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
- 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.
Top comments (0)