Enables Claude Code to utilize OpenRouter models through delegated sub-agents to optimize context window efficiency.
This skill provides a specialized framework for integrating the Claudish CLI with Claude Code, allowing developers to leverage alternative models like Grok, GPT-5, and Gemini while strictly maintaining context window hygiene. It includes a mandatory decision tree for agent delegation, specialized file-based instruction patterns, and a curated selection matrix to ensure that external model calls happen within isolated sub-agents. By following these patterns, users can prevent token pollution in their main conversation while harnessing the specialized capabilities of various state-of-the-art language models.
Key Features
01Strict Sub-Agent Delegation Patterns
02File-Based Instruction Workflow
03OpenRouter Model Gateway (Grok, GPT-5, Gemini, MiniMax)
04Interactive Model Selector and Fuzzy Search
050 GitHub stars
06Context Window Pollution Prevention
Use Cases
01Scaling complex refactoring jobs through isolated sub-agent environments.
02Benchmarking different LLM outputs for the same implementation task.
03Executing specialized coding tasks with Grok-Code or GPT-5 without losing main context.