Show HN: Spm – a package manager for agent skills per-repository (camunda.github.io)

Spm, a new package manager designed for agent skills, has been introduced to streamline the integration of AI functionalities into projects. Developers can now declare AI skills as git dependencies within an ai.json file, facilitating seamless access to tools like Amp, Claude Code, and OpenAI Codex without permanently adding these skills to their codebase. The skills are managed similarly to traditional code dependencies, allowing users to pin them by tag, branch, or immutable SHA in ai.lock, ensuring that collaborative projects maintain consistency in their AI skill sets. The significance of Spm lies in its innovative approach to dependency management, offering a multi-vendor projection system where a single declaration can independently resolve across different AI platforms. Notably, the skills are stored in a gitignored format, avoiding clutter in repositories. This system promotes reproducible installs, ensuring every team member has access to the same skill versions through a shared global fetch cache, thereby enhancing workflow efficiency. With its cross-platform compatibility as a single self-contained binary, Spm positions itself as a valuable tool for the AI/ML community, simplifying the integration of diverse AI capabilities across various development environments.