JetBrains shipped two new features for its AI Assistant plugin: Skill Manager and a curated Skill Repository. Together they create an IDE-level management layer for agent skills, the reusable capability bundles that tell AI agents how to handle specific tasks like CI debugging, code migration, or browser automation, according to the JetBrains AI blog.
The Portability Problem
Agent skills today tend to get locked to a single setup: one agent, one repository, one machine. A developer who builds a skill for debugging CI failures in one project has to rebuild or manually copy it to use in another. Skill Manager addresses this by adding three scope levels for each skill:
IDE-level skills stay available across all projects for an individual developer. Install once, use everywhere. Project-level skills travel with the repository via version control and can be shared with teammates. Agent-specific skills stay tied to a particular workflow like code review or frontend work.
The practical result: a developer installs a Postgres optimization skill at the IDE level and it is available in every project they open, regardless of which agent they are using in AI Assistant Chat.
Security-Screened Repository
The Skill Repository on GitHub provides a JetBrains-curated starting point. New additions are screened for prompt injection, data exfiltration, and malicious code patterns, according to the blog post. Attribution is preserved using the skill’s original author metadata or upstream maintainer credit.
Launch skills include React and Next.js best practices, Postgres query and schema optimization, Playwright browser automation, pnpm workspace management, and Java-to-Kotlin migration tooling.
Current Limitations
JetBrains is transparent about the current boundaries. IDE-wide skill storage works in AI Assistant Chat for Codex and Claude Agent. Support for Junie (JetBrains’ own coding agent) and other ACP-compatible agents is coming. CLI agents cannot access IDE-installed skills at all, so terminal-based workflows still depend on project-level or agent-specific installation.
There is also a fragmentation detail: Claude Agent uses its own .claude folder rather than the shared .agents folder due to harness configuration differences. That means cross-agent portability is not fully seamless yet.
The Emerging Skill Ecosystem
JetBrains is not the only company investing in agent skill infrastructure. OpenAI’s Codex supports skills via its developer documentation, and the open Agent Skills format is gaining traction as a cross-client portability standard. The pattern is clear: IDEs are evolving from code editors into control planes for agent deployment, and reusable, security-vetted skill bundles are becoming the unit of agent capability distribution.