Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Agent skills are portable packages of instructions, scripts, and resources that give agents specialized capabilities and domain expertise. Unlike tools, which connect to external services, skills are self-contained sets of instructions and logic that you can create, share, and reuse across agents. Skills follow an open specification and implement a progressive disclosure pattern so agents load only the context they need, when they need it.
Use skills when you want to:
- Package domain expertise - Capture specialized knowledge (expense policies, legal workflows, data analysis pipelines) as reusable, portable packages.
- Extend agent capabilities - Give agents new abilities without changing their core instructions.
- Ensure consistency - Turn multistep tasks into repeatable, auditable workflows. Unlike tools, which connect to external services, skills are self-contained sets of instructions and logic that you can create, share, and reuse across agents.
What are skills?
A skill is a capability defined by a name, a description, and instructions that tell the agent how to handle specific types of tasks or scenarios. The orchestration runtime invokes a skill when a user's request matches the skill's purpose. Think of skills as specialized "roles" or "modes" that your agent can activate when needed.
Note
Agents powered by the GitHub Copilot harness use the skill definition described in this article: a name, description, and Markdown instructions that you create from blank, generate with AI, or upload as a file or package. For details, see Create a skill for an agent (GitHub Copilot).
Note
Agents powered by the Customer Conversation harness support two types of skills: Generative skills, which work the same way as described in this article (instructions-based, written in Markdown, and the agent reasons over them in the moment), and Topic skills, which are deterministic, authored flows of steps and conditions (matching the existing Copilot Studio Topics concept). For details on both types, see Create a skill (Customer Conversation).
Why use skills?
- Reusability: Create a skill once and add it to multiple agents.
- Modularity: Break complex agent behavior into focused, manageable pieces.
- Shareability: Export skills as Markdown files or as packages and share them with others.
- Clarity: Each skill has a clear purpose, making it easier to understand and maintain agent behavior.
How skills differ from other agent components
| Component | Purpose | Managed by |
|---|---|---|
| Instructions | General agent behavior and personality | Identity configuration |
| Knowledge | Data the agent can reference | Knowledge sources |
| Tools | Actions via external services | Connectors, APIs, MCP servers |
| Skills | Reusable task-specific capabilities | Skill files or packages (Markdown-based), or, for Customer Conversation harness Topic skills, an authored flow |
How the agent uses skills
The orchestration runtime decides when to activate a skill based on the user's message and the skill's description. When the runtime activates a skill, the skill's instructions guide your agent's behavior for that specific task. Skills can work alongside tools and knowledge. For example, a skill might instruct your agent to use a specific tool in a particular way. Well-written skill descriptions help the orchestrator invoke the right skill at the right time.
Built-in skills
The GitHub Copilot harness provides built-in skills for analyzing document and data files. These skills contain task-specific instructions and supporting resources. The agent selects relevant skill guidance as it works on a request.
Examples of built-in document-analysis skills include:
| File format | What the skill helps the agent do |
|---|---|
CSV (.csv) |
Look up values, filter rows, and calculate summaries from tabular data. |
Word (.docx) |
Extract text, document structure, tables, and images to answer questions about a document. |
HTML (.html, .htm) |
Read page content and analyze tables and other structured information. |
Markdown (.md) |
Find information in headings, lists, and tables, and calculate summaries from table data. |
PDF (.pdf) |
Extract text and tables to answer questions about document content. |
PowerPoint (.pptx) |
Analyze slide content, including tables, charts, and speaker notes. |
Excel (.xlsx) |
Look up and summarize data across worksheets and workbooks, and create charts. |
These built-in skills are provided by the harness, rather than uploaded or authored by a maker. You can add your own skills to provide other task guidance or domain expertise.
Skills provide instructions; packages provide code libraries the agent can use for file and data operations. Learn more in Available packages for file and data tasks.
Adding a new skill
You add a skill to an agent from the Build tab in Copilot Studio. The available creation methods depend on which harness powers your agent:
- Agents powered by the GitHub Copilot harness support three methods: uploading an existing skill file or package, creating a skill from blank, or generating one with AI. Learn more in Create a skill for an agent (GitHub Copilot) and Add and manage an uploaded skill (GitHub Copilot).
- Agents powered by the Customer Conversation harness support two skill types, Generative and Topic, both created directly from the Add skill dialog. Learn more in Create a skill (Customer Conversation).