skill-creator: run_loop.py fails with Claude Max (OAuth) - missing API key support

[noejon]

Summary

The run_loop.py description optimisation script in the skill-creator plugin fails when the user authenticates via Claude Max (OAuth) instead of an API key. The evaluation step (run_eval.py) works fine because it uses claude -p, but improve_description.py calls the Anthropic Python SDK directly via anthropic.Anthropic(), which requires ANTHROPIC_API_KEY.

Steps to reproduce

1. Authenticate with Claude Max via /login (OAuth flow)
2. Run the description optimisation loop:

   python3 -m scripts.run_loop \
     --eval-set eval_set.json \
     --skill-path path/to/skill \
     --model claude-opus-4-6 \
     --max-iterations 5

3. The eval runs complete successfully (using claude -p)
4. The improve step crashes:

   TypeError: "Could not resolve authentication method. Expected either api_key or auth_token 
   to be set. Or for one of the `X-Api-Key` or `Authorization` headers to be explicitly omitted"
   

Root cause

improve_description.py (line 115) creates an anthropic.Anthropic() client directly, which looks for ANTHROPIC_API_KEY in the environment. Claude Max users authenticate via OAuth tokens managed by the CLI, not API keys.

Suggested fix

Route the improve_description call through claude -p (like run_eval.py already does), or read the OAuth token from the CLI's credential store. This would make the full loop work for both API key and Claude Max users.

Environment

• Claude Code with Claude Max subscription (OAuth authentication)
• macOS, Python 3.9
• anthropic SDK installed via pip