Multi-Model Skill

Run a multi-model “battle” loop where:

• Each iteration uses a different model (rotating by index).
• The active model produces a candidate answer.
• Other models judge and critique it.
• The process keeps the best-scoring answer as the current best.

Configuration (.env)

This skill reads .env (searching upward from the current working directory) to find multi-model config.

Single-endpoint setup (OpenAI-compatible)

• ARENA_OPENAI_BASE_URL (e.g. http://localhost:11434/v1 for Ollama, or https://api.openai.com/v1)
• ARENA_OPENAI_API_KEY (optional for Ollama)
• ARENA_MODELS (comma-separated model names)

Example:

ARENA_OPENAI_BASE_URL=http://localhost:11434/v1
ARENA_OPENAI_API_KEY=
ARENA_MODELS=qwen3:8b,deepseek-r1:14b

Multi-provider setup (optional)

• ARENA_MODELS=provider:model,provider:model2,...
• ARENA_PROVIDER_<PROVIDER>_BASE_URL=...
• ARENA_PROVIDER_<PROVIDER>_API_KEY=...

Example:

ARENA_MODELS=ollama:qwen3:8b,openai:gpt-4o-mini
ARENA_PROVIDER_OLLAMA_BASE_URL=http://localhost:11434/v1
ARENA_PROVIDER_OLLAMA_API_KEY=
ARENA_PROVIDER_OPENAI_BASE_URL=https://api.openai.com/v1
ARENA_PROVIDER_OPENAI_API_KEY=YOUR_KEY

How to run

1. Ensure .env exists and contains the variables above.
2. Run the multi-model script.

python3 "${CLAUDE_PLUGIN_ROOT}/skills/multi-model/scripts/multi_model.py" --prompt "<your task>" --iters 5

Options you can use:

• --iters N: number of iterations (each iteration rotates the writer model)
• --max-judges N: cap number of judge models per round
• --json: output machine-readable JSON
• --out path.json: save JSON transcript

Output expectations

• Treat model identity as an internal numeric ID (Model 0, Model 1, ...).
• In any prompts to models, do not disclose provider/model names; only IDs.
• Never print secrets from .env (API keys).