GitHub - Balcon2/claude-harness-skills: A Claude Code skill library for faster, cheaper, and more controlled AI coding workflows. · GitHub
Skip to content

About

A Claude Code skill library for faster, cheaper, and more controlled AI coding workflows.

Resources

Contributing

Stars

0 stars

Watchers

1 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation

Claude Harness Skills

A big collection of Claude Code skills for cleaner, faster and more controlled AI coding workflows.

I made this repo because Claude Code can become much more useful when it follows small, focused workflows instead of trying to figure out everything from scratch every time.

The goal is simple:

  • save tokens
  • avoid unnecessary context
  • make Claude Code work more structured
  • improve debugging, refactoring, testing and repo analysis
  • keep workflows reusable across projects

What is inside?

This repo contains a large set of Markdown-based skill files for Claude Code.

The skills cover areas like:

  • token saving
  • repo exploration
  • debugging
  • refactoring
  • frontend work
  • backend work
  • testing
  • Git workflows
  • security checks
  • database tasks
  • DevOps basics
  • project planning

Most skills are written as small, focused instructions. They are not meant to be huge prompts. They are meant to guide Claude Code in a specific situation.

Why?

Claude Code is powerful, but sometimes it reads too much, explains too much, or changes more than needed.

These skills are made to keep it focused.

Instead of saying:

“Fix my project”

you can use a skill that tells Claude exactly how to approach the task, for example:

/bug-fix-fast
/repo-scout
/token-budget
/safe-refactor

That makes the workflow more predictable and usually cheaper in tokens.

Token saving

A lot of these skills are manual-only.

That means Claude does not automatically load them all the time. You call the skill only when you need it.

This helps because loading too much context defeats the whole purpose of saving tokens.

Most skills use:

disable-model-invocation: true

So you stay in control.

Folder structure

.claude/
└─ skills/
   ├─ token/
   ├─ repo/
   ├─ coding/
   ├─ frontend/
   ├─ backend/
   ├─ testing/
   ├─ git/
   ├─ security/
   ├─ database/
   ├─ devops/
   └─ project/

Each skill has its own folder with a SKILL.md file.

How to use

Copy the skills into your project:

mkdir -p .claude/skills
cp -r .claude/skills/* your-project/.claude/skills/

Or use them globally in your Claude Code setup if you prefer.

Then call a skill inside Claude Code:

/token-budget
/repo-scout
/bug-fix-fast

Example skills

Some examples:

/token-budget

Keeps Claude short and focused.

/repo-scout

Helps Claude understand a repo without reading everything.

/bug-fix-fast

Focuses on fixing a bug with minimal changes.

/safe-refactor

Refactors code without randomly rewriting the whole project.

/commit-clean

Helps create clean commits and summaries.

Who is this for?

This repo is useful if you use Claude Code a lot and want more control over how it works.

It is especially helpful for:

  • developers
  • students
  • indie hackers
  • people building AI-assisted workflows
  • anyone experimenting with harness engineering

Note

This is not meant to be a perfect official framework.

It is a practical skillpack for experimenting, improving workflows and making Claude Code easier to control.

Feel free to edit, remove, rename or improve the skills for your own setup.

License

Use it, change it, remix it.

Just don’t pretend you wrote everything from scratch if you directly copy the whole thing.

About

A Claude Code skill library for faster, cheaper, and more controlled AI coding workflows.

Resources

Contributing

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors