VBW plugin
Vibe Better With Claude Code
A Claude Code plugin where a team of agents builds the work and a bash kernel accepts nothing until checks pass.
- 88 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Agents write code, make commits and run commands and stop less often in overnight and auto modes
- Guards limit dangerous commands but do not replace the Claude Code sandbox and permission rules
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/vibe-better-with-claude-codeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents write code, make commits and run commands and stop less often in overnight and auto modes
- Guards limit dangerous commands but do not replace the Claude Code sandbox and permission rules
This is third-party code. Review the repository files before installing.
What it does
VBW wraps Claude Code in a process where a requirement counts as done only after an approved check passes on a clean copy of the code or you accept it yourself. One `/vbw:vibe` command runs the loop: agree requirements, plan, write tests that fail first, implement and verify. A team of agents with one role each works: Architect makes the decisions that deserve a person, Lead splits work into plans and writes tests, Dev implements one plan with a commit per task, QA reads a finished phase, Scout looks things up, Debugger reproduces a bug and Docs writes documentation. Parallelism is set by the `vbw` kernel in bash, jq and git, and agents do not control it: it decides how many Devs work at once and how much checking is needed. `/vbw:loop` works overnight through named items with a spend ceiling.
Who it is for. People who build a project with Claude Code and do not want to take done at face value, both non-readers of code and seasoned developers.
Good fit when
- A new feature needs requirements with acceptance criteria before the first line of code.
- Work must be split into plans with disjoint files and run in parallel by several Devs.
- Before merging, an independent check of a phase against its goal and recorded decisions is needed.
- A list of small tasks should run overnight with a spend ceiling and a report in the morning.
Not a fit when
- A one-line fix is needed: the agreement and check process is too heavy.
- Work happens on Windows without WSL: the kernel is written in bash and does not run there directly.
- Another agent is needed: the plugin targets Claude Code 2.1.286 or newer on Sonnet, Opus or Fable.
Example request
/vbw:vibe add magic-link sign-in to the app and show the requirements before work starts.Limitations
Needs Claude Code 2.1.286 or newer on Sonnet, Opus or Fable, plus macOS or Linux; on Windows run Claude Code inside WSL. The kernel uses bash, jq and git. Updates do not reach the plugin by themselves: VBW makes no network call, and auto-update is enabled in Claude Code for `vbw-marketplace`. Guards against `rm -rf`, force pushes and `git reset --hard` are guard rails. The Claude Code sandbox and permission rules remain the security boundary. In auto mode agents work longer between stops, and the plugin does not make `--dangerously-skip-permissions` safe.
How to disable. Disable the plugin with `/plugin uninstall vbw@vbw-marketplace` and restart Claude Code. The `.vbw/` directory in the project with requirements and decisions is deleted by hand.
FAQ
How do I install it?
With two separate commands: `/plugin marketplace add swt-labs/vibe-better-with-claude-code-vbw` and `/plugin install vbw@vbw-marketplace`. Pasted together they read as one and break the URL.
How do I update the plugin?
With `/vbw:update`, then `/reload-plugins`. For Claude Code to install new versions itself, enable auto-update for `vbw-marketplace` in the Marketplaces tab of `/plugin`.
What does the kernel do?
`vbw`, built on bash, jq and git, decides what counts as done, how many Devs work at once and how many checks a phase gets. Agents do not overrule it.
What protects against dangerous commands?
Guards check every command before it runs and read what the shell would actually execute. While `/vbw:loop` is on, `git push`, releases, publishes and deploys are denied to everyone.
How do I verify the installation?
`/vbw:doctor` checks the rest, including whether auto-update is on.
Discussion
No questions or reviews yet.
Related
PluginsA workflow template for Claude Code and five more agents: a backlog task, a diff review and a commit.
A Claude Code plugin links planning, implementation and task checks, retains progress and controls how independently agents work.
A Claude Code plugin for development with separated roles: spec, review, TDD, QA and verification by different agents
A plugin for Claude Code: you read one spec and parallel agents ship it to a commit with green tests.
Claude Code plugin with 17 skills: task sizing, design review, TDD, fresh-context subagents and a PR after three human decisions.
A plugin that turns Claude Code into a 16 agent team: documentation first planning, TDD and 9 pre-merge quality gates.
README in short
Covers the idea, agent roles, the agree and verify loop, installation, supervised and auto modes, the night loop, safety and configuration.