claude-code-model

Configure Claude Code to use custom models (DeepSeek, GLM, Qwen, etc.). Use when user wants to change Claude Code's model, switch API provider, or set up cus...

Security Scan

The skill does what it says: it updates Claude Code config files and environment variables to point at a custom model endpoint; nothing in the files indicates hidden exfiltration or unrelated privileges, but it will store your API key in config/env in plaintext so only install if you trust the provider and the script.

Purpose & Capability

Name/description align with the included script and instructions: the SKILL.md and scripts/configure_model.py only change ~/.claude/config.json, ~/.claude/settings.json, and user-level environment variables to point Claude Code at a different API endpoint and model. The requested inputs (base_url, model, api_key) are appropriate for this purpose.

Instruction Scope

Instructions are scoped to configuring Claude Code: they show how to set environment variables, edit the two config files, and run a local 'claude' command to verify. They do not attempt to read unrelated system files or send data to third-party endpoints beyond invoking the local 'claude' tool.

Install Mechanism

No install spec or external downloads; the skill is instruction + a small included Python script. Nothing is fetched from remote URLs or installed to system-wide locations.

Credentials

The skill asks for an API key and writes it into user environment variables and config files (plaintext). That is proportionate to changing the model provider, but is a sensitive action: storing credentials in shell profiles and JSON files is expected here but has privacy/security implications the user should consider.

Persistence & Privilege

The skill does not request always: true and does not modify other skills or system-wide agent settings. It writes to the user's ~/.claude config and (on Windows) sets per-user environment variables — expected for this functionality.

Assessment

This skill appears to be what it claims: it will add or replace entries in ~/.claude/config.json and ~/.claude/settings.json and (on Windows) set user-level environment variables, or print shell lines for you to add on Unix. Before running: 1) review the script to confirm you trust it (it runs PowerShell on Windows and calls the local 'claude' binary); 2) back up ~/.claude/config.json and settings.json so you can restore original settings; 3) be aware your API key will be stored in environment variables and JSON files in plaintext — only use keys you trust to share with the configured provider; 4) prefer running the script locally rather than pasting commands from unknown sources, and verify the base_url points to the intended provider.

Claude Code Model Configuration

Switch Claude Code to use custom model providers (DeepSeek, GLM, Qwen, OpenAI-compatible endpoints).

Configuration Locations

Claude Code reads config from multiple sources (in priority order):

1. Environment variables (highest priority)
2. ~/.claude/config.json
3. ~/.claude/settings.json

Required Parameters

User must provide:

• base_url: API endpoint (e.g., https://api.deepseek.com/anthropic)
• model: Model name (e.g., deepseek-v4-flash, glm-5)
• api_key: API key

Workflow

Step 1: Update Environment Variables

Set user-level environment variables (persist across restarts):

[Environment]::SetEnvironmentVariable("ANTHROPIC_API_KEY", "<api_key>", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_BASE_URL", "<base_url>", "User")
[Environment]::SetEnvironmentVariable("ANTHROPIC_MODEL", "<model>", "User")

Clear conflicting variables:

[Environment]::SetEnvironmentVariable("ANTHROPIC_AUTH_TOKEN", "", "User")

Step 2: Update config.json

Edit ~/.claude/config.json:

{
  "env": {
    "ANTHROPIC_API_KEY": "<api_key>",
    "ANTHROPIC_BASE_URL": "<base_url>",
    "ANTHROPIC_MODEL": "<model>"
  }
}

Step 3: Update settings.json

Edit ~/.claude/settings.json, add/update:

{
  "model": "<model>",
  "env": {
    "ANTHROPIC_BASE_URL": "<base_url>",
    "ANTHROPIC_API_KEY": "<api_key>"
  }
}

Step 4: Verify

Test the configuration:

claude --print "hi, what model are you?"

Common Issues

• Auth conflict: Both ANTHROPIC_AUTH_TOKEN and ANTHROPIC_API_KEY set → Clear ANTHROPIC_AUTH_TOKEN
• Old model still shows: Check config.json for stale ANTHROPIC_MODEL value
• Changes not生效: New terminal window required for env var changes

Popular Provider Examples

Provider	base_url	Models
DeepSeek	https://api.deepseek.com/anthropic	deepseek-v4-flash, deepseek-v4-pro
GLM (阿里云)	https://coding.dashscope.aliyuncs.com/apps/anthropic	glm-5
Qwen	Same as GLM	qwen-*

Script

Use the bundled script for automated configuration:

python scripts/configure_model.py --base-url <url> --model <name> --api-key <key>