groundworklisted

Use for any FastAPI, Pydantic, LangChain/LangGraph, RAG, SSE streaming, OpenAPI, Next.js, Flutter or Arabic/RTL work — including one change to an existing production service ("add a streaming endpoint to our FastAPI service", "add an endpoint", "add RAG over our documents", "review this code") as well as a new product ("new project", "scaffold", "MVP", "which stack should I use", "build the frontend", "Flutter app"). Also uv, Deep Agents, hybrid RAG, Qdrant, pgvector, vLLM, middleware, TanStack, shadcn, Riverpod, and spec workflows (GitHub Spec Kit, OpenSpec, Superpowers). Use it even if it is not named, and even when the request looks answerable without it.
abdelrahmaan/groundwork · ★ 0 · AI & Automation · score 72

Install: claude install-skill abdelrahmaan/groundwork

# Groundwork One standard for the whole product, for a **Python backend + AI** service with a **React / Next.js** web app and a **Flutter** mobile app, joined by a single OpenAPI contract. Two rules govern everything below: - **MVP first.** Build the smallest thing that proves the goal. Nothing exists until a stated need asks for it. - **Docs, not memory.** Library details come from current official docs (Context7 / the framework's docs MCP). Pairs with **any spec workflow** — GitHub Spec Kit, OpenSpec, Superpowers, or another: the workflow owns the process (spec → plan → tasks → implement → review), this kit owns the engineering standard. See `references/spec-workflows.md`. Last verified against official docs: 2026-09-20. Version floors live in each reference file. **Works on any agent.** This is a plain Agent Skill: `SKILL.md` plus Markdown reference files. On agents that don't auto-load skills, the user names it ("use groundwork"); the agent then follows §9 and opens the reference files itself. --- ## 1. Operating modes (how the agent works) ### 1.0 Language of work - **Conversation:** English by default. Switch only if the user's latest message is written in another language — then reply in that language. - **Technical terms stay in English** in every language (RAG, embedding, reranker, middleware, rung, endpoint, stack guide…). Translating them creates wrong words and ambiguity. - **Every generated artifact is English, always:** code, comments, identifiers, `docs