Text of docs.github.com › en › copilot › responsible-use › agents as Styx crawled it on Nov 20, 2026. The page may have changed since. Only its words are kept, without pictures or layout, and only the first part of a long page.

Open the live page

Application card: GitHub Copilot Agents - GitHub Docs

Skip to main content Skip to content Collapse sidebar Expand sidebar Application card: GitHub Copilot Agents Learn how to use GitHub Copilot agentic features responsibly by understanding their purposes, capabilities, and limitations. In this article What is an Application Card? GitHub’s application and platform cards are intended to help you understand how our AI technology works, the choices application owners can make that influence application performance and behavior, and the importance of considering the whole application, including the technology, the people, and the environment.

Application cards are created for AI applications and platform cards are created for AI platform services. These resources can support the development or deployment of your own applications and can be shared with users or stakeholders impacted by them. As part of its commitment to responsible AI, GitHub adheres to Microsoft's six core principles : fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.

These principles are embedded in the Responsible AI Standard , which guides teams in designing, building, and testing AI applications. Application and Platform Cards play a key role in operationalizing these principles by offering transparency around capabilities, intended uses, and limitations. For further insight, readers are encouraged to explore Microsoft’s Responsible AI Transparency Report and GitHub Terms . 1.

Overview GitHub Copilot includes several agentic features that go beyond suggestion and conversation—they can review code, take action on your behalf, and build applications. This card covers the following experiences: Copilot code review : Reviews pull request diffs and metadata on GitHub.com, producing feedback comments and suggested changes. Copilot cloud agent : An asynchronous agent on GitHub.com that can create branches, write code, and open pull requests in response to assigned issues.

The cloud agent runs in an ephemeral, firewalled environment with automated security scanning. Copilot Cloud : Copilot in the Cloud (GitHub Copilot on GitHub.com, GitHub Mobile, or anywhere that "Cloud" is selected as a session environment) can answer questions about your repositories, plan and make changes, review or create pull requests, and automate repository tasks. A single interface built on the Copilot SDK combines chat and agentic work.

Copilot CLI : A command-line tool that can create and modify files, execute commands, and perform multi-step tasks. By default, filesystem access is scoped to the directory where the CLI was started. Permission prompts depend on the active permission mode. When computer use is enabled, Copilot CLI can also interact with desktop applications outside that directory scope. Copilot SDK : A programmatic library that allows developers to build custom AI-powered applications using Copilot.

The SDK communicates with Copilot CLI over JSON-RPC and supports custom agents, MCP server integrations, lifecycle hooks, and session management. The GitHub Copilot app : A desktop application for directing agent sessions across local repositories, git worktrees, and cloud sandboxes. The app uses GitHub Copilot CLI and GitHub Copilot SDK as its foundation and adds a user interface for parallel sessions, quick chats, GitHub issue and pull request workflows, automations, and canvases.

When computer use is enabled, the app can also interact with desktop applications. These features share common principles—human oversight, review of outputs, and responsible use—but differ in their execution environments, permissions, and data flows. The sections below describe each in context. 2. Key terms The following list provides a glossary of key terms related to GitHub Copilot Agents: Code suggestion : A specific code change proposed by Copilot code review as part of its feedback on a pull request.

Code suggestions are presented as suggested changes that can be applied with a couple of clicks. Content filtering : A safety system that scans prompts and responses to detect and block harmful, offensive, or insecure content before it is shown to the user. Computer use : An optional capability in GitHub Copilot CLI and the GitHub Copilot app that allows an agent to interact with desktop applications. Its tools can read accessible application content and visual context through the operating system's accessibility tree or screenshots when visual context is needed, click controls, enter and edit text, press keys, scroll, drag, and navigate workflows across applications.

Custom instructions : Natural language descriptions of coding style and best practices that a repository maintainer can configure to guide Copilot code review's feedback. Custom instructions help Copilot understand the conventions and standards of a specific codebase. Hallucination : A phenomenon where a language model generates output that sounds plausible but is factually incorrect, unsupported by the provided context, or entirely fabricated.

In code review, hallucination can manifest as feedback that highlights problems that do not exist or are based on misunderstandings of the code. Large language model (LLM) : A type of neural network trained on a large body of text data that can generate, analyze, and transform natural language and code. Copilot Agents use one or more LLMs to process context and produce responses. Ephemeral development environment : A temporary, isolated compute environment created for each cloud agent session.

The environment is destroyed after the session ends, ensuring that no state persists between runs. Firewall : A network-level control enabled by default for the cloud agent that prevents outbound connections to unauthorized hosts, protecting against accidental or malicious exfiltration of code or data. AI credit : A unit of consumption for Copilot features. Each use of Copilot code review consumes AI credits. Permission prompt : An interactive confirmation step in Copilot CLI that asks the user to approve an action—such as modifying a file, executing a command, or accessing files outside the current directory—before the agent proceeds.

Permission prompts are a key safety mechanism for local agentic execution. Copilot SDK : A programmatic library ( @github/copilot-sdk ) that lets developers build custom AI-powered applications by creating sessions, sending prompts, and receiving streaming responses from Copilot. The SDK communicates with Copilot CLI over JSON-RPC. JSON-RPC : The communication protocol used between the Copilot SDK and Copilot CLI. The SDK sends structured requests to the CLI process, which handles model interaction and tool execution.

Custom agent (SDK) : A named agent configuration within the SDK that has its own system prompt, scoped tools, and optional MCP servers. The SDK runtime can automatically delegate to sub-agents based on user intent. Hooks (SDK) : Lifecycle callbacks in the Copilot SDK that let developers inject custom logic at specific points during a session, such as before or after tool use, on session start or end, and on error. 3. Key features or capabilities The key features and capabilities outlined here describe what GitHub Copilot Agents are designed to do and how they perform across supported tasks.

Automated code review feedback : When a user requests a review from Copilot, it scans the code changes plus additional relevant context and provides feedback on the code. Feedback may include natural language comments and specific code suggestions linked to particular lines and files. Customizable review guidance : Copilot code review can be customized with custom instructions—natural language descriptions of coding style and best practices—so that feedback reflects a repository's conventions and standards.

Pull request creation : Copilot can pick up a task from an issue or a pull request comment, create a branch, generate tailored code changes, and open a pull request. After the

The rest of this page is on the live site.