Gemini — OpenClaw Skill

What This Skill Does

The Gemini skill enables your OpenClaw AI agent to call Google's Gemini models as a secondary AI for specific tasks. It uses the Gemini CLI in one-shot mode, meaning your agent can send a prompt to Gemini and receive a response without entering an interactive session. This is ideal for cross-referencing information, getting a second opinion on generated content, or leveraging Gemini's particular strengths.

Your agent can specify which Gemini model to use, request structured JSON output for downstream processing, and manage Gemini extensions for enhanced capabilities. The skill runs non-interactively, so it integrates seamlessly into multi-step agent workflows where Gemini is just one component of a larger task.

Common use cases include delegating factual lookups, generating alternative drafts, extracting structured data from unstructured text, and comparing outputs between different AI models. The skill acts as a bridge between your OpenClaw agent and Google's AI ecosystem.

Example Prompts

"Ask Gemini to explain the difference between gRPC and REST APIs"

"Use Gemini to generate a JSON schema for a user registration form"

"Have Gemini summarize this research paper in three bullet points"

"Cross-check your answer about quantum computing with Gemini"

"Use Gemini to translate this paragraph to Japanese"

"Ask Gemini to list all Gemini extensions available"

Requirements

• Gemini CLI installed (gemini-cli)
• Google account authentication (one-time browser login)

On KiwiClaw: Pre-installed. Authenticate your Google account during setup and the skill is ready.

Self-hosted: Install via Homebrew: brew install gemini-cli. Run gemini once interactively to complete the login flow.

Setup on KiwiClaw

1. Open your KiwiClaw dashboard
2. Navigate to Skills in your agent settings
3. Enable the Gemini skill
4. Complete the one-time Google authentication
5. Ask your agent to query Gemini -- it responds instantly

Setup Self-Hosted

1. Install the CLI: brew install gemini-cli
2. Authenticate: Run gemini and follow the browser-based login flow
3. Test: gemini "What is the capital of France?"
4. Optionally specify a model: gemini --model gemini-pro "Your prompt"
5. Ensure the skill is enabled in your OpenClaw configuration

Related Skills

• OpenAI Image Generation -- generate images using OpenAI's DALL-E
• Summarize -- summarize URLs and files using multiple AI providers
• Coding Agent -- AI-powered code generation and debugging
• Canvas -- display generated content visually on connected devices

FAQ

Why would I use the Gemini skill if my OpenClaw agent already has an LLM?

The Gemini skill gives your agent access to a second AI model for specific tasks. This is useful for cross-referencing answers, leveraging Gemini's strengths in areas like code generation or factual retrieval, or getting structured JSON output. Think of it as your agent delegating a subtask to a specialist.

Does the Gemini skill require a Google API key?

The Gemini CLI requires authentication via a Google account. On first run, it will prompt you to complete a browser-based login flow. On KiwiClaw, this is handled during skill setup. Self-hosted users need to run gemini once interactively to authenticate.

Can I choose which Gemini model to use?

Yes. The skill supports the --model flag to specify any available Gemini model. You can ask your agent to "use Gemini Pro for this query" or specify the model name directly. The default model is used when no model is specified.

Can the Gemini skill return structured JSON output?

Yes. The skill supports the --output-format json flag, which instructs Gemini to return structured JSON. This is useful for data extraction, API response formatting, and integration with other tools and scripts.