Your plugin is live in ChatGPT. So are 13,000+ others. Now what?
Do you know which prompts you rank for? Or what your top competitor changed in their listing last week?
We tracked 30,000+ plugins across ChatGPT, Claude, Grok, Muse and more every day, and turned that data into skills your agent can run:
> simulate whether an agent picks you over competitors
> find the keywords competitors' listings cover and yours doesn't
> design widgets from what live ChatGPT apps render
> watch competitors' ranks, rewrites and tool changes
We just open sourced all of it as Plugin Stack, so you can run it on your own plugin.
Start with /plugin-stack. Give it to your agent, and it reads your repo, finds one thing to fix, and suggests what to do next.
npx skills add theopenapps/plugin-stack
github.com/theopenapps/plugi…
8
98
the web had shopify, agents have openapps
Did you notice that everybody on the timeline is 10x'ing their users through MCP's?
I saw these guys launch an intelligence layer for MCP's and all plugins out there
ChatGPT, Claude, Grok, Muse...
While browsing I came across an Escape from Tarkov Plugin.
You heard that right :D
It really shows that EVERYTHING can become MCP's.
I would have killed for something like this back in the day grinding Overwatch, used to be top 500 eu btw
Anyways, you got an MCP anon?
2
8
401
fellow builders of plugins!
point your agents to theopenapps.com/connect to find out
> which stores your competitors are on and you're not
> what they shipped this week, down to new MCP tools
> which features they have that your plugin doesn't
ChatGPT, Claude and Cursor each run an app store, and each one only ranks its own apps
This is the Sensor Tower for AI app stores.
@atanizamoglu, @OzSilahtar and I built Open Apps: all AI App Stores in one place. Top charts, new launches, every category's #1
theopenapps.com
Ata 🌲 retweeted
omux'u github'ıma yükledim
eğer cmux'u beğendiyseniz bunu 10x daha beğenirsiniz öyle söyleyeyim.
şuradan build alabilirsiniz: github.com/onurozcan/omux
Ata 🌲 retweeted
Bir bardak çayın arkasında binlerce insanın emeği var. Birbirini hiç tanımayan bu insanları birlikte çalıştıran şey ne?
Yeni videomuz "Para Nedir?": takastan altına, kralın damgasından Bitcoin'e paranın hikâyesi. 👇
Ata 🌲 retweeted
Introducing ManimGX, the 3D animation engine for agents.
- Blazingly fast: 90× faster than ManimCE
- Rust + wgpu core
- Compatible with Manim's Python API
- Lightweight: one pip install, no system dependencies
Open source: github.com/academa-labs/mani…
Docs: manimgx.academa.ai
For 40 years we built GUIs for humans. Now we build APIs for agents.
Try it. Paste this into your favorite coding agent:
"Follow manimgx.academa.ai/llms.txt to make me a 3D animation of the solar system, and open it."
Ata 🌲 retweeted
Last week we ran an AI meetup on personal agents at @DotshubIstanbul with @OzSilahtar and @atanizamoglu, two of the sharpest AI people I know, the kind of talent that's hard to find anywhere. Demand was high, the event got overbooked. And you know the interest is real when nobody in the audience is checking their phone ⚡️
We covered assistant harnesses, what makes an agent personal (memory, tools, skills), where it lives (surface, runtime, inference, sandbox) and the latest trends. Then the community took over with show & tell: people demoed their own harnesses and agentic setups.
I'm back home in Istanbul for a few short weeks and trying to share as much of San Francisco as I can while I'm here.
Next one is early September. Stay tuned 👾
Ata 🌲 retweeted
We increased our weekly outbound capacity from 80 prospects to 500 this week.
Previously, our team selected 80 people and wrote each message manually. That generated an average of 20 replies so far, a 25% response rate.
Our new GTM engine divides the process across specialized skills.
It evaluates ICP fit, selects the most relevant sales thesis, and generates each message accordingly. We can also A/B test alternatives at every stage.
The same team can now run outbound for 500 prospects on LinkedIn per week.
Our next test is whether we can maintain that 25% reply rate at this volume.
For us, scaling B2B outbound comes from systematically choosing the right prospect, sales thesis, and message, then continuously testing each decision.
We'll run the first test using Gojiberry for lead generation and distribution! I'll share some results later on.
html for control + generative for polish direction is brilliant
must have skill for app makers and beyond
Anyone tried using image models to make App Store screenshots? Mine looked cool but got rejected. Wrong sizes, warped text, frames Apple won't take.
So I built Hypershots.dev:
- HTML renders the panels perfect to spec — sizes, frames, fonts, every language
- image models add the fun: stickers, pop-outs, whole-set makeovers that never break a spec
- fastlane ships it, end to end
It's a free, open-source skill for Claude Code & Codex. Uses @fal in the backend: bring your own key.
15
285
Ata 🌲 retweeted
I didn’t open a video editor to make our MCP feature video.
I opened Telegram.
I was at a pub and sent Hermes the rough draft.
Hermes already had access to the Remotion skill, the NewFrame CLI, our product context, and the tools to render the thing.
By the end of the day, we had:
> a Claude connector flow
> the in-chat campaign moment
> generated creative cards
> an editable Remotion video
Normally, this job starts with a brief, then someone hunts for assets, then it gets passed between people and tools.
This one stayed in one production loop.
I could steer it from Telegram, review the frames, change the story, and keep moving while Hermes handled the boring production work.
That is a very different feeling from “we should make a product video.”
Still needs taste, review, and someone to stop it when it starts getting too clever :D
One annoying lesson from debugging lately:
Users do not experience your product as separate models, tools, prompts, and APIs.
They experience one system.
So when the assistant understands the request but the image model fails, the whole product failed.
You have to debug the chain, not just the model.
The boring parts between models matter a lot more than they look.
At that point, isn’t the chain the product?
1
5
171
MCPs might be back.
We're seeing it in customer feedback, users don't always want another dashboard. They want the product to be reachable where they already work.
Products are becoming less tied to their own UI and improving.
Madison MCP + CLI soon.
2
7
134
Everyone keeps talking about it and they are right, the prompt is no more the real system. The loop is.
> what signal starts the work?
> what context does the agent get?
> what can it change?
> where does it stop?
> who reviews it?
> what gets saved for the next run?
Better prompts do help but I can clearly say that most of the leverage is now in designing the loop around the model.
4
84
We spent a good amount of time on the boring parts behind this
> turning a link into reusable brand context
> pulling in your products for future use
> setting up a workspace environment
so that every generation you make going forward stays aligned with your brand.
Feels simple? that's the whole point.
The painful part of ongoing AI creative is repeating yourself.
> Explaining the brand.
> Explaining the products.
> Explaining the style.
> Explaining what should be generated next.
Again and again and again.
We are making this easier in Madison. You can now paste your website and Madison starts working from there.
It reads your brand, pulls your products, understands the context, and suggests where to begin.
Then you can start generating in minutes.
New landing + hero is live.






















