Gemini CLI: Built-in Agent Skills, PR Creator, and Smarter Help Agent

Built-in Skills Go Live by Default

Gemini CLI v0.25.0, released January 20, 2026, marks the point at which Agent Skills moved from an opt-in experimental feature to a first-class capability enabled for all users. Skills are self-contained instruction bundles that the CLI loads on demand, extending the agent with specialized workflows without bloating the base context window. With this release, skills are active by default — no configuration changes required.

At the same time, Google enhanced the activate_skill tool that powers skill invocation, making it more reliable and consistent across different terminal environments.

PR Creator: Automated, Standards-Aware Pull Requests

The most developer-focused addition in v0.25.0 is the built-in pr-creator skill. When a developer asks Gemini CLI to create a pull request, the skill takes over: it first checks whether the current branch is main and creates a new descriptive branch if needed, then scans the repository for PR templates at standard paths like .github/pull_request_template.md. It drafts a description that follows the template structure, marks completed checklist items, and ultimately invokes the gh CLI to open the PR — writing the body to a temporary file to avoid shell-escaping issues.

This removes a common friction point: developers who previously had to context-switch between the terminal and a browser to write PR descriptions can now issue a single natural-language instruction and have Gemini CLI handle the entire workflow.

A Dedicated Help Agent

v0.25.0 introduced a specialized cli_help agent that intercepts Gemini CLI-specific queries — questions about commands, configuration keys, or extension management — and routes them to an agent trained on the CLI's own documentation. Previously, these queries were handled by the general-purpose model, which sometimes produced imprecise answers. The dedicated help agent delivers more accurate, configuration-aware responses.

/agents refresh and Runtime Updates

A new /agents refresh command allows developers to reload available agents and skills at runtime without restarting the session. This is particularly useful in extension development workflows, where a developer modifying a skill definition no longer needs to close and reopen the CLI to pick up changes.

UX Improvements: Tab Focus and Terminal Titles

v0.25.0 brings two quality-of-life improvements to the terminal experience. The interactive shell focus toggle was moved from Ctrl+F to Tab, making it faster to switch between the input area and a running shell process. The terminal tab title now reflects the current agent state dynamically: "◇ Ready", "✦ Working…", or "✋ Action Required" — giving developers a glanceable status indicator without needing to watch the CLI output.

Other UX additions include @-based file selection in slash commands, in-CLI extension management via /extensions install and /extensions uninstall, and the ability to disable automatic LLM-based edit correction by setting tools.disableLLMCorrection to true for faster, deterministic edit workflows.

DeepLearning.AI Course Launch

Alongside v0.25.0, Google partnered with DeepLearning.AI to publish a free short course titled "Code & Create with an Open-Source Agent," taught by Jack Wotherspoon, Developer Advocate at Google. The course covers installation, context management, MCP server integration, extension orchestration, and creative automation workflows — offering a structured path from beginner to proficient Gemini CLI user in under two hours.