kumiko eval
kumiko eval [args]kumiko eval drift [args]Wraps the enterprise evaluation pipeline. It requires the enterprise evaluation package to be available in the configured maintainer workspace and fails loudly when that dependency is missing.
Common args (kumiko eval)
Section titled “Common args (kumiko eval)”| Flag | Meaning |
|---|---|
--smoke | 3-fixture subset |
--live | Hit a real provider (default Anthropic) |
--baseline | Write checked-in baseline JSON |
--model <name> | e.g. claude-sonnet-4-6, claude-haiku-4-5 |
--provider openai-compat | Ollama / vLLM |
--endpoint <url> | OpenAI-compat endpoint |
kumiko eval drift --baseline docs/l2-eval-live-baseline.json --current /tmp/new.jsonExit 1 when regressions are detected.
Args are forwarded 1:1 to the underlying enterprise scripts, see those scripts for the full surface.