agentpack

This Skill addresses the chaotic management of skills and plugins across different AI coding agents (such as Claude Code, Cursor, OpenCode, etc.). In real-world development, each agent has its own distinct file layout, configuration format, and discovery rules. When teams use multiple agents or need to share a common set of tools, they often resort to manually copying and pasting the same files into each agent's configuration, leading to version drift, redundant effort, and a lack of a unified lock file to confirm the exact versions everyone is using.

Usage is straightforward: first install agentpack (recommended via a one-click Homebrew install), then create an agentpack.toml file in your project root to declare the required skill and plugin dependencies. Running the agentpack command will automatically resolve dependencies, lock them to precise commit versions (generating a pack.lock file), cache the content, and then re-render the corresponding artifacts into the target agent's format. When you launch the agent, all skills will be ready. The entire process eliminates the need to manually copy any files into the project directory or global agent configuration.

It is ideal for teams or individuals who maintain multiple AI coding agents (like Claude Code, Cursor, Codex, etc.) simultaneously. This is especially useful for developers already using various tools for coding assistance who want to unify skill versions and configurations, or for teams needing to share a validated set of skills across projects. For development workflows aiming to avoid manual synchronization and reduce configuration errors, agentpack offers an elegant solution.

It is recommended to use agentpack uniformly for managing skill dependencies in team collaboration projects and to regularly run agentpack update to keep skill versions synchronized. Note that it primarily targets CLI and terminal environments and does not currently support GUI operations. It also relies on a Rust-compiled binary, so initial installation may require environment variable configuration. For simple scenarios using only a single agent, introducing an additional package manager might be unnecessary.