Gemini CLI (New Upgrades - 5.0): Better Free Limits, Skills, Conductor & MORE!

Gemini CLI Updates: November - January & Skills Feature

Key Concepts:

• Gemini CLI: A command-line interface for interacting with Google’s Gemini models.
• Gemini 3: The latest generation of Gemini models, offering improved performance.
• Skills: Self-contained packages of expertise (instructions, scripts, references) for the Gemini CLI.
• Conductor: An extension for detailed AI planning and context management.
• Policy Engine: A system for defining and enforcing permissions for tool calls.
• MCP Tools: A category of tools integrated into the Gemini CLI, likely referring to Model Call Protocol tools.
• Yak Shaving: A humorous term for a series of seemingly small tasks that are required to complete a larger task, often taking a disproportionate amount of time.

UI & Core Improvements (V0.15.0 - V0.16.0)

The initial updates, starting with V0.15.0 in early November, focused on enhancing the user experience. A significant UI facelift introduced a seamless, scrollable interface with sticky headers and a stable input prompt, resolving previous jarring behavior during long sessions. Mouse support was added, allowing users to click to type. Rendering quirks were fixed for full-screen tools.

Alongside the UI improvements, “to planning” was introduced. This feature breaks down complex questions into manageable checklists, providing visible progress tracking for multi-step tasks. Users gained more control with the ability to disable GitHub extensions and validate local extension formatting using the /extensions restart and validate commands, respectively. Angular workflow support was also improved. New extensions were released: Arise (AI observability), Chronosphere (logs & metrics), and Transmit (O workflows), all installable with single commands.

V0.16.0 marked the launch of Gemini 3, initially available to paid tiers and enabled via /settings under preview features. Gemini 3 offers improved performance in complex reasoning and coding tasks. The Data Common extension was also introduced, enabling natural language queries of open-source statistical data from datacoms.org, requiring a Data Common API key and UV installation.

Security & Ecosystem Expansion (V0.18.0 - V0.19.0)

V0.18.0 introduced an experimental policy engine, allowing users and administrators to create fine-grained policies for tool calls, functioning like a firewall for AI actions. This feature is currently behind a flag. The ecosystem continued to expand with the addition of Google Workspace integration (Docs, Slides, Chat, Sheets), Reedus (data management & search), and Anomal (data quality monitoring). Gemini 3 support was extended to API key users, Google AI Pro/Ultra, and Code Assist Enterprise. The UI updates were temporarily rolled back to refine scrolling and mouse support, re-enableable via /settings with “use alternate screen buffer.” Chat history now displays the model used, and multi-uninstall functionality was added for extensions.

V0.19.0 brought the 11 Labs extension for audio creation, playback, and management. Zed integration now supports Gemini 3 when preview features are enabled, synchronizing the editor and terminal. The interactive shell was improved with click-to-focus functionality (when using alternate buffer) and a loading phrase to indicate processing.

Enhanced Functionality & Gemini 3 Access (V0.20.0 - V0.22.0)

V0.20.0 introduced multifile drag and drop functionality, automatically prefixing valid paths with at for batch operations. Persistent “always allow” policies were added, eliminating repetitive approval prompts for safe operations.

V0.21.0 released Gemini 3 Flash, described as faster, cheaper, and in some cases, outperforming 3 Pro. It was made available to paid tiers and free tier users on the waitlist via preview features. Quota visibility was improved with the /stats command displaying quotas for all models. Fuzzy setting search was implemented for quicker option finding. MCP resource support was added for resource discovery and viewing with at commands, and auto-execution on enter was enabled. New extensions included Real (natural language data analysis) and BrowserBase (web interaction, screenshots, automation).

V0.22.0 granted free tier users access to Gemini 3 Pro and Flash by toggling preview features in /settings. Gemini CLI was pre-installed in Google Colab, enabling headless use in notebooks and interactive use in the terminal. The Conductor extension was released, emphasizing a planning-first approach with detailed plans, context gathering, and guardrails stored in markdown files (product.md, plan.md) for team collaboration. Endor Labs was added for code analysis, vulnerability scanning, and dependency checks.

Agent Skills & Recent Updates (V0.23.0)

The January release, V0.23.0, introduced “agent skills” in preview. Skills are self-contained directories packaging instructions, scripts, and references, offering discoverable expertise activated when relevant. Skills are discovered in three tiers: Workspace, User, and Extension, with higher tiers overriding lower ones. Management commands include /skills list, /skills disable, /skills enable, and /skills reload. Installation and uninstallation are done via gemini skills install and gemini skills uninstall, respectively. Creating a skill requires a skill.md file with YAML front matter (name, description) and a body containing procedural guidance. Security requires explicit consent before skill activation, displaying purpose and permissions.

Additional V0.23.0 features include Gemini CLI Wrapped (usage statistics visualization via npx gemini-wrapped), clipboard image support for Windows users (Alt+V), terminal background color detection for theme optimization and accessibility, and a /log command for clearing credentials.

Notable Quotes:

• “The UI changes from v0.15.0 are genuinely great. Sticky headers, no flicker, and mouse support make long sessions feel modern.”
• “Conductor is a standout. I've been using it heavily and the planning first approach dramatically reduces wasted iterations on complex projects.”
• “Skills are an elegant abstraction for packaging expertise. They're portable, discoverable, and secure.”

Limitations & Considerations:

The presenter acknowledges limitations including the experimental nature of skills (potential rough edges, documentation gaps), the policy engine being behind a flag (not production-ready), the initial configuration complexity of extensions, the learning curve associated with Conductor’s workflow, and the need to manage API quotas.

Synthesis/Conclusion:

The recent Gemini CLI updates represent a significant evolution from a simple chat tool to a comprehensive developer platform. The improvements in UI, security, planning (Conductor), and the introduction of skills demonstrate a clear vision for AI-assisted development workflows. The expanded access to Gemini 3, particularly for free tier users, further democratizes access to powerful AI models. While some features remain experimental or require initial setup effort, the overall trajectory indicates a rapidly maturing and increasingly valuable tool for developers.