VibeWise skill
VibeWise turns coding with an agent into learning: it asks for your approach first, you make the design decisions, and the agent writes the code
- 4,028 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- A hook runs a Python script at the start of every session and feeds the saved notes into the agent's context
- Writes and updates note files in the project's `.vibe-wise/`, and the reset command copies them into `backups/`
Install
- Works in
Manual install
/plugin install vibe-wise@anthropic-plugin-directoryPick an install scope, confirm and restart Claude Code in the project. Then run /vibe-wise:learn.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- A hook runs a Python script at the start of every session and feeds the saved notes into the agent's context
- Writes and updates note files in the project's `.vibe-wise/`, and the reset command copies them into `backups/`
This is third-party code. Review the repository files before installing.
What it does
VibeWise installs as a Claude Code or Codex plugin and turns on a learning mode for development. The `/vibe-wise:learn` command (`$vibe-wise:learn` in Codex) starts setup: the agent asks one question at a time about what you are doing (a new project, an existing repository or a familiar one), your programming and stack experience, how often to place checkpoints and what question style you prefer. You can pick Use defaults and skip setup. For an existing repository the agent first reads the code and sketches a small system map. After that, an ordinary request to build something gets a question about your approach, and the agent waits for your reasoning. It explains unfamiliar concepts and points out concrete errors and risks, but offers solution options only when you ask for help or get stuck. Work runs through three kinds of checkpoints. At Build you work out the approach with the agent. At Design you review the design: Confirm and continue records it and moves on to planning, with no code yet. At Implementation you review the specific changes, and Implement this step authorizes the agent to make them. Afterwards the agent briefly describes what changed, how the key code works, why it follows your decision, which tests were added and what the checks showed. The amount of guidance follows your experience: a beginner gets more explanation and diagrams, an experienced developer gets more questions about constraints and failure modes. Preferences, learning notes and a project map live in a `.vibe-wise/` folder at the Git repository root, or in the folder where you started the agent when there is no Git. The mode resumes in later sessions and after context compaction: at session start a Python hook restores the notes. `/vibe-wise:reset` copies the notes into `backups/` after confirmation and restarts setup; it leaves project code alone.
Who it is for. For beginners and junior developers, and for experienced engineers learning an unfamiliar stack who want to make the design decisions themselves while the agent writes the code.
Good fit when
- You are building your first project with Claude Code or Codex and want to understand why the system is built that way instead of getting finished code
- You joined an unfamiliar repository and want the agent to sketch a system map first and then walk you through it with questions
- You are learning a new stack and are ready to work through table relationships, access boundaries and failures before writing code
- You want the agent to make no edits without your approval at each step: at the Implementation checkpoint it waits for an explicit Implement this step
- You run a learning project and want notes on covered concepts and open points kept in a `.vibe-wise/` folder next to the code
Not a fit when
- You need to close a task fast or fix production urgently: the mode puts learning above speed, and only an explicit request to skip or pause bypasses it
- You have no Python 3: it powers note restoration in Claude Code and the reset command, and in Codex the reset
- You want the agent to propose an architecture and implement it at once: the plugin waits for your decisions and gives hints only on request
- You look for post-commit exercises with memory checks: the learning-opportunities skill is closer to that
Example request
Turn on learning mode and help me build a notes app with folders: ask first how I would design itLimitations
- The plugin is version 0.1.44, an early project, and the mode's behavior may change between updates.
- Invocation is manual: the `learn` and `reset` commands are marked as not callable by the model on its own. In Codex, `$vibe-wise:learn` has to be repeated at the start of each session and whenever the agent loses learning mode.
- Notes from `.vibe-wise/` enter the agent's context, so your usual Claude Code or Codex data settings apply. The plugin does not edit `.gitignore` silently: you add the folder to the ignore list yourself.
- Without Git the notes are created in the launch folder, so start the agent from the project root. The plugin does not use state from a parent repository or another worktree.
- Claude Code needs a restart after install, and auto-update for third-party marketplaces is off by default. MIT license.
How to disable. Telling the agent "Pause learning" suspends the mode, and `/vibe-wise:learn` brings it back. To remove the plugin, open `/plugin` in Claude Code and delete VibeWise from the installed list. The notes stay in `.vibe-wise/` in your project, and you can delete that folder by hand.
FAQ
How does VibeWise differ from ordinary work with an agent?
An ordinary agent takes the requirements and picks a structure itself. VibeWise asks for your approach first, checks it against the requirements and the code, and writes only the decision you made. A feature preference sets requirements but does not approve an architecture.
What do the Build, Design and Implementation checkpoints mean?
Build: you work out the approach with the agent. Design: you review the design, confirming records it and no code exists yet. Implementation: you see the specific changes, and Implement this step authorizes them. None of the three is mandatory: when you are ready to code, the Implementation checkpoint also confirms the design.
Where are my learning notes kept and can I keep them out of commits?
The profile, progress and project map live in `.vibe-wise/` at the Git repository root. To keep them out of commits, add that folder to `.gitignore`: the plugin recommends it but does not edit the file. Claude Code and Codex share the same folder.
How do I set question difficulty and checkpoint frequency?
Experience level (Beginner, Intermediate, Advanced) changes how much the agent explains and what it asks. Checkpoint frequency (Light, Normal, Frequent) is set separately. You change either with a phrase like Use fewer checkpoints or Use multiple-choice questions, and no reset is needed.
What does the reset command do and does it touch my code?
The command shows the project and asks Cancel / Reset learning. After confirmation it copies the profile, progress and project map into `backups/` inside the notes folder and restarts setup. Source code, dependencies, Git history and other projects stay untouched.
Discussion
No questions or reviews yet.
Related
SkillsA vibe coding workflow with skills: idea research, PRD, tech design, AGENTS.md, build, verification and debugging
A spec-driven dev process for agents: gated phases, context memory, and a set of agents and skills
A skill that explains and audits code, telling you what it does and why instead of just accepting it
A Claude Code and Codex skill that offers short learning exercises during coding so the developer builds expertise, not just accepts generated code
A Grid Dynamics plugin that loads a team's shared engineering standards, architecture and workflows into a coding agent session
A CLI that gives six coding agents project context, over 80 skills, verification and resumable tasks across sessions
README in short
The repository's front page opens with the idea of you build, AI writes and the requirements: Claude Code or Codex and Python 3. Two ways to install in Claude Code, the Codex install and a sample learning session about a notes app with folders follow, showing all three kinds of checkpoints. Separate sections explain tuning for your experience level, note storage in `.vibe-wise/`, reset and updating. The repository also holds tests for the reset script and the startup hook, a write-up of the demo session, a development guide and Codex-specific guides.
README badge
Original SKILL.md
--- name: learn description: Activate or resume learning-first development. You lead the design; Claude gives feedback, explains concepts, asks follow-ups, and writes the agreed code. disable-model-invocation: true --- # VibeWise Learn mode Activate learning mode in the main conversation. Read [behavior.md](behavior.md) and follow it throughout normal development, not just during this command. The learner owns the design. Ask for their approach and wait. Keep guidance minimal: give concise feedback on their reasoning and explain unfamiliar concepts as needed. Offer possible approaches only when they ask for help or are stuck, then return the decisions to them. Learning and learner control take priority over build speed.