I Built an Open-Source AI Toolkit for ChatGPT, Claude, Gemini and Grok — Here’s What’s Inside

Alptuğ Harun — Wed, 30 Sep 2026 19:39:04 +0000

I spend a lot of time working with AI tools.

ChatGPT, Claude, Gemini, Grok, automation tools, content systems, research workflows, visual production, social media...

And after a while I noticed a problem:

I had plenty of prompts, notes, experiments and useful workflows — but they were scattered everywhere.

A good prompt lived in one document.

An automation idea lived somewhere else.

An assistant configuration was buried in another chat.

A workflow that worked really well once might be difficult to reproduce a month later.

So I started turning those experiments into something more structured.

Not another giant list of prompts.

A reusable AI workspace.

That project became AI Social Media Toolkit.

What I’m trying to build

The idea is relatively simple:

Instead of treating AI as a collection of isolated chat sessions, I want to organize the useful parts into reusable components.

The project currently includes:

structured prompt workflows
assistant blueprints
Agent Skills
MCP and integration guidance
API bot starters
automation recipes
learning paths
creator workflows
validation and testing utilities

The main ecosystems I’m experimenting with are:

ChatGPT / OpenAI, Claude, Gemini and Grok.

But the goal is not to create four separate libraries.

The goal is to understand which parts of an AI workflow can remain portable even when the model or provider changes.

1. Prompt Library

One of the first things I wanted to fix was prompt chaos.

A prompt is much more useful when it explains:

when to use it
when not to use it
what inputs it requires
what output structure to expect
how to evaluate the result
what to do when the output is weak

So instead of storing prompts as isolated blocks of text, the repository uses a more structured format.

Some of the current workflows cover:

source synthesis
trend evidence auditing
assistant creation
approval-gated automation
Reels planning
Pinterest research
brand voice adaptation
output quality auditing

The interesting part for me isn't the wording of the prompt itself.

It's the workflow around it.

2. Portable Assistant Blueprints

Another area I'm exploring is reusable assistants.

Many people create a useful GPT, Claude Project or Gemini Gem and then essentially lock the workflow inside that platform.

I wanted to separate the assistant's logic from the provider.

So the toolkit contains blueprints for thinking about things like:

role
context
instructions
inputs
tools
boundaries
expected outputs
quality checks

There are currently examples for ChatGPT-style assistants, Claude Projects, Gemini Gems and Grok-oriented workflows.

They are not intended to pretend every platform works identically.

They are meant to make the underlying system easier to understand and migrate.

3. Agent Skills

The repository also contains reusable Agent Skills.

Right now there are skills around areas such as:

creator operations
content signals
Reels
Pinterest
content repurposing
brand voice
comment intelligence
GitHub opportunity research
local business intelligence
commercial opportunity research

An important rule I’m trying to follow here:

a skill should do something specific.

I don't want the repository to become a folder full of impressive-sounding AI terminology with no practical use.

4. MCP and integrations

MCP is another area I’m actively learning and experimenting with.

For me, the interesting question isn't simply:

“Can an AI connect to another tool?”

It is:

“Can we create repeatable workflows where context, tools and permissions are understandable and controllable?”

The repository therefore includes MCP and integration guidance alongside the prompt and assistant layers.

This is still an area I expect to evolve significantly.

5. Bots and automation

I also wanted to experiment with moving beyond the chat window.

The project contains starter structures for working with multiple AI providers and automation workflows.

This includes a multi-provider assistant starter and examples of approval-gated automation.

One principle matters a lot to me here:

automation should not automatically mean autonomy.

There are many situations where a system can research, prepare or recommend something, while a human should still approve the final action.

That boundary is something I'm deliberately keeping visible.

6. 10 Quick Wins

A large repository can become useless if a new visitor doesn't know where to begin.

So one of the most recent additions is 10 Quick Wins.

The idea is to let someone open the repository and quickly try a useful workflow without studying the entire project first.

Examples include:

synthesizing sources
auditing a trend
creating an assistant blueprint
testing an API assistant in mock mode
designing an approval-gated workflow
planning a Reel
building a Pinterest content cluster
adapting brand voice
auditing an AI-generated output

This is probably the part I want to improve most based on outside feedback.

Why am I building this?

My background is primarily in social media, digital content and creative strategy.

I'm not trying to present myself as a traditional software engineer.

What interests me is the layer between creative work and technical AI systems.

How can a creator take a useful process and make it repeatable?

How can research become a workflow?

How can prompts become systems?

How can assistants become portable?

How can automation save time without removing human judgment?

That's the territory I'm exploring.

A rough mental model I use is:

Signal → Research → Strategy → Content → Distribution → Measurement

AI can participate in almost every part of that chain.

But only if the workflow around it is designed properly.

What I don't want this project to become

I don't want to optimize the repository around having the largest possible number of prompts, skills or tools.

More files do not automatically make a better project.

I'd rather discover which workflows people actually use and improve those.

So the next phase is less about adding everything I can think of and more about seeing where real usage appears.

I’d genuinely like feedback

The project is still in alpha.

If you explore it, I’m especially interested in four things:

What feels immediately useful?
What feels confusing or unnecessary?
Which workflow would you expand first?
Is there anything here you would actually return to use again?

The project is open source and available here:

👉 AI Social Media Toolkit on GitHub

I’m building it in public, so useful criticism is much more valuable to me than polite praise.

If you try something from it, I’d love to know what happened.

DEV Community: Alptuğ Harun