Where are my model preferences stored?

Model preferences are learned based on your task context and saved in the .claude/multimodel-team.json file for automatic reuse in future sessions.

Can I pass Claude Code flags like --effort to external models?

Yes, claudish v5.3.0+ supports a two-pass parser that recognizes its own flags and forwards unknown flags like --effort or --permission-mode directly to Claude Code.

How does this skill handle different AI providers?

It uses the claudish CLI, which internally routes requests to providers like OpenAI, Google, and xAI using bare model names without needing provider prefixes.

What happens if an external model fails during a task?

The skill follows a strict 'Stop and Report' protocol. It will never silently substitute a model; instead, it reports the exit code and stderr, then waits for a user decision.

Does it support parallel execution for speed?

Yes, it provides specific Bash patterns for running multiple external models in the background within a single /team message to maximize efficiency.

External Model Orchestrator

Name: External Model Orchestrator
Author: MadAppGang

byMadAppGang

•

数据科学与机器学习

Orchestrates external AI models like Grok, Gemini, and GPT-5 within Claude Code using deterministic Bash commands and the claudish CLI.

This skill provides a standardized framework for integrating non-Claude LLMs into your Claude Code environment. By utilizing the claudish CLI, it enables reliable model invocation with full flag passthrough for settings like effort levels and permission modes. It features automatic context-aware model selection, parallel execution patterns for team workflows, and a robust error escalation protocol to ensure multi-model tasks are executed with total transparency and reliability. Whether you are cross-validating code or running complex research, this skill ensures your external model integrations are predictable and 100% reproducible.

主要功能

01Robust error verification and escalation protocol for failed requests

02Automatic context-aware model preference learning and reuse

03Deterministic model invocation via Bash + claudish CLI

043 GitHub stars

05Full flag passthrough for Claude Code parameters like effort and permission-mode

06Parallel execution support for multi-model /team orchestration

使用场景

01Cross-validating complex logic or architectural decisions across multiple SOTA LLMs

02Automating multi-step research tasks that require specific external model capabilities

03Running parallel code reviews using Grok and Gemini alongside Claude

主要功能

01Robust error verification and escalation protocol for failed requests

02Automatic context-aware model preference learning and reuse

03Deterministic model invocation via Bash + claudish CLI

043 GitHub stars

05Full flag passthrough for Claude Code parameters like effort and permission-mode

06Parallel execution support for multi-model /team orchestration

使用场景

01Cross-validating complex logic or architectural decisions across multiple SOTA LLMs

02Automating multi-step research tasks that require specific external model capabilities

03Running parallel code reviews using Grok and Gemini alongside Claude