skillzero

skillzero is an open-source CLI tool built by Jacob "Kurt" Groß (kurtextrem) that helps developers manage AI agent skills more efficiently by reducing unnecessary token consumption. It runs via npx skillzero and works with popular agent harnesses including Claude Code, Codex, Cursor, OpenCode, and Devin. The project is licensed under MIT and available on GitHub.

What It Is

skillzero addresses a specific problem in AI-assisted development: most agent harnesses expose every installed skill's name and description to the model at the start of each session, even when those skills are irrelevant to the current task. With many skills installed, this bloats the context window, increases cost, and can trigger accidental skill invocations. skillzero lets developers categorize their skills so only the most relevant ones are visible to the agent by default.

Three-Category Skill Management

skillzero divides skills into three categories:

• Unmanaged: Skills left as-is; their names and descriptions remain visible to the agent at all times. Best for everyday guidance like repo rules, accessibility standards, or frequently used framework skills.
• Hidden: Skills completely hidden from the agent's automatic discovery. They consume no tokens in context but remain manually invokable like commands. Useful for migration guides, niche framework rules, or one-off platform guidance.
• Collections: A bundle of related skills grouped under a single entry point. The agent sees only the collection's name and description; it reads the individual skills lazily only when the topic matches. Examples include design, writing, marketing, or frontend/backend skill sets.

How Collections Work

When a collection is created, skillzero generates a SKILL.md file that tells the agent when to use the collection and where to find its member skills. The original skills remain in their own folders but are hidden from automatic discovery. The generated collection skill acts as an index, listing each member skill's name, description, and source path in a Markdown table. This lazy-loading approach means the agent only reads the full skill content when it determines the topic is relevant to the current task.

CLI Workflow

skillzero is invoked from the command line with no installation required beyond a package manager:

• npx skillzero — interactive picker for skills folder and management settings
• npx skillzero --dry-run — preview changes without writing any files
• npx skillzero collections — edit existing collections
• npx skillzero undo / skillzero redo — revert or re-apply changes
• skillzero update — update skills via Vercel's skills CLI, then re-apply skillzero settings

skillzero stores its managed state in a skillzero/ directory inside each skills root and writes generated collection skills as skillzero-<collection-id>/ folders alongside ordinary skills.

Update: v0.2.0

The latest release is v0.2.0, published in September 2026. The repository was created in August 2026 and last pushed in September 2026, indicating active early development. The project is written in TypeScript and uses the Aube package manager for its own build and test pipeline. The GitHub repository currently has MIT licensing and is open for community contributions.