Set Up an Entire AI Development Pipeline with a Single Command

AI Factory Overview

AI Factory is an npm package that adds a slash‑command system to AI‑centric development environments. It automatically detects the project stack (e.g., Next.js, Python, Laravel), installs the required skills and MCP services, and equips the chosen AI agent with a structured workflow (plan → implement → verify → fix). The core principle is spec‑driven development: the AI must receive a plan before generating code, and each step can be paused, resumed, and audited.

Problem Addressed

When using Claude Code, Cursor, or Codex, developers often spend 30 minutes re‑configuring skills, MCP servers, commit conventions, and CI for each new project. This repetitive setup erodes momentum before actual coding begins.

Workflow Architecture

phase_0: configuration (once)
  └─ /aif → generate DESCRIPTION / ARCHITECTURE / ROADMAP / RULES / AGENTS.md

development_loop (repeatable)
  ├─ /aif-roadmap   → split requirements into milestones
  ├─ /aif-plan      → detailed task breakdown per milestone
  ├─ /aif-improve   → refine the plan
  ├─ /aif-implement → execute with automatic checkpoint commits
  ├─ /aif-verify    → verify each task is complete
  ├─ /aif-fix       → fix issues
  └─ /aif-evolve    → accumulate patches and evolve

Running the loop creates persistent planning documents (DESCRIPTION, ARCHITECTURE, ROADMAP, etc.) that can be reused across sessions.

Installation

# npm global install
npm install -g ai-factory
# or with mise
mise use -g npm:ai-factory

Usage

From a project directory, start the interactive wizard or pass parameters directly:

# interactive wizard
ai-factory init
# with explicit parameters
ai-factory init --agents claude,codex --mcp playwright,github

The wizard asks for:

AI agent (e.g., Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI)

Skills to install

MCP servers to configure (e.g., Playwright, GitHub)

After configuration, invoke the AI with the slash command /aif (or $aif for Codex CLI).

Complete Workflow Example

# explore requirements (optional)
/aif-explore Add user authentication with OAuth
# grounding check
/aif-grounded Does this repo already support OAuth providers?
# planning
/aif-plan Add user authentication with OAuth
# optional refinement
/aif-improve
# implementation with automatic commits
/aif-implement
# provide reference documentation
/aif-reference https://docs.example.com/api-reference --name example-api
# fix a bug
/aif-fix TypeError: Cannot read property 'name' of undefined
# generate CI configuration
/aif-ci github
# generate project documentation
/aif-docs

This sequence produces code, CI pipelines, and documentation from a single specification.

Design Highlights

Reflex Loop : Four‑step cycle PLAN → PRODUCE → EVALUATE → REFINE that repeats until quality stabilizes.

Quality Gates : Each step emits a machine‑readable aif-gate-result; the pipeline blocks progress until the gate passes.

Skill Ecosystem : Integrates with the skills点sh marketplace; external skills undergo two‑level security scans.

Multi‑Agent Support : Single configuration works with 15+ agents (Claude Code, Cursor, Windsurf, Roo Code, Kilo Code, Antigravity, OpenCode, Warp, Zencoder, Codex CLI, Codex app, GitHub Copilot, Gemini CLI, Junie, Qwen Code, etc.).

AIF Handoff : Tasks can be placed in a backlog where the AI autonomously plans, implements, reviews, and ships without human intervention.

Cost‑Benefit Claim

The official claim states that spending roughly 20 % more tokens on planning saves about 60 % on patching, rewrites, and context‑restoration costs.

Token usage : Planning is token‑expensive; subsequent fixes become token‑cheap.

Documentation : Automatically maintained within the workflow versus manual, error‑prone syncing.

Context persistence : Roadmap/plan/architecture specifications survive across sessions, preventing “forgotten” context.

Bug fixing : /aif-fix reads the original plan, eliminating the need to re‑explain intent.

Quality assurance : /aif-verify validates each task, reducing reliance on luck.

CI/Automation : CI configurations are generated and kept up‑to‑date automatically.

Limitations

Learning curve is higher than using an AI agent directly; users must remember slash commands.

Strict spec‑driven flow can feel heavyweight for rapid‑experimentation scenarios.

Effectiveness depends on clear, well‑written requirements; poor input leads to poor output.