Claudecode Glm Stack — AI skill for Claude Code

Run GLM-5.2 1M in Claude Code

This guide routes Claude Code's API calls through **Z.ai GLM-5.2 1M instead of Anthropic** -- same Claude Code UI, same skills, same workflow.

It uses Z.ai's Anthropic-compatible endpoint, so Claude Code can talk to GLM directly. No local proxy is required.

Most importantly: **your normal Claude Code stays untouched**. Plain `claude` keeps using your usual backend, such as Opus on Vertex AI. GLM only activates when you pass `--bare --settings`.

---

Setup

1. Get a Z.ai API key

1. Go to https://docs.z.ai
2. Sign in to Z.ai / Zhipu AI
3. Create an API key
4. Copy the key for the local setup below

2. Save the key locally

mkdir -p ~/.config/mg-glm && chmod 700 ~/.config/mg-glm
echo 'export ZAI_API_KEY="your-zai-key-here"' > ~/.config/mg-glm/key.env
chmod 600 ~/.config/mg-glm/key.env
source ~/.config/mg-glm/key.env

3. Create a Claude Code settings override

Most Claude Code installs already have a provider configured: Anthropic login, Vertex AI, AWS Bedrock, or an API key. Inline env-var prefixes can lose that precedence fight. The reliable approach is a dedicated `--settings` override file.

cat > ~/.config/mg-glm/claude-glm-settings.json <

The empty strings neutralize Vertex / Bedrock defaults for this one command. `ANTHROPIC_BASE_URL` points Claude Code at Z.ai, and the model keys map Claude Code's n