According to the DynaBeat Monitor, Matthew Phillips, co-founder of the modern front-end framework Astro, has open-sourced Rosie, a command-line tool designed to manage an AI agent skill set. Using rosie install owner/repo to install a skill from GitHub, Rosie will automatically detect which coding agents are locally installed and synchronize the skill over.
Rosie currently supports 10 types of agents, including Claude Code, Cursor, Codex, Windsurf, Aider, Zed, Continue, Cline, OpenCode, and GitHub Copilot. The key feature is the lockfile mechanism: installation records are written to .agents/rosie.lock, one line per entry, in a format friendly to git diff, allowing direct submission to the version control system. The skill folder itself can be placed in .gitignore, and team members can clone and run rosie install once to restore the setup.
Version management is divided into two modes: when no version is specified, it uses auto, and rosie update will automatically upgrade to the latest semver tag; when a version is specified with @ref, it uses pin, only refreshing the version with an unchanged SHA. Upon local installation, skills exist in the .agents/skills/ directory and are synchronized to each agent's skill directory using symbolic links; it also supports --global to install directly to the user-level directory.
The entire tool is written in C, relies on libcurl and libarchive, supports Homebrew, apt, AUR, FreeBSD pkg, and source compilation, and is under the BSD-3-Clause license.
