# AI Skill Distribution The practice of packaging, sharing, and deploying [[AI Agent Skills]] beyond the project or person that created them. As skills become the building blocks of AI agent systems, distribution becomes essential for reuse at every scale: personal, team, enterprise, and community. ## Distribution scopes - **Personal reuse**: sharing skills across your own projects and vaults. The simplest form: copy skill files between repos, or maintain a personal skill library - **Team sharing**: distributing skills within a team via shared repositories, internal registries, or monorepo conventions - **Enterprise deployment**: governed rollout of approved, vetted skills across an organization, with access control and compliance review - **Community/marketplace**: publishing skills for others to discover and use. Emerging ecosystem of skill registries and marketplaces ## Current state Today, AI skills are files. Distribution is file-based: - Skills live as Markdown files (SKILL.md) in project repos or dedicated skill libraries - Sharing means copying files, cloning repos, or using git submodules - [[Claude Code Plugins]] provide a more structured packaging format with plugin.json manifests - [[Skills.sh]] is an early skill registry for discovery and sharing - No universal standard yet for skill packaging, metadata, or dependency declaration ## Emerging ecosystem The ecosystem is evolving toward: - **Skill marketplaces**: browsable registries where skills can be discovered, rated, and installed - **Standardized packaging**: consistent metadata (name, description, version, dependencies, compatibility) - **Versioning and updates**: semantic versioning for skills, update notifications, and backward compatibility - **Trust and curation**: verified publishers, community ratings, security audits for skills that access tools or data ## Challenges - **Portability**: a skill written for [[Claude Code]] may not work in [[Cursor.com]] or other AI tools without adaptation - **Context assumptions**: skills often assume specific vault structures, tool configurations, or domain knowledge - **Quality assurance**: no standardized testing framework for skills yet - **Security**: skills can execute arbitrary tool calls. Distributing untrusted skills is a security risk ## References - ## Related - [[AI Agent Skills]] - [[AI Agent Distribution]] - [[AI Skill Composability]] - [[Claude Code Plugins]] - [[Skills.sh]] - [[AGENTS.md (File Convention)]] - [[AI Agent Harness]] - [[Context-as-Code]] - [[Model Context Protocol (MCP)]] - [[Convention over Configuration]] - [[Barrel Pattern]]