What problem does it solve? Measuring whether Omni AI answers questions correctly is hard to do manually, and model-context changes can silently regress accuracy. This Skill drives Omni's built-in eval system to define prompt sets, run judged evals against a model or branch, and read per-prompt pass/fail verdicts from the accuracy judge. ## Core Features & Use Cases - Prompt Set Management: Create, update, list, and archive reusable prompt sets of up to 25 natural-language prompts with optional judge expectations. - Judged Eval Runs: Execute prompt sets against main or a branch, poll for completion, and read per-prompt verdicts, costs, timing, and judge rationales. - A/B Branch Comparison: Run the same prompt set on main and a branch, then diff verdicts to catch regressions before promoting model-context changes. - Use Case: After adding ai_context to a model on a branch, run your regression prompt set against both main and the branch to confirm accuracy improved from 75% to 91.7% with no newly broken prompts. ## Quick Start Ask the AI to run an eval of Omni AI on your key prompts and report the pass rate with per-prompt judge verdicts.
