Multi-Agent Intelligence Blueprint
AGENT_README.md Generator for Autonomous AI Agents.
Standard READMEs are built for humans, not AI. AGENT_README.md provides deep architectural invariants, non-negotiable guardrails, and subagent delegation boundaries so AI agents never hallucinate or break your codebase.
Sample AGENT_README.md BlueprintMulti-Agent Ready
# 🤖 AGENT_README.md — Codebase AI Intelligence Dossier
## 🎯 1. Executive Mission & System Invariants
- Purpose: High-performance production microSaaS application.
- Paradigm: Modular Next.js 16 (App Router) + TypeScript + Edge API runtimes.
- Truth Policy: Verified commands in this dossier override generic LLM assumptions.
## 🛡️ 2. Non-Negotiable Guardrails ("The Red Lines")
- ❌ NEVER expose private API keys (ANTHROPIC_API_KEY, SUPABASE_SERVICE_ROLE).
- ❌ NEVER edit generated build directories (.next/, dist/, out/).
- ❌ NEVER use implicit 'any' types in TypeScript.
- ✅ ALWAYS run 'npm run build' and 'npm test' to verify changes before completion.
## 👥 3. Multi-Agent Delegation Blueprint
- 🎨 Frontend Specialist: Owns /src/components and /src/app. Handles UI/UX and styling.
- ⚙️ Backend Specialist: Owns /src/app/api and /src/lib/actions.ts. Handles data validation.
- 🛡️ QA & Security Specialist: Owns test suites and executes typecheck validation.Adaptive Sizing
Generates a unified master dossier for small repos, or a modular subagent folder structure for large multi-package repos.
Anti-Hallucination Guardrails
Explicit "NEVER do X" and "ALWAYS do Y" rules prevent agents from rewriting working architectural patterns.
Multi-Agent Swarm Ready
Partitions folder ownership and task roles across Frontend, Backend, and QA subagents seamlessly.