GitHub Copilot for JetBrains: Agentic Capabilities GA, Agent Hooks, and Instruction File Support

Core Agentic Capabilities Reach General Availability in JetBrains

GitHub has promoted three foundational agentic features in the JetBrains Copilot plugin to general availability: custom agents, sub-agents, and plan agent. These capabilities, previously available in preview, are now production-ready for all paid Copilot subscribers using IntelliJ IDEA, PyCharm, WebStorm, GoLand, and other JetBrains IDEs.

Custom agents allow teams to define specialized personas and workflows tailored to specific roles — such as a dedicated planning agent or a review agent — using agent definition files stored in .github/agents/*.agent.md. Sub-agents enable one agent to delegate subtasks to another, allowing complex multi-step workflows to be decomposed and executed in sequence. Plan agent provides a structured reasoning step before action, prompting Copilot to lay out a plan before executing code changes.

Agent Hooks Now in Public Preview

A new mechanism called agent hooks is entering public preview for JetBrains users. Hooks allow developers and teams to run deterministic shell commands at key points during an agent session, defined as JSON files placed in .github/hooks/ within the repository.

Supported hook events include userPromptSubmitted, preToolUse, postToolUse, and errorOccurred. The preToolUse hook is particularly powerful because it can approve or deny a tool call before it executes, giving teams a way to enforce policies or integrate with external approval systems without interrupting the entire session.

MCP Auto-Approve Support

The update also adds the ability to configure MCP (Model Context Protocol) auto-approval at both the server and tool level directly from JetBrains settings. Developers can navigate to Settings > GitHub Copilot > Chat > MCP Server and Tool Auto-approve Configuration to designate which MCP tools should proceed without manual confirmation, reducing interruptions during long-running agent sessions.

AGENTS.md and CLAUDE.md Instruction File Support

GitHub Copilot for JetBrains now supports AGENTS.md and CLAUDE.md instruction files, which can be placed anywhere in a repository to provide natural-language guidance to agent workflows. Copilot automatically discovers and loads these files into context during agent sessions, including support for nested files across subdirectories.

A new /memory slash command provides quick in-chat access to the instruction file settings. Developers can also generate an initial AGENTS.md directly from within the IDE, lowering the barrier to getting started with agent customization. All instruction files and custom instructions are viewable and manageable under Settings > GitHub Copilot > Customizations.

Additional Model and UX Improvements

The release brings auto model selection to general availability in JetBrains for all Copilot plans, allowing Copilot to choose the best available model based on real-time performance and availability rather than requiring manual switching. A dedicated thinking panel has been added for extended-reasoning models, with configurable budget controls, and the chat interface now includes a context window usage indicator.

On the UX side, sign-in flows have been streamlined, the chat panel opens automatically after authentication, and device code login is now prioritized for Enterprise users. The update also deprecates edit mode in the chat mode dropdown, reflecting the maturation of agent mode as the primary interactive workflow.