computerlovetech/agr

What it solves

It eliminates the manual process of copying and sharing AI agent "skills" across teams and tools. Instead of managing skill files by hand, which leads to version drift and inconsistency, it provides a centralized way to treat these skills as software dependencies.

How it works

It functions as a package manager for agent skills. Users can install skills directly from Git repositories (using a owner/repo/skill handle) or from local directories. The tool manages a manifest file (agr.toml) and a lock file (agr.lock) to ensure every team member is using the exact same version of a skill. It automatically detects and installs these skills into supported AI tools like Claude Code, Cursor, and GitHub Copilot.

Who it’s for

Teams of developers and AI users who share a common set of agent skills and want to ensure consistent behavior across different machines and AI tools.

Highlights

• Cross-tool compatibility: Installs skills into multiple tools including Claude Code, Cursor, Codex, and Copilot.
• Version pinning: Uses a lock file to prevent silent upstream changes from breaking agent workflows.
• Git-based distribution: No central registry required; any public Git repo can serve as a source for skills.
• One-command onboarding: New team members can set up their entire agent environment using agr sync.
• Ephemeral execution: The agrx command allows users to try a skill once without permanently installing it.