Agent Dev Flow — AI skill for Claude Code

AI Engineering System

**A reusable, tool-neutral operating system for planning and building software with AI agents.** It defines *how* work is classified, planned, approved, implemented, tested, reviewed, and released — without prescribing any application type or technology stack. The stack and applications are chosen per project, with user approval.

It is **documentation and governance**, not application code: no dependencies, no committed stack. It works with **Claude Code, OpenAI Codex, Cursor, Windsurf, GitHub Copilot, Antigravity**, or any agent that can read files.

Version **1.0.0** · MIT · The full guide lives in [`.ai/README.md`](.ai/README.md).

---

What's inside

Layer	Count	Where
Skills (reusable capability modules)	178	`.ai/skills/`
Agents (roles for multi-agent runs)	13	`.ai/agents/`
Hooks (tool-neutral lifecycle checklists)	13	`.ai/hooks/`
Workflows (per request type)	12	`.ai/workflows/`
Templates (fill-in documents)	31	`.ai/templates/`
Prompts (tool-neutral starters)	19	`.ai/prompts/`
Checklists (verifiable gate/hook checks)	18	`.ai/checklists/`

Skills are organized into **8 packs**: core (28), mobile (36), web & dashboard (26), backend (30), database (15), testing (15), devops (16), security (12). Plus system rules, knowledge, memory, and references — all in [`.ai/`](.ai/README.md).

Install

Fastest — one command (all editors)

Set up any project with a single command. It copies `.ai/`, the `AGENTS.md` entry point, and your editor adapter(s) — no dependencies, no stack, no repo:

npx github:ahtishamshahzad/agent_dev_flow init

Only the editors you use: `--editor claude,cursor` (values: `claude`, `cursor`, `windsurf`, `copil