dw-dwain/ai-gen-company-stack

AI Gen Company Stack is a bundle of plugins and skills for AI coding assistants, specifically Claude Code and Codex. Instead of writing out a long list of separate instructions and prompts every time you want an assistant to help build an AI product, this project gives the assistant a set of ready made routines it can call on for common tasks. The stack covers a wide range of jobs. It helps plan the architecture of an AI agent app using popular frameworks such as Pydantic AI, LangGraph, LangChain, LlamaIndex, and vLLM. It can help design a retrieval based question answering system, often called RAG, where the assistant looks things up in your documents before answering. It includes a skill that pushes back against generic looking AI generated interfaces and pushes for more distinctive design instead. Other skills keep code from becoming overly complicated, keep written output short while staying accurate, and add routing for defensive cybersecurity tasks. There are also skills for hardening prompts against misuse, and for adding fallback, rollback, and learning loop behavior so an assistant can recover from mistakes and improve over time. Installation for Claude Code is done through the plugin marketplace system, either by pointing it at this GitHub repository or by running an included PowerShell install script. Codex support works by copying the included skills into the user's Codex skills folder, with the installer automatically backing up anything it overwrites. The project also ships small PowerShell scripts for checking that everything installed correctly, rolling back Codex skills to an earlier backup, recording notes from a completed task, and generating a starter scaffold for a new AI project. The author is explicit that this is an independent, original project. It is not an official product from OpenAI, Anthropic, Claude, or Codex, and its similarly named skills are not affiliated with any other tool that shares those names. It references other open source tools as integration targets but does not copy their source code. The project is written mainly in PowerShell scripts and is released under the MIT license.