Agentic Workflow

Standardize AI agent setup across your entire team. One interactive CLI, uniform configuration for everyone.

The biggest barrier to team-wide AI adoption isn't the tools it's the inconsistent setup across team members. One developer has a polished CLAUDE.md, another has nothing. One has MCP servers configured, another doesn't know they exist. Skills that work brilliantly for one person are invisible to the rest of the team.

agentic-workflow fixes this. A project lead answers 10 questions about their stack, and the script generates a complete set of configuration files that every team member gets identically same rules, same skills, same agent capabilities.

Supported Platforms

Platform	Config files
Claude Code	CLAUDE.md, .claude/skills/, .claude/settings.json
Cursor	.cursorrules
GitHub Copilot	.github/copilot-instructions.md
Windsurf	.windsurfrules
Codex	.codex/instructions.md
Antigravity	AGENTS.md
Roo Code	.roo/rules/project-rules.md
Kilo Code	.kilocode/rules/project-rules.md

The Three Pillars

This tool generates configuration across three pillars that form a complete agentic setup:

Pillar	Purpose	What gets generated
Project rules & skills	Encode team conventions so the agent behaves consistently for everyone	CLAUDE.md, .claude/skills/, .cursorrules, etc.
Knowledge files	Give agents project-specific context to avoid hallucination	Architecture docs, API references, coding conventions
MCP & tool integration	Connect agents to external systems (GitHub, Context7, etc.)	MCP server configs, IDE settings

Why this works:

• Skills-first. Skills are portable, platform-agnostic instruction files. Standardizing them early prevents fragmentation as team members use different IDEs or agent platforms.
• Knowledge files are the underrated piece. Most teams skip this. Auto-scaffolding knowledge structures means agents produce output aligned with your actual codebase from day one.
• MCP integration as a standard. Baking GitHub MCP, Context7, and others into the setup ensures everyone has the same agent capabilities out of the box.

Getting Started

Install

curl -fsSL https://raw.githubusercontent.com/ADORSYS-GIS/agentic-workflow/main/install.sh | bash

Alternatively, clone and run directly:

git clone https://github.com/ADORSYS-GIS/agentic-workflow.git
cd agentic-workflow
bash setup.sh

Run the setup

agentic-workflow

The interactive questionnaire walks you through 10 questions:

#	Question	Examples
1	Project name	my-saas-app
2	Languages	TypeScript, Python, Java, Go, Rust, ...
3	Frameworks	React, Django, Spring Boot, Axum, ...
4	Package managers	pnpm, poetry, Gradle, Cargo, ...
5	Repo structure	Monorepo or single repo
6	AI agent platforms	Claude Code, Cursor, Copilot, Windsurf, Codex
7	IDEs	VS Code, IntelliJ, Neovim
8	MCP servers	GitHub, Context7, Filesystem, PostgreSQL
9	Testing frameworks	Vitest, pytest, JUnit 5, go test, ...
10	AI-assisted workflows	PR review, testing, docs, debugging, security, refactoring

Answers are saved to agentic-config.conf so you can regenerate or share with the team.

Regenerate from config

agentic-workflow --from-config agentic-config.conf --output /path/to/project

Re-run generation without answering questions again useful when updating the tool or applying the same setup to a new repo.

What Gets Generated

your-project/
├── CLAUDE.md                         # Project rules (composed from language + framework templates)
├── .cursorrules                      # (if Cursor selected)
├── .github/copilot-instructions.md   # (if Copilot selected)
├── .codex/instructions.md            # (if Codex selected)
├── .windsurfrules                    # (if Windsurf selected)
├── .claude/
│   ├── skills/                       # AI workflow skills
│   │   ├── pr-review/SKILL.md
│   │   ├── testing/SKILL.md
│   │   ├── documentation/SKILL.md
│   │   ├── debugging/SKILL.md
│   │   ├── security-audit/SKILL.md
│   │   └── refactoring/SKILL.md
│   └── settings.json                 # MCP server configuration
├── docs/knowledge/                   # Project knowledge scaffolds (fill in the TODOs)
│   ├── architecture.md
│   ├── coding-conventions.md
│   ├── api-reference.md
│   └── development-setup.md
└── GETTING_STARTED.md                # Onboarding guide for team members

How CLAUDE.md is composed

The generated CLAUDE.md isn't a static template it's composed from fragments matched to your answers:

1. Base rules (always included) git conventions, security principles, error handling, code review standards
2. Language rules one section per selected language with idiomatic patterns, naming conventions, and pitfalls
3. Framework rules one section per selected framework with project structure, routing, state management, and anti-patterns
4. Testing rules test structure, coverage expectations, and mocking philosophy for your chosen test frameworks

A TypeScript + React + Vitest project gets a completely different CLAUDE.md than a Python + FastAPI + pytest project both tailored to their stack.

Supported Stacks

11 languages and 60 frameworks across all major ecosystems:

Language	Frameworks
TypeScript / JavaScript	React, Next.js, Angular, Vue, Nuxt, Svelte, SvelteKit, Express, NestJS, Fastify, Hono, Remix, Astro
Python	Django, Flask, FastAPI, Starlette, Litestar, Pyramid, Sanic, Tornado
Java	Spring Boot, Quarkus, Micronaut, Jakarta EE, Vert.x, Dropwizard, Blade
Kotlin	Ktor, Spring Boot (Kotlin), Exposed
Go	Gin, Echo, Fiber, Chi, Gorilla Mux, Buffalo, Beego
Rust	Axum, Actix-Web, Rocket, Warp, Tide, Poem
C#	ASP.NET Core, Blazor, .NET MAUI, Entity Framework
Ruby	Rails, Sinatra, Hanami, Grape
PHP	Laravel, Symfony, Slim, CodeIgniter, CakePHP
Swift	SwiftUI, UIKit, Vapor

After Setup

Once the files are generated:

1. Review CLAUDE.md the generated rules should match your team's actual conventions. Edit anything that doesn't fit.
2. Fill in docs/knowledge/ these are scaffolds with TODO markers. Replace them with your real architecture, API docs, and conventions.
3. Update MCP tokens open .claude/settings.json and replace placeholder tokens with real values.
4. Share GETTING_STARTED.md distribute to every team member so they can verify their setup.
5. Commit to your repo these files should live in version control so every team member gets them automatically.

Reporting an Issue

If you believe you have discovered a defect in agentic-workflow, please open an issue. Please provide a clear summary, description, and steps to reproduce.

Contributing

Before contributing to agentic-workflow, please read through the codebase and existing patterns. The project is pure Bash with no external dependencies keep it that way.

Key areas for contribution:

• New language or framework templates in templates/claude-md/
• New skill definitions in templates/skills/
• New MCP server configurations in templates/mcp/
• Improvements to the interactive questionnaire

Uninstall

curl -fsSL https://raw.githubusercontent.com/ADORSYS-GIS/agentic-workflow/main/install.sh | bash -s -- --uninstall

Or if cloned locally:

bash install.sh --uninstall

License

MIT License