What I found after digging deeper: I explored the codebase and discovered some existing capabilities I initially missed: • The "Edit & Replay" feature in PromptEditor.tsx - allows editing prompts and re-running them • The /eval-creator page - a GUI for building configs • Test Case Dialog - add/edit test cases visually These are great! But they seem oriented around post-eval editing rather than pre-eval exploration. The specific gap I see: Current workflow: Write YAML config → Run eval → View results Proposed workflow: Tinker with prompt → See output → Save as test case Basically moving from "Formalize first" to "Explore first". Why I think this aligns with promptfoo: Better onboarding - Lower barrier for new users who just want to "try something". Competitive parity - LangSmith, Humanloop, and PromptLayer all have playground modes. Doesn't cannibalize Cloud - Enterprise value (team collab, history, compliance) stays intact. Concrete proposal: I could extend the existing /eval-creator page to add: • A "Run Single Prompt" button (no full eval needed) • Live variable substitution preview • "Mark as Test Case" with auto-suggested assertions • Export to promptfooconfig.yaml Before I start: A few questions to avoid wasted effort: Is this direction something you'd accept as a PR, or would you prefer it as a standalone community tool? Any architectural preferences? (e.g., should it reuse the existing runAssertion pipeline or be separate?) Who's the right person to loop in for design review?