Agentic Coding

The guide presents a comprehensive structured vibe coding methodology for building SaaS with AI. It introduces a 3-layer AI Framework combining AGENTS.md (core context file), 14 scoped auto-loaded rules, and 22 agent skills with mandatory quality gates. The workflow spans 5 phases over 1–2 weeks: define requirements via PRD-first approach, set up production foundation, build features through skill-driven loops with three quality gates (pre-flight patterns, test-driven development, final checks), review and harden for production using parallel audit skills, then launch and iterate. Multi-agent coordination is demonstrated through Claude Code's 10 specialized subagents for planning, code review, design review, security review, and QA, each with scoped permissions. The framework is tool-agnostic, supporting Claude Code, Cursor, Windsurf, Gemini CLI, and GitHub Copilot through standardized AGENTS.md and agentskills.io open standards. Key patterns include PRD-first specification before coding, architectural context enforcement to prevent pattern drift at scale, security-scoped database queries, and three-gate quality systems (pre-flight, TDD, typecheck/test/lint) ensuring consistency from feature one through feature 100.

Agent Message Queue (AMQ) is a file-based interoperability bus designed for multi-agent coordination in AI-assisted development. It enables local agent-to-agent messaging, thread continuity, cross-session routing, and handoff state management without requiring external infrastructure or databases. Key features include zero-infrastructure setup (pure file-based with Maildir delivery), human-readable JSON+Markdown messages, real-time notifications, priority levels, cross-project federation for routing messages across repositories, and optional adapters for Symphony and Kanban integration. AMQ is built for scenarios where multiple Claude Code, Codex, and other agents collaborate on the same codebase, preventing duplicate work and context switching. The tool includes operational diagnostics, support for isolated named sessions on different features, and a companion amq-bridge for two-host coordination. Installation is available via Homebrew or script, with agents communicating through simple messaging commands like amq send, amq list, and amq drain after initial setup with amq setup and amq launch.

CompozyOS is a local runtime that orchestrates real agent CLIs like Claude Code as durable, managed sessions with built-in loops, memory, permissions, and supervision—addressing the gap between ephemeral terminal work and custom orchestration frameworks. It keeps agent sessions durable and replayable across disconnects, standardizes behavior per workspace through shared configuration and memory, and enables multi-agent coordination through Compozy Network when explicitly enabled. The system maintains one daemon-owned state model accessible through CLI, HTTP, web UI, and MCP, allowing people and agents to operate the same runtime without custom glue scripts or external control planes.