Introducing Skills in Port

TL;DR

Skills are reusable instruction sets (like an incident triage playbook or PRD checklist) that load automatically into AI agents based on context. Instead of generic responses, your agents follow your organization's specific processes. 

Create skills in Port's catalog, and they're instantly available across Cursor, Claude, and any MCP-compatible client.

Why we built it

Your engineering team has an AI agent. Developers ask questions like "help me analyze this incident" or "scaffold a new service." The agent responds, but the answers are generic. Three developers ask the same question and get three different approaches, none matching your actual process.

-

You've tried MCP tools. They give access to PagerDuty, AWS, GitHub. But tools are just pipes to data and actions. They don't know how to do things your way.

You've tried prompts. Developers type /scaffold-service and get a template. But prompts require users to know they exist, and one prompt can't cover every variation of a question.

So you decide to build skills yourself. Now you need a repository to maintain them, a CI/CD pipeline to keep agents updated, and a way to distribute them across every AI client your teams use. Each client works differently: Claude needs a zip file uploaded manually, Cursor needs a .cursor/skills folder in every repo. There's no sync, no permissions, no visibility into who owns what.

What started as "let's give the agent some expertise" becomes a maintenance burden. Skills quickly get out of control.

- We think there's another way. When skills live in Port, you get single-source distribution to any AI client. Time to distribute a skill across your team drops to minutes, with clear governance and access controls built in.

-

What it is

Skills are domain expertise that loads automatically based on context.

A skill contains:

• Description: When to trigger it (e.g., "when the user asks about incident analysis")
• Instructions: What the agent should do, step by step
• References: Supporting docs, templates, and formats the agent can use

When a developer asks "help me write an RCA," the agent checks available skills, finds one matching that intent, loads its instructions, and responds using your RCA process. Same question, consistent answer, every time.

In Port, skills are entities in your software catalog. You create a Skills blueprint, add skills as entities, and Port's MCP server distributes them to connected clients.

Managing skills in Port means:

• Available everywhere: Any MCP-compatible AI client (Cursor, Claude, VS Code) gets your skills through one connection
• Permission-controlled: Users get only skills relevant to their role through dynamic permissions
• Fully discoverable: Git integration lets you manage skills alongside your code, with clear ownership in the catalog

Where it's useful

• Deploy to production A developer asks "how do I deploy my service to production?" The agent loads your deployment skill, which includes your specific approval process, environment checks, and rollback procedures.
• Day planner A developer asks "what should I work on today?" The agent loads your day planner skill, pulls their open PRs, assigned Jira tickets, and on-call status, then formats a prioritized daily plan.
• RCA reporting After an incident, a developer asks "help me write the RCA." The agent loads your RCA skill with your template, required sections, and links to relevant runbooks.

How to start using it

To start using skills, follow our Skills documentation through these steps:

1. Create the Skills blueprint in Port's data model

2. Connect your AI client to Port MCP and ask it to create a new skill in Port (we have a Skill Creator skill!)

‍

3. (optional) Define permissions on who can access each skill

4. Done! The skill is available to your engineers through Port's MCP server

What's next

Skills bring your organization's expertise directly into AI agents, no matter which client your developers prefer. 

Next, we’re looking into supporting scripts as part of skills, and integration skills into the IDE using Port CLI. We are looking forward to your feedback on our public roadmap.

Check out our Skills documentation to get started, or see working examples in our public demo.