🚀 QuanuX: A "Self-Aware" Trading Platform Extension for Gemini

[quantDIY]

Hey everyone! 👋

We’ve been building QuanuX (a high-performance distributed trading platform), and we used Gemini to build an extension... for Gemini.

The QuanuX Extension turns the Gemini CLI into an expert developer for our specific architecture. It doesn't just read files; it has "Native Skills" and tools to operate the platform. Generate your own UI in Figma and have Gemini plug it into QuanuX, fully compliant with our server-side mandate for business logic.

🧠 What it does:
Architectural Guardrails: It knows our strict rules (e.g., C++ for Execution, Python for Strategy) and enforces them in conversation.
Native Skills: We successfully integrated the Agent Skills protocol. The extension dynamically loads react-native-standards as a skill, allowing Gemini to "specialize" in mobile UI generation on demand.
Platform Control: It can bridge the CLI to our runtime to generate strategies (
generate_strategy
) and run system commands (
run_quanuxctl
) directly from the chat.
Self-Configuration: It even knows how to configure Gemini itself (teaching users about .geminiignore, /model, and trusted folders).
📦 Try it out:
If you are curious about how we structured a complex domain extension, check out the repo or install it (requires a QuanuX environment to actually run platform tools, but the context features work anywhere). Full featured version 0.0.1 will be dropping very soon (check https://github.com/QuantDIY/QuanuX for feature previews of the upcoming release!):

bash
gemini extension install https://github.com/QuanuX/quanux-gemini-cli-mcp
Feedback welcome! We're mostly excited about the potential of Native Skills bundled inside extensions. 🚀

[aniruddhaadak80]

From my point of view, domain specific extensions like this become most credible when they are explicit about where the guardrails stop and where normal agent autonomy begins. The interesting part is not only that the extension can operate the platform, but that it can encode architectural policy in a way that stays inspectable instead of becoming opaque prompt magic. That is the kind of pattern that can make extensions genuinely valuable rather than just larger prompt packs.

[quantDIY]

Yes, and you're free to edit or add guardrails. I've been thinking that I'll go a step further and build an integrated AI as well. Decided yesterday that it will need to be very focused on core platform features/workflows and also open-sourced so others can use it as a jumping off point for custom inference projects. Regardless, anyone using the platform should grep for skill files as they're integrated in almost every directory/subdirectory to explain to agents the purpose each serve serves, the rationale for architectural decisions and engineering concepts, so that they do not break any functionality.