arjar88Claude Code
curl -fsSL https://raw.githubusercontent.com/vibe-md/vibe.md/7cb0620fbd80f9829155d8866328b58cd18d5aa7/scripts/install.sh \
| bash -s -- --skill code-hygiene \
--agent claude-code --ref 7cb0620fbd80f9829155d8866328b58cd18d5aa7Code Hygiene
AvailableKeeps AI-written code real, used, and written once. Checks for an existing helper before creating a new one, verifies every third-party API against the installed version before using it, and cleans up code the change left orphaned. Use when adding functions, hooks, components, or utilities; when calling or upgrading a third-party package or editing framework config; after finishing a feature or abandoning an approach; or when an error says something is undefined, not a function, not exported, or an unknown option.
Code Hygiene keeps what your agent writes real, used and written once. Before it creates a new helper, it searches your project for one that already exists. Before it uses a third-party API, it checks that the API exists in the version you have installed. When the work is done, it lists the code the change left unused and asks before removing anything.
ProductivityAI / AgentsFrontend
- Claude Code
- Codex CLI
- GitHub Copilot
- Cursor
- Gemini CLI
Maintained by
The problem
AI agents write before they search, so you end up with a third formatDate or a second Button. They code from memory of an older library, so you get hooks, props and config keys that don't exist in your version. And when they change direction, the abandoned files and packages stay behind. The project gets bigger, harder to read and more likely to break.
What it does
Search before creating
Your agent looks for an existing function, hook or component, by name and by behavior, and checks the packages you already have. It reuses or extends what exists instead of writing a copy.
Verify every third-party API
Each import, prop, option and config key is checked against the installed version, using its type definitions and changelog, not the agent's memory.
Sweep what the task orphaned
When the work is finished, it lists the files, exports, imports and packages that nothing uses anymore. It proposes the removals and waits for your approval.
A short hygiene report
Each task ends with what was reused, which APIs were verified, what was cleaned up, and which checks passed. You can see exactly what was done.
Installation
One command
Install directly into your project.
Open Terminal in your project folder, then run this command. No ZIP, moving files, or prompt needed.
- 1Open your project folder in Terminal
- 2Run the command above
- 3Reload your agent and ask to use Code Hygiene
Works on macOS, Linux, and Windows through WSL or Git Bash. The installer checks destinations and stops if existing files conflict.
Usage examples
Type these to your agent in plain language. Here's what to expect.
you › Add a debounced search input to the products page.
Expected workflow
- 01Searches the project and finds an existing useDebounce hook.
- 02Reuses it instead of writing a new one.
- 03Checks the input component's props in its installed type definitions.
- 04Reports what was reused and that no new helper was created.
you › Let's drop the custom carousel and use FlashList.
Expected workflow
- 01Switches to FlashList, then searches routes, config, tests and Storybook for anything still using the old carousel.
- 02Lists the unused component and package, and asks before removing them.
- 03After you approve, uninstalls the package with the package manager and reruns the build.
How it works
The skill follows the order of a normal task: search before writing, verify while writing, and sweep after writing. When an error says something is not a function, not exported or an unknown option, your agent treats it as a missing API and checks the installed version first.
It works in any language, and the examples target JavaScript and TypeScript projects. It can use knip and depcheck to find unused code, and it asks before installing either.
Safety and limitations
Required permissions
- Read your project and its installed dependencies.
- Run searches and your type check, lint and build commands.
- Edit the files that the current task changes.
Potentially destructive operations
Read what your agent is about to run before approving it. These can lose work:
- Files, exports and packages are only deleted after you approve the list.
- Packages are removed with the package manager, and your agent will not install, upgrade or downgrade one just to make an API exist.
Limitations
- It works through instructions to your agent, so it relies on the agent following them. Review the cleanup list before approving.
- Unused-code tools can report false positives, so each item is checked by hand before it is proposed.
- Duplicates that existed before the task are reported, not merged.