The Problem: AI Coding Without Guardrails

Using the raw Claude Code CLI is powerful, but it often leads to:

• 💸 Wasted Tokens: Claude aggressively reads node_modules, dist/, or .git/, eating up your limits.
• 🔓 Security Risks: Accidental exposure of your .env, .pem keys, and credentials into AI context.
• 🍝 Spaghetti Code: AI outputting code without structure, tests, or documentation updates.

The Solution: Agentic Base

Agentic Base is a highly-optimized, 14-file configuration kit that seamlessly injects specialized agents, security gates, and structured workflows straight into your terminal-based AI environment.

🌟 Core Features

• 🛡️ Zero-Leak Privacy Gate: Cryptographically blocks the CLI from reading or editing .env, keys, or credentials without your explicit, mid-session Y/N approval.
• 🧹 Scout Gate (Token Saver): Automatically ignores bloated directories, slashing token burn and keeping the AI context window laser-focused.
• 🧠 Auto-Context Bootstrapper: Instantly detects your current framework (Next.js, Vite, etc.), runtime, and Git branch to spoon-feed optimal context to Claude on startup.
• 🤖 4 Specialized Sub-Agents: Ready-to-use personas: planner, tester, reviewer, and debugger. Invoke them naturally: "Use the tester agent to cover this component."
• 🏛️ The Three Pillars Enforcement: Built-in lifecycle hooks ensure strict adherence to Document-Driven (DDD), Test-Driven (TDD), and Observability-Driven (ODD) development. Claude will literally remind you if tests aren't written!

🚀 Quick Start (Zero Config)

Run this once inside any project repository:

npx @namtroi/agentic-base init

This cleanly scaffolds the .claude/ directory and injects the CLAUDE.md golden prompt.

Then, just start coding as usual:

claude

Claude will automatically ingest the new structured rules and optimized context.

🛠 Workflow Superpowers

With Agentic Base installed, you unlock micro-orchestration commands directly in Claude Code:

• /plan Add OAuth2 authentication Forces Claude to halt, research your stack, and output a structured plan.md (complete with test scope and docs impact) before writing any code.
• /preview --diagram payment architecture Instantly generates Mermaid or ASCII visual diagrams to explain complex systemic changes before you approve them.

📚 Documentation

Ready to customize hooks, write your own agents, or block specific directories?

👉 Read the Full Documentation Guide

🤝 Open Source

Built with ❤️ for the AI developer community.

License: MIT