soumatheusgomes/vibe-coding-toolkit

What it solves

It addresses the inefficiency and unpredictability of "prompt-and-hope" AI-assisted development. Instead of relying on luck, it provides a structured, production-tested workflow to integrate AI agents (specifically Claude Code and OpenAI Codex) into a development team as reliable contributors rather than unpredictable assistants.

How it works

The toolkit implements a multi-layered system that moves from planning to execution:

• Planning & Intent: Uses tools like "Superpowers" to force a cycle of brainstorming, planning, and review before any code is written.
• Execution via Subagents: Employs a "parallel wave" orchestration protocol where a main agent plans and delegates tasks to specialized subagents (e.g., reviewers, DB experts, testers) to prevent file collisions and commit disputes.
• Quality Control: Integrates "Quality Gates" using ESLint and Biome to promote warnings to errors, ensuring architectural boundaries are maintained.
• Context & Memory: Uses a two-tier memory system (a MEMORY.md file for quick lessons and an Obsidian vault via MCP for long-term project knowledge) and tools like Graphify for knowledge graph mapping of the codebase.
• Communication: Applies personas like "Ponytail" (the efficient, lazy senior engineer) and "Caveman" (concise, no-fluff communication) to govern how the AI thinks and speaks.

Who it’s for

Developers of all skill levels who use Claude Code or OpenAI Codex and want a professional, structured workflow to replace trial-and-error prompting.

Highlights

• Superpowers Plugin: A core discipline that ensures agents explore intent and plan before implementing.
• Parallel Wave Orchestration: A protocol for running multiple independent subagents simultaneously without overlapping files.
• Two-Tier Memory: Combines an immediate project index with a long-term Obsidian-based knowledge base.
• MCP Integration: Leverages Model Context Protocol (MCP) servers like Context7 and Chrome DevTools for real-time documentation and browser diagnostics.
• Ready-to-use Prompts: Includes templates for project sanitation, lint burndown, and multi-agent code reviews.