When you ask your agent to do work in gloria.dev, you usually start with a skill. A skill is a versioned, shareable coding-agent instruction in the Skill Library; see Concepts for the shared vocabulary behind that model.
The simplest way to think about the flow is: the skill thinks, MCP tools move the result, and the web app keeps the record visible. That turns a plain-language request into something stored, shared, and ready for the next step.
The skill is the unit you reach for
Ask your agent: “Use documenting-service-dependencies to map the services in this repository.”
Your agent reads the repository, looks for the signals that matter, and returns a structured result in human terms. Other skills follow the same pattern. using-the-skills-library helps a team work with the shared library, identifying-skills-for-a-project finds useful gaps, defining-the-documentation-site-map turns source into a site map, and writing-doc-holiday-prompts turns that plan into a prompt set.
MCP tools move the result
After the skill finishes, your agent uses MCP tools to hand the result to gloria.dev or Doc Holiday. That is the bridge between analysis and action. The skill stays focused on reading source and deciding what matters. The MCP layer takes the structured findings and records them in the right place, updates shared state, or requests the next piece of work.
If you want the exact tool list or the direct integration details, see MCP Tools.
What you get back
You do not need to think about the wiring. Your agent gives you a concise reply that says what it found, what changed, and where the work now lives. Depending on the skill, that might mean a refreshed dependency inventory, a published skill, a documentation request, or a conversation thread you can follow in Doc Holiday.
Why the split matters
This split keeps the experience clean. Skills stay close to the source and do the thinking. MCP tools move the result into the shared system. The web app then shows the durable state so the next request starts from the latest picture, not from guesswork.
Next
Go to Skill Library for the author → publish → install model, or open MCP Tools for the reference index.