Back to directory

Vibe Coding Toolkit

A curated set of Claude Code plugins, subagent orchestration patterns, quality gates and ready-to-copy prompts, extracted from a real production project, for developers who want a structured AI-coding workflow instead of pasting one large prompt.

Vibe Coding Toolkit

Vibe Coding Toolkit is a collection of practices for working with an AI coding agent as an actual part of a development team rather than an unsupervised assistant that needs constant correction. Each piece exists because it solved a specific real problem: sessions that lost track of what they were doing, an agent that built more than was asked for, lint warnings nobody ever cleared, and lessons that had to be relearned every new session because nothing was recorded.

The toolkit centers on the open-source Superpowers plugin for Claude Code, which the author considers its most important single component: the discipline of exploring a request's intent, planning, and only then writing code, instead of acting on the first plausible reading of an open-ended request. The practices themselves come from a private production fintech project under a non-disclosure agreement, so what is published here is the extracted and generalized method rather than that project's own history.

It targets developers using Claude Code or, to a lesser extent, OpenAI's Codex, and is licensed under MIT.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.