether it's recyclable, how to dispose of it properly, and mentions
local Coimbatore recycling options.

3. Dynamic Earth Day tip generator

On every page load, a separate Gemini API call generates a fresh,
surprising eco fact relevant to India — so the app always feels alive.

Key Features

💬 Multi-turn AI chat with Coimbatore-specific knowledge
📷 Image recognition for trash and product labels
✅ Commitment tracker — "I'll commit to this!" button with live counter
💡 Gemini-generated Earth Day fact on every page load
📱 Fully mobile responsive

Why hyper-local?

Generic sustainability advice ignores local context. By telling Gemini
about Coimbatore specifically, the app gives tips like:

"Many Coimbatore markets accept cloth bags"
"Collect AC condensate water for plants"
"Use city buses for regular commutes"

This makes EcoChat genuinely useful for our community.

Prize Categories

Best Use of Google Gemini

I used Gemini 2.5 Flash for three distinct features:

Multi-turn text chat with a custom hyper-local system prompt
Multimodal image analysis of trash and product labels
Dynamic Earth Day tip generation on page load

What I Learned

Gemini's multimodal API is surprisingly easy — image analysis took less than 20 lines of code
Hyper-local system prompts dramatically improve response quality
Small UX details like the commit button transform a passive chatbot into an active accountability tool

Do Not Press -- an extremely important button that you should not press

Sarvesh — Sat, 11 Apr 2026 13:17:16 +0000

This is a submission for the DEV April Fools Challenge

What I Built

I built a button. One button. A big, red, very important button that says DO NOT PRESS.

It does nothing. There is no easter egg. There is no secret. There is no reward. There is only the button, your poor decisions, and a Regret Level meter that fills up every time you make the wrong choice — which is every time you press it.

Why did I build this? That's the wrong question. The right question is: why did YOU press it?

Demo

🔴 Try it live: https://sarvesh555.neocities.org/do_not_press_button

Code

The entire project is a single HTML file. No framework. No dependencies. No npm install. No build step. No CI/CD pipeline. No cloud infrastructure. Just a button, some JavaScript, and 25 handcrafted messages of escalating disappointment.

<!-- The most important part -->
<button id="big-btn" onclick="pressIt()">DO NOT<br>PRESS</button>

The pressIt() function:

Increments a counter
Displays a new consequence message (there are 25)
Updates the Regret Level meter
Judges you via an escalating verdict system
Changes the page title after 10, 20, and 30 presses
Files a complaint with HR on your behalf

How I Built It

HTML, CSS, and vanilla JavaScript. That's it. The most technologically complex part was deciding how many times someone would press a button they were explicitly told not to press before giving up. The answer, based on extensive research, is: way more than you'd expect.

The regret meter starts red, turns amber at 70%, and goes fully dark red at 100% -- a subtle UX touch to indicate that you have gone too far, and yet here you are.

Total build time: less than an hour. Total usefulness: zero.

Prize Category

Community Favorite -- because nothing brings a community together like being told not to do something and then all doing it anyway.

This button is a mirror. What does it say about you that you pressed it?

Press it and find out.

AI Study Planner — Groq AI + Notion MCP

Sarvesh — Sun, 29 Mar 2026 04:47:45 +0000

This is a submission for the Notion MCP Challenge

What I Built

I built an AI Study Planner that generates personalized day-by-day study schedules using Groq AI and automatically syncs them to Notion via Notion MCP.

Students enter:

📖 Exam name (e.g. JEE Mains, GATE, MBA)
📅 Number of study days
⏰ Hours available per day
🗓 Exam date

The AI (Groq LLaMA 3.3) generates a smart topic-by-topic study plan and pushes it directly into a Notion database — creating structured entries with Task, Date, Exam Date, Hours, and Subject columns filled automatically.

Video Demo

Show us the code

GitHub: https://github.com/sarveshm555/ai-study-planner

1. User fills the form

const res = await fetch("http://localhost:5000/add-task", {
  method: "POST",
  body: JSON.stringify({ subject, days, hours, examDate })
});

2. Groq AI generates the study plan

const aiResponse = await groq.chat.completions.create({
  model: "llama-3.3-70b-versatile",
  messages: [{
    role: "user",
    content: `Create a ${days}-day study plan for ${subject}. 
    Return ONLY a JSON array with day, topic, description.`
  }]
});

3. Notion MCP creates pages automatically

await notion.pages.create({
  parent: { database_id: DATABASE_ID },
  properties: {
    Task: { title: [{ text: { content: `Day ${item.day}: ${item.topic}` } }] },
    Date: { date: { start: taskDate } },
    "Exam Date": { date: { start: examDate } },
    Hours: { number: parseInt(hours) },
    Subject: { rich_text: [{ text: { content: subject } }] }
  }
});

How I Used Notion MCP

Notion MCP is the core output layer of this project:

✅ Connects to Notion workspace database
✅ Creates structured pages for each study day
✅ Fills Task, Date, Exam Date, Hours, Subject columns
✅ Adds AI-generated descriptions inside each page
✅ Everything happens automatically — zero manual work!

Tech Stack

Tool	Purpose
Groq AI (LLaMA 3.3)	Generate smart study topics
Notion MCP	Auto-sync plans to Notion
Node.js + Express	Backend server
HTML/CSS/JS	Beautiful dark UI

What I Learned

How to integrate Notion MCP with AI workflows
How to structure AI prompts to return clean JSON
How to build a full-stack AI automation tool

👨‍💻 Team Details

Solo Submission

Name: Sarvesh M
DEV Username: @sarveshofficial
GitHub: https://github.com/sarveshm555

Built with ❤️ from India 🇮🇳