soumatheusgomes/vibe-coding-toolkit

🎧 Vibe Coding Toolkit – A practical workflow for AI‑assisted development

What it is – A collection of documented tools, prompts, and configuration templates that let developers use large‑language‑model coding assistants (Claude Code, OpenAI Codex, etc.) as a disciplined member of a software team. The repo does not ship a new model; it ships a process and the glue (plugins, hooks, lint rules, sub‑agent orchestration) that make the process repeatable in production.

Why it matters – Many AI‑coding guides stop at “paste a big prompt and hope for the best.” This toolkit shows a real‑world flow that solves concrete pain points the author faced in a production fintech project:

• Session continuity (memory between runs)
• Preventing agents from over‑generating or diverging from the request
• Parallel sub‑agents that specialise (review, DB, tests) without stepping on each other’s files
• Automatic quality‑gate enforcement (ESLint/Biome) that turns warnings into tracked errors
• A lightweight “Superpowers” discipline that forces a brainstorm → plan → implementation → review loop before any code is written.

Core pieces

Piece	What it does	How you use it
Superpowers (docs/tools/01‑superpowers.md)	Enforces the brainstorm‑plan‑code‑review cycle; the single most impactful habit.	Install the Claude‑Code plugin superpowers@claude-plugins-official and invoke the skill before asking the model to code.
Sub‑agent orchestration (02‑subagent‑orchestration.md)	Runs multiple specialised agent instances in waves so they never edit the same file concurrently.	Follow the protocol described in the doc; the repo includes a template for the wave‑dispatch prompt.
RTK token proxy (03‑rtk-token-proxy.md)	CLI that rewrites repetitive commands (git status, diff, log…) into a compact token‑saving form.	Use the provided script as a drop‑in replacement for those commands inside a session.
Ponytail & Caveman personas (04‑ponytail.md / 05‑caveman.md)	Simple decision‑making and communication policies that keep the agent focused and concise.	Add the persona prompts to your Claude/Codex session before work begins.
Quality gates (ESLint/Biome) (06‑eslint‑biome‑quality‑gates.md)	Turns selected lint warnings into hard errors, enforces a 350‑line‑per‑file ceiling, and records violations.	Run the ready‑made prompt 08‑eslint‑quality‑gates‑install.md (or copy the config files from templates/eslint/).
Graphify & Context7 (07‑graphify.md / 12‑context7.md)	Builds a persistent knowledge graph of the codebase and injects up‑to‑date library docs into the model’s context.	Deploy the optional MCP servers described; the toolkit shows how to hook them into Claude.
Obsidian memory (08‑obsidian‑memory.md)	Long‑term project notes stored in an Obsidian vault, accessed by the agent via an MCP protocol.	Initialise the vault and point the MEMORY.md index to it.
Prompt library (docs/prompts/*.md)	Ready‑to‑copy prompts for common tasks: lint clean‑up, multi‑agent code review, file‑size refactoring, etc.	Paste the whole prompt into Claude/Codex (or any LLM that can read a URL) and let the agent execute it.
Templates (templates/…)	Boiler‑plate CLAUDE.md, ESLint configs, hook examples, and a minimal settings.json for the CLI.	Copy the needed files into your project and adjust paths as described in the Playbook.

Getting started

1. Install Claude Code and the official plugin marketplace (the README’s one‑liner).

   npm install -g @anthropic-ai/claude-code
   /plugin marketplace add anthropics/claude-plugins-official
   /plugin install superpowers@claude-plugins-official
   cp templates/CLAUDE.md.template CLAUDE.md
   

2. Read the Playbook – docs/02-playbook-onboarding.md. It walks you through a full end‑to‑end example project, showing how each tool fits into the timeline (install → setup → brainstorm → plan → sub‑agent waves → review → ship).
3. Pick a single tool if you just need a quick win. For example, to enforce the 350‑line rule, run the one‑line prompt shown in the README or copy the ESLint config from templates/eslint/.

Who should use it

• Developers who already have access to Claude Code, Codex, or a comparable LLM‑coding assistant and want a repeatable, production‑grade workflow.
• Teams that have struggled with “prompt‑and‑hope” sessions, lost context, or agents overwriting each other’s files.
• Anyone comfortable with a modest amount of CLI work and willing to read short Portuguese docs (the core concepts are explained inline, and the actual prompts are in English).

Maturity & status

• The repository is a methodology extracted from a private fintech project (the original code is under NDA). The author says the practices have been used in production, but the repo itself contains only the documentation, prompts, and configuration templates – no compiled binaries.
• The history is short because the repo is meant to be a living guide; updates are expected when the author refines the workflow or adds new plugins.

License – MIT (see LICENSE).

---

Bottom line – Vibe Coding Toolkit is a concrete, open‑source playbook for turning AI coding assistants into disciplined teammates. It supplies the missing glue (plugins, prompts, lint gates, sub‑agent orchestration) that lets developers move from ad‑hoc prompt‑hacking to a repeatable, production‑ready development flow.