claude-switch-models-setup

Configure isolated Claude Code CLI profiles to run multiple LLM providers in separate terminal windows.

1.4k|216|Updated Oct 22, 2025
One-click install
npx skills add https://github.com/daymade/claude-code-skills --skill claude-switch-models-setup
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: claude-switch-models-setup
Source: https://github.com/daymade/claude-code-skills/tree/main/daymade-claude-code/claude-switch-models-setup
Command: npx skills add https://github.com/daymade/claude-code-skills --skill claude-switch-models-setup

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

Running Claude Code with third-party LLM providers (Kimi, DeepSeek, GLM, MiniMax, StepFun) alongside Anthropic normally causes configuration bleed: shared settings, plugin state, and credentials collide when you want several provider windows open at once. This Skill builds an isolated-but-shared profile system so each provider runs as a fully independent Claude Code process while skills, projects, hooks, and plugin content stay shared.

Core Features & Use Cases

  • Isolated profile management: Creates per-provider directories under ~/.claude-profiles with isolated .claude.json state, symlinked shared content, and shell aliases (csk, csd, csg, css) for one-command launches.
  • Automatic settings convergence: A SessionStart hook script syncs settings.json (hooks, marketplaces, permissions, env flags) and behavior keys from the default profile into every profile, preventing silent drift.
  • Per-profile plugin repair: claude-plugins-sync.py rebuilds per-profile known_marketplaces.json and mirrors enabledPlugins, fixing "corrupted installLocation" and missing-skill errors.
  • Provider templates and context-window guidance: Ships templates for Kimi, MiniMax, GLM, DeepSeek, StepFun, and Anthropic with verified context-window configuration ([1m] marker vs explicit token limits).
  • Use Case: A student wants Kimi in one terminal and DeepSeek in another, both with the same skills and plugins as their Anthropic setup. The Skill installs the profile scripts, generates provider settings from templates, initializes profiles, and verifies isolation with claude-profiles-doctor.

Quick Start

Ask the AI to set up Claude Code profiles so you can run Kimi and DeepSeek in separate terminal windows alongside your default Anthropic setup.

Frequently Asked Questions about claude-switch-models-setup

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run multiple Claude Code providers in separate terminal windows?

Create one profile per provider under ~/.claude-profiles using claude-profiles-init, with a provider settings file in ~/.claude/settings/<name>.json. Launch each with claude-profile <name> or aliases like csk for Kimi and csd for DeepSeek; each window gets isolated state via CLAUDE_CONFIG_DIR.

How do I switch Claude Code between Kimi, DeepSeek, and Anthropic?

Use shell aliases defined by claude-profiles.sh: csk launches Kimi, csd launches DeepSeek, csg launches GLM, css launches StepFun, and plain claude uses the default Anthropic profile. Each alias points at a separate profile with its own provider settings file.

Why does Claude Code report corrupted installLocation for marketplaces?

Claude validates installLocation with path.resolve(), which does not resolve symlinks, so a shared known_marketplaces.json fails for every profile except the one that wrote it. Run claude-plugins-sync.py to rebuild a per-profile copy with each profile's own config-dir prefix.

Why is a skill missing in my Kimi or DeepSeek profile but present in the default profile?

Claude Code stores enabledPlugins in each config directory's settings.json, so sharing the plugin cache is not enough. Run claude-plugins-sync.py to mirror enabledPlugins from the default profile, then restart the affected window.

How do I configure the context window size for a third-party Claude Code provider?

Append the literal [1m] suffix to model values for verified 1M-context providers, or set explicit CLAUDE_CODE_MAX_CONTEXT_TOKENS and CLAUDE_CODE_AUTO_COMPACT_WINDOW values for smaller windows. Verify the provider's real limit in its own docs rather than copying another template.

What are the limitations of sharing Claude Code config across profiles?

Session history, credentials, and known_marketplaces.json must stay per-profile and cannot be shared. Settings changes apply only at the next session start, and running sessions never hot-reload skill metadata.