Overview
This glossary defines the EN-only domain terms used across the workshop. The full EN↔FR translation glossary lives on the French side at fr/labs/glossary.md because its primary audience is FR readers reconciling English tooling with French documentation.
Terms
Agent. A Copilot Studio entity that talks to users through one or more channels. Authored as YAML files under a cloned folder (for example, HelloWorldAgent/agent.mcs.yml).
Topic. A unit of conversational logic inside an agent — a trigger plus the actions to run when it fires. Lives at topics/<Name>.topic.mcs.yml.
Action. A step that an agent runs as part of a topic. Built-in kinds include SendActivity, Question, and SetVariable. Custom kinds back MCP servers, Power Automate flows, and connectors. Lives at actions/<Name>.action.mcs.yml.
Knowledge source. A retrieval-augmented-generation source the agent grounds its replies in — a public URL, a SharePoint site, a Dataverse table, or a custom source. Lives at knowledge/<Name>.knowledge.mcs.yml.
Variable. A typed value the agent can read and write across topic invocations. Global variables live at variables/<Name>.variable.mcs.yml; topic-scoped variables are inline inside the topic file.
Trigger. The condition that starts a topic. The most common kind is OnRecognizedIntent, which fires when the user’s message matches one of the topic’s triggerQueries.
Skill (in this workshop). A plugin capability inside an AI coding tool such as GitHub Copilot CLI or Claude Code. The repository microsoft/skills-for-copilot-studio ships several skills (clone, push, author, add-knowledge, and so on) that the four @copilot-studio:* sub-agents orchestrate.
Skill (Copilot Studio “Skill”, capital S). A separate, older Bot Framework feature where one agent calls another agent as a sub-conversation, registered in the portal under Settings → Skills. This workshop does not use that feature. It is mentioned here only to disambiguate the word “skill” — see the disambiguation table in the hello-world walkthrough.
Sub-agent. A @<plugin>:<name> invocation inside copilot that delegates to a specific plugin capability. This workshop uses four: @copilot-studio:copilot-studio-manage, @copilot-studio:copilot-studio-author, @copilot-studio:copilot-studio-test, and @copilot-studio:copilot-studio-advisor.
Plugin. A bundle of skills installed into an AI coding tool. The plugin marketplace entry for this workshop is microsoft/skills-for-copilot-studio; the installed plugin slug is copilot-studio.
LSP binary. The LanguageServerHost executable that ships inside the Copilot Studio VS Code extension. The plugin shells out to it for every cloud-sync operation (clone, push, pull).
Power Platform environment. A Microsoft Power Platform isolation boundary that hosts Copilot Studio agents, Power Apps, Power Automate flows, and Dataverse data. Pick one in the portal’s top-right environment switcher.
Clone. Download an existing cloud agent into a local folder as YAML. @copilot-studio:copilot-studio-manage clone is the workshop’s entry point.
Pull. Re-sync your local folder against the latest cloud version of the agent. Use this when push fails with ConcurrencyVersionMismatch.
Push. Upload local YAML changes to the Copilot Studio cloud as a draft revision. The agent does not yet talk in the new shape until you publish.
Publish. Promote the current draft revision so the Test pane and deployed channels invoke it. A required step after every push that you want to test.
Test pane. The conversation panel inside the Copilot Studio portal that lets you talk to the published agent without leaving the editor. Opened via the Test button in the top-right of an agent edit view.
Activity map. A debugging view inside the Test pane (three-dot menu → Show activity map when testing) that shows which trigger, actions, and generative-answers nodes fired in response to a message. The primary way to confirm that retrieval against a knowledge source is actually wired up.
Adaptive Card. A JSON-described UI snippet the agent can return inside a SendActivity action. The Copilot Studio schema supports Adaptive Cards v1.6.
Generative answers. Copilot Studio’s name for retrieval-augmented generation: the agent grounds its response in attached knowledge sources rather than free-running model output.