Introduction: New Collaboration Paradigm Between Claude Code and Gemini CLI

gemini-delegate is a Claude Code skill plugin that intelligently identifies task types and automatically delegates suitable work to Google Gemini CLI. It fully leverages Claude's coding capabilities and Gemini's advantages in million-level context, real-time search, and multimodal processing to achieve a dual-AI collaborative workflow, letting AI assistants do their respective jobs.

Background: Capability Boundaries of Single AI and Collaboration Needs

In daily development, Claude Code is recognized for its excellent code understanding and generation capabilities, but it has bottlenecks: limited by context window when processing large files, unable to perform real-time online searches, and lacking native support for multimodal content. Google Gemini CLI, on the other hand, has million-level token context, built-in real-time search, and multimodal reading functions. How to make them collaborate has become a new exploration direction.

gemini-delegate Project: Trigger Scenarios for Intelligent Task Delegation

gemini-delegate is a task scheduling layer with intelligent judgment capabilities, core to which is "let the right tool do the right thing". Delegation is automatically triggered in the following scenarios:

1. Large file processing: When a code file exceeds about 500 lines, use Gemini's large context for complete analysis;
2. Full codebase analysis: Call the codebase_investigator tool to deeply map the project architecture;
3. Multimodal processing: Support reading of images, videos, PDFs and other files;
4. Real-time information acquisition: Query time-sensitive content such as the latest software versions and CVE vulnerabilities;
5. Double verification: Call Gemini for cross-verification on key tasks (code generation, review, etc.) to improve quality.

Technical Implementation: Complete Workflow From Trigger to Result Return

gemini-delegate's workflow:

1. Task trigger: Automatic feature recognition trigger or user explicit instruction;
2. Config reading: Read config.md to get user subscription tier and model settings;
3. Model selection: Use DEFAULT_MODEL (e.g., auto-gemini-3) for complex tasks, FAST_MODEL (e.g., gemini-3-flash-preview) for fast tasks, and Pro-exclusive PREVIEW_MODEL for cutting-edge tasks;
4. Command construction and execution: Build Gemini CLI command and attach the instruction "Execute immediately, do not show a plan." to avoid hanging;
5. Result return: Claude receives and integrates Gemini's output to present a complete answer.

Practical Applications: Typical Scenario Examples

gemini-delegate's 6 typical usage scenarios:

1. Large file summary: Generate a structured summary for a 3000-line legacy configuration file;
2. Codebase architecture analysis: Output module relationships and dependency graphs for newly taken-over projects;
3. UI screenshot analysis: Describe components and layout in interface screenshots;
4. Real-time security audit: Query the latest CVE vulnerabilities and repair suggestions for libraries like lodash;
5. Video bug analysis: Identify issues in screen recording videos;
6. PDF report extraction: Structured extraction of key findings and recommendations from technical whitepapers.

Installation and Configuration Guide

1. Install Gemini CLI: npm install -g @google/gemini-cli;
2. Install the skill: git clone https://github.com/charlienew/gemini-delegate.git, copy to Claude's skill directory: cp -r gemini-delegate ~/.claude/skills/gemini-delegate;
3. Config adjustment:
   • Free version: Default configuration (model gemini-2.5-flash);
   • Pro version: Modify config.md to unlock Gemini3 series: TIER=pro, DEFAULT_MODEL=auto-gemini-3, etc.

Limitations and Future Trends of AI Collaboration

Limitations: Requires stable network (for real-time search), free version quota limits, OAuth token expiration requiring re-authentication, and large codebases needing .geminiignore to exclude redundant directories. Future trends: gemini-delegate represents the multi-AI collaboration paradigm, where different AIs perform their respective roles and are seamlessly connected via standard interfaces. This will become the evolution direction of development workflows, unlocking the potential of cross-platform intelligent collaboration.