Agent Skills in GitHub Copilot for JetBrains IDEs

Agent Skills in GitHub Copilot for JetBrains IDEs

Copilot Agent Skills are customizable, context-aware instruction sets that allow developers to tailor GitHub Copilot’s behavior within JetBrains IDEs. By enabling Agent mode, teams can load specific project guidelines, standard libraries, or automation scripts into the AI's context, significantly reducing manual prompting and improving workflow automation stability.

From Manual Runbooks to Contextual Agents

Back in 2009, I was working at a boutique consulting firm handling data migration for a Fortune 100 client. We had a "runbook"—a massive text file containing specific regex patterns and Hadoop configuration flags we had to use for every single ETL job. If you missed a flag, the job failed six hours later. I spent half my day just copying and pasting these snippets into my terminal or editor.

If I could have automated that context loading, I would have saved myself years of gray hair. That is essentially what the new Agent Skills update for GitHub Copilot brings to the table, specifically for those of us living inside the JetBrains ecosystem (IntelliJ, PyCharm, WebStorm).

I have been testing the new preview features released in the latest update. While I remain skeptical of tools that promise to "code for you," the shift toward Agent Skills is different. It moves us away from generic text prediction toward specific, context-aware workflow automation. Instead of treating the AI like a generic search engine, you can now teach it your specific project constraints once and have it remember them. Here is how it works and why it actually matters for your engineering team.

Understanding Agent Skills: Tailoring the AI

The core of this update is Agent Skills support (currently in public preview). Until now, Copilot in JetBrains was somewhat stateless regarding your broader team rules unless you manually pasted them into the chat window. Agent Skills allow you to import specific capabilities—think of them as specialized instruction sets or tools—into the chat context.

You can create custom skills for your specific repo or pull from community lists like github/awesome-copilot. For example, if I am working on the backend for SocketStore, I do not want Copilot guessing how my API authentication wrapper works. I want to load a "SocketStore Auth Skill" that forces the AI to use our internal libraries correctly every time.

How to Enable Agent Skills

Getting this running requires a bit of menu diving. It is not on by default.

1. Update the Plugin: Ensure you are on the latest version of the GitHub Copilot plugin for your JetBrains IDE.
2. Access Settings: Go to Settings/Preferences > GitHub Copilot > Chat > Agent.
3. Toggle Agent Mode: You should see options to enable Agent mode.
4. Select Skills: You can choose which specific skills to load into the context window.

Common Gotcha: If you use Copilot Business or Enterprise (like many of my consulting clients), you might not see these options even if you update the plugin. Your organization administrator must enable the "Editor preview features" policy in the GitHub organization settings first. I wasted twenty minutes debugging my local setup before realizing my admin account had the policy locked down.

Enhanced Inline Chat and Editor Automation

Beyond the Agent Skills, the day-to-day interactions in the editor have shifted. The "Inline Chat" feature—where you press a shortcut to talk to Copilot directly in the code editor rather than a side panel—has become the standard way I work. The update makes this smoother by allowing you to grab selected code and instantly add it as contextual input.

When I was building the initial data scrapers for SocketStore, I often had to refactor massive JSON parsing functions. The old workflow involved copying the function, pasting it into a chat window, writing a prompt, and then pasting the result back. The new improvements allow you to trigger the agent directly on the selection.

Comparison: Standard Chat vs. Agent Mode

Refined Settings and Observability

One of my biggest gripes with previous versions of the JetBrains plugin was the lack of granularity. It was often an "all or nothing" situation. The changelog indicates a significant refactor of the settings page.

Individual Toggles

You now have independent toggles for Agent mode, Coding Agent, and Custom Agent. This is crucial for observability and debugging. Sometimes the Agent gets too "creative" or aggressive. Being able to disable the Agent mode while keeping the standard completion features active allows you to dial in the right level of assistance.

Persistent Mode Settings

Previously, I noticed that certain configuration states would reset after restarting PyCharm. This update brings persistent mode settings. It sounds minor, but when you restart your IDE multiple times a day (which happens when you are managing heavy Java processes or dealing with memory leaks), having your environment stay consistent is a quality-of-life necessity.

Quality, Reliability, and Safety

GitHub also noted improvements in "safety checks in file handling tools." In the context of AI developer tools, safety usually refers to preventing the AI from hallucinating package imports that do not exist or suggesting code that violates security policies.

In my experience analyzing data pipelines, reliability is worth more than speed. I have seen junior engineers accept AI suggestions that introduced subtle race conditions in Python multi-threading scripts. The new update claims increased stability for "next edit suggestions." While I have not run a statistical analysis on the acceptance rate of these new suggestions yet, the tool feels less "jumpy" during file navigation. The cursor movement logic using Home and End keys during prompting is also much more intuitive now—it behaves like a standard text editor rather than a constrained input box.

Integrating Copilot into Data Workflows

If you are running a data-heavy operation, these features offer a specific advantage. At SocketStore, we maintain 99.9% uptime on our social media data streams. We cannot afford bad commits.

I recommend setting up a "Data Schema Skill." Create a skill definition that includes your core database schemas (Postgres/Snowflake) and your mandatory API response formats. When a developer asks the Agent to "write a function to fetch user tweets," the Agent won't invent a schema—it will use the strict definitions provided in the skill.

This moves the tool from being a fancy autocomplete to a compliance assistant. It is not perfect, and you still need code review, but it raises the baseline quality of the initial draft.

Big Data Consulting & SocketStore

Implementing these tools requires a solid underlying architecture. If your data infrastructure is a mess, AI will just help you write bad code faster.

At SocketStore, we provide a unified API for social media data (Instagram, TikTok, YouTube, etc.) that handles the complexity of data ingestion for you. If you are building analytics products, our API ensures you get clean, normalized data without maintaining your own scrapers.

For teams struggling with scaling their data infrastructure or implementing effective DevOps workflows, I offer specialized consulting. I help organizations move from fragile, manual data pipes to robust, automated systems that can handle terabytes of throughput.

Frequently Asked Questions