Evaluate with promptfoo

yori is the source of truth for your prompts; for model-graded evaluation it hands off to promptfoo rather than reinventing it.

yori export promptfoo <name> resolves an artifact's composition (includes, slots) but leaves variables as {{ placeholders }}, and emits a ready-to-run config:

yori export promptfoo review > promptfooconfig.yaml
promptfoo eval

Test cases live next to the artifact

Cases go in <name>.cases.yaml (or cases.yaml in a skill bundle) — a plain list of promptfoo test objects, so authoring yori cases is authoring promptfoo tests:

# review.cases.yaml
- description: flags a real bug
  vars: { lang: go, input: "func f() int { }" }
  assert:
    - { type: contains, value: "return" }
    - { type: llm-rubric, value: "explains the root cause" }

The provider comes from --provider or the artifact's model: hint.

yori export promptfoo review --provider openai:gpt-4o-mini
yori export promptfoo pr-bot --type agent

yori manages and ships your prompts; promptfoo grades them.