AI Dev Stack

A production-grade kit of rules, docs, prompts, and patterns for AI-native development. It teaches your AI coding tools — Cursor and Claude Code — to think like a principal architect: pick the right layer, follow clean architecture, and ship deploy-ready code instead of demos. Drop it into any project and the assistant inherits a consistent operating model on the first prompt.

Quick Start

Cursor

curl -fsSL https://raw.githubusercontent.com/aiagentwithdhruv/ai-dev-stack/main/install.sh | bash

Claude Code

curl -fsSL https://raw.githubusercontent.com/aiagentwithdhruv/ai-dev-stack/main/claude/CLAUDE.md -o CLAUDE.md

Both (recommended)

curl -fsSL https://raw.githubusercontent.com/aiagentwithdhruv/ai-dev-stack/main/install.sh | bash
curl -fsSL https://raw.githubusercontent.com/aiagentwithdhruv/ai-dev-stack/main/claude/CLAUDE.md -o CLAUDE.md

The mental model: substrate × two axes

Everything in the kit sits on one substrate and is organized along two axes — how you build and what you build. Reserved space (_frontier/) holds patterns that aren't stable yet.

ai-dev-stack/
├── foundations/      # SUBSTRATE — non-negotiables every build inherits
│   ├── rules/                  # clean architecture, security, response style
│   ├── docs/                   # PRD / ARCHITECTURE / API / SCHEMA / DEPLOY templates
│   ├── evals/                  # measure before you trust — task + regression evals
│   ├── observability/          # tracing, cost, latency, structured logs
│   ├── guardrails/             # layered policy → input → output → monitor
│   └── prompts/                # reusable system + task prompt patterns
│
├── pillars/          # AXIS 1 — HOW you build
│   ├── software-development/   # backend, frontend, data, API contracts, DevOps
│   ├── agents/                 # tools, schemas, orchestrator–worker, supervisor loops
│   └── automation/             # event/scheduled pipelines  → companion repo below
│
├── domains/          # AXIS 2 — WHAT you build
│   ├── rag-knowledge/          # ingestion, chunking, retrieval, grounded answers
│   ├── data-analytics/         # NL-to-SQL, metrics, reporting, BI assistants
│   ├── voice/                  # STT, TTS, real-time voice agents
│   ├── vision-doc-ai/          # OCR, document extraction, multimodal pipelines
│   ├── content-generation/     # long-form, structured, and media generation
│   └── decisioning-forecasting/# scoring, ranking, prediction, recommendations
│
└── _frontier/        # RESERVED — emerging patterns, not yet production-stable

Read it as a grid. Any project picks one or more pillars (the how) and one or more domains (the what), then stands the whole thing on foundations. A RAG support assistant is pillars/agents + domains/rag-knowledge on foundations/{rules,evals,guardrails}. A nightly report bot is pillars/automation + domains/data-analytics. The substrate never changes; the axes compose.

The substrate — foundations/

The defaults every build inherits, regardless of pillar or domain. Rules and doc templates tell the AI how to write code and what you're building; evals, observability, and guardrails keep it honest in production. Start here — see foundations/.

Axis 1 — pillars (HOW you build)

Pillar What it covers
software-development/ Thin routes, services, repositories, typed API contracts, caching, CI/CD.
agents/ Tool schemas, validated outputs, orchestrator–worker and supervisor patterns.
automation/ Event-driven and scheduled pipelines — see the companion repo below.

Axis 2 — domains (WHAT you build)

Domain What it covers
rag-knowledge/ Separate ingestion from generation; chunk metadata; grounded, cited answers.
data-analytics/ NL-to-SQL, read-only query agents, metrics, dashboards.
voice/ Speech-to-text, text-to-speech, low-latency voice agents.
vision-doc-ai/ OCR, document extraction, multimodal understanding.
content-generation/ Long-form, structured, and media content with quality gates.
decisioning-forecasting/ Scoring, ranking, forecasting — classical models before LLMs for tabular data.

Companion repos

Focused repos that pair with this stack (kept separate so each stays searchable and reusable on its own):

  • ai-automation-kit — n8n-first + general workflow-automation patterns. The automation pillar links out to it.
  • skills — installable AI-agent skill packs (npx skills add …), cross-tool (Claude / Cursor / Copilot / Codex / Windsurf / Cline). The foundations/ rules-and-skills layer links here.
  • ghost-browser — AI-powered browser automation (web scraping, auto-posting); a runnable example that pairs with the automation pillar.

Suggested GitHub Topics

ai-agents · rag · llm · prompt-engineering · automation · mcp · llmops · evals · claude-code · cursor · ai-development

Contributing & changelog

  • Adding a rule, prompt, or pattern? See CONTRIBUTING.md — including the generic, de-identified content rule.
  • Version history lives in CHANGELOG.md.

License

MIT — use it, fork it, ship better code.