Expand your skills registry
A skills registry starts as one catalog of every skill in your organization, with owners and groups. If you haven't set one up yet, start with get started with a skills registry.
The skill lifecycle
Every skill in the registry moves through three stages, from the moment it is shared until it is retired. The capabilities on this page let you manage each stage.
| Stage | What happens | Covered by |
|---|---|---|
| Publish | Developers share skills with the organization through one golden path workflow. A new skill appears in the registry right away with a pending status. | Ship new skills through one golden path |
| Certify | When a skill is created or updated, a certification workflow assesses its quality, records the result on scorecards and recommends whether to approve, iterate or reject it. | Certify skills to meet industry and org standards |
| Retire | Skills with no usage, negative ROI (they cost more than the work they replace) or low scorecard levels are suggested for deprecation. | Measure the ROI of skills in your org registry |
Add capabilities to your registry
Once the registry is in place, you can add governance and insight one capability at a time. Each guide below builds on the registry, and you can pick the ones that matter to you, in any order.
Ship new skills through one golden path
The ship new skills through one golden path guide gives developers a single self-service way to publish a skill. An AI node extracts the skill's metadata, a pull request opens automatically, and the skill is registered as pending until the pull request is merged. Reviewers approve or reject pending requests from the Skills Lifecycle Control dashboard.

Distribute skills through a virtual MCP server
The seamlessly distribute skills across your organization guide builds a virtual MCP server with three tools: list_skills returns the name and description of every skill the connected user can see, load_skill loads a skill when a task matches it, and publish_skill runs the publishing golden path. You roll it out to coding agents, Port AI agents or individual users, and the server's instructions make the agent sync the list of skills at the start of every session.
Certify skills to meet industry and org standards
The certify skills to meet industry and org standards guide adds an AI review of each skill's description, a discoverability scorecard, and an approve, iterate or reject recommendation with a Slack notification. The Skills Registry Health dashboard shows the result across the organization.

Avoid duplicate skills in your org registry
The avoid duplicate skills in your org registry guide adds a similarity check before a skill is published, with a prompt to publish anyway or cancel, and a uniqueness check during certification that links similar skills.

Bundle skills into plugins
The bundle skills into plugins guide adds a plugin blueprint and AI suggestions for which plugin a skill belongs in, so one install covers a whole role or team.

Measure the ROI of skills in your org registry
The measure the ROI of skills in your org registry guide links usage and spend data to each skill to calculate cost per run, an ROI scorecard and total savings, so you know which skills to invest in, standardize or retire. To track adoption, sharing and spend over time, add the visualize Claude skills adoption dashboard.

Skill versions
We base our approach to skill versions on how you manage versions of software packages, such as npm packages. You tag each release, consumers can pin a specific version when they need to, and the latest version is the default.
- Serve the latest version. The version on a skill's default branch is the one developers and agents should load. Evaluate a new or updated skill while it is still
pending, before its pull request is merged, not after it reaches users. - Track each skill's version in the registry. The
versionproperty is mapped from theversionfield in the skill'sSKILL.mdfrontmatter, and the production readiness scorecard from set up a skills registry checks that every skill has one. - Pin a specific version when you need to. Tag each skill release in Git. Skill installers such as
npx skillsby Vercel and APM by Microsoft let you install and pin a specific version in a repository.
The APIs of agent providers such as Claude and Cursor don't currently report which version of a skill an agent loaded or when it was installed, so the registry can't identify agents that still run an older version.
To track this, model each skill version as its own entity with the version in its identifier, for example check-pr-v1.2.3, and relate every versioned entity to the main skill entity.
Related docs
- Skills overview - the skill blueprint, built-in skills and best practices.
- Ingest skills from Git - mapping configuration for GitHub and GitLab.
- Enrich Port AI with skills - how Port AI loads skills at runtime.
- Skills usage analytics - skill usage across Cursor and Claude.