nykooi1/vibe-wise

VibeWise – AI‑guided coding with a learning focus

What it is

• A plugin for Anthropic’s Claude Code and OpenAI’s Codex that adds a learning mode to the code‑generation workflow.
• The AI first asks you to outline the approach, discusses trade‑offs, and explains any unfamiliar concepts before it writes any code.
• After you confirm the design, the AI generates the implementation, then gives a concise report of what changed and why.

Who it’s for

• Junior developers or anyone who wants to deepen their understanding while building.
• Experienced engineers who need a structured way to explore a new stack or architecture.
• Teams that want to keep design decisions explicit and retain ownership of the code.

How it works

1. Installation – Install via the built‑in Anthropic Directory in Claude Code (/plugin install vibe-wise@anthropic-plugin-directory) or through the GitHub marketplace. For Codex, add the marketplace and install with codex plugin add vibe-wise@vibe-wise.
2. Start a learning session – Run /vibe-wise:learn (Claude Code) or $vibe-wise:learn (Codex). The plugin walks you through a series of checkpoints:
   • Build – You reason out the problem with the AI (e.g., how notes and folders should relate).
   • Design – The AI proposes a design; you can Confirm and continue or Discuss.
   • Implementation – Once you approve, the AI writes the code, runs tests, and reports the changes.
3. Notes & context – All decisions, diagrams, and explanations are saved in a hidden .vibe-wise/ folder inside your project. These notes are automatically added to the AI’s context in future sessions, so the assistant remembers your prior discussions.
4. Preferences – You can set a learning level (Beginner, Intermediate, Advanced) and tweak checkpoint frequency, question style, etc., on the fly.
5. Reset / Update – Use /vibe-wise:reset to start over, or update the plugin via the usual Claude Code or Codex plugin‑update commands.

Key features

• Interactive design discussions before any code is generated.
• Explain‑first approach – AI explains concepts, trade‑offs, and unfamiliar pieces.
• Checkpoint system – Structured yet optional stages (Build, Design, Implementation) with clear “Confirm” or “Discuss” actions.
• Persistent learning notes – Stored locally, versioned, and ignored by Git by default.
• Zero external dependencies – Only Python 3 is required; the plugin runs inside the host IDE/assistant.

Why it matters

• Turns AI‑assisted coding from a black‑box code dump into a teach‑you‑while‑you‑build experience.
• Helps developers retain ownership of architectural decisions and understand the generated code.
• Provides a reusable workflow that can be applied to any project, from simple scripts to full‑stack apps.

License

• Open‑source under the MIT license, allowing commercial use, modification, and redistribution.