What is the claudish CLI used for in Claude Code?

The claudish CLI is a tool that allows Claude to invoke external AI models directly through bash commands, ensuring a more reliable and deterministic integration than prompt-based delegation.

Does this skill support local AI models?

Yes, it includes routing prefixes for popular local backends such as Ollama, LM Studio, vLLM, and MLX, as well as custom HTTP endpoints.

How do I troubleshoot failed external model calls?

The reference guide recommends capturing exit codes and redirecting stderr to log files (e.g., 2>stderr.log) to identify issues like missing API keys or rate limits.

Can I save my preferred models for specific tasks?

Yes, preferences are automatically learned per context (e.g., debug, research, coding) and saved in .claude/multimodel-team.json for future reuse.

How do I avoid model ID collisions when using OpenRouter?

You can use specific routing prefixes like 'oai/' for OpenAI, 'g/' for Gemini, or 'mmax/' for MiniMax to ensure the system routes directly to the intended provider API.

Multi-Model Proxy Reference

Name: Multi-Model Proxy Reference
Author: MadAppGang

byMadAppGang

•

230

•

Herramientas para Desarrolladores

Provides a comprehensive technical reference for routing and invoking external AI models through the claudish CLI.

This skill serves as the essential documentation for developers using the multimodel plugin to integrate external LLMs like Gemini, GPT-4, Kimi, and GLM into their Claude Code workflows. It details the deterministic routing mechanism of the claudish CLI, explains how the system learns and reuses model preferences based on task context, and provides an exhaustive list of routing prefixes to prevent provider collisions. By following these implementation patterns, users can reliably execute parallel model reviews, optimize API costs via direct provider access, and troubleshoot integration issues using standardized bash exit codes and error logs.

Características Principales

01Standardized patterns for parallel model execution and background processing.

02Context-aware model preference learning and automatic reuse via local storage.

03Compatibility with local AI runners including Ollama, LM Studio, vLLM, and MLX.

04Multi-backend routing support for OpenAI, Google, MiniMax, Kimi, and GLM.

05Deterministic CLI-based model invocation for 100% reliable external calls.

06230 GitHub stars

Casos de Uso

01Debugging connectivity and rate-limiting issues for external LLM integrations in development workflows.

02Optimizing API expenses by routing requests directly to provider endpoints instead of aggregators.

03Conducting automated multi-model code reviews and voting systems via the /team command.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/claude-code proxy-mode-reference

For use in Claude.ai and ChatGPT

Download Skill