eval-engineerlisted

# Eval Engineer Use this skill as the front door for Galileo-backed eval engineering. Keep it as a router: inspect current project readiness, educate the user only enough to choose the workflow, and route to the focused command skill that matches the job. If a focused skill clearly applies, use that focused skill directly instead of running the full core loop here. ## Current Project State Start by checking what the user gave you and what the project already has: - Galileo URL, project/log-stream/experiment/session/trace ID, or symptom - `.galileo/config.yml` - `.galileo/current/debug-packet.json` - `.galileo/current/verification-debug-packet.json` - configured verification commands - metric profile or expected-output contract Then report one compact status block: ```text Eval Engineer works by closing the loop: evidence -> diagnosis -> bounded change -> verification. Current project state: - workspace: ready/missing - evidence: ready/missing/ambiguous - measurement: ready/missing - best next command: /eval-... ``` Do not dump general docs. Name the smallest useful next step. ## Route - Use `/eval-setup` when `.galileo/` or verification config is missing. - Use `/eval-fetch` when the user provides a Galileo URL/ID or needs evidence pulled into `.galileo/current/debug-packet.json`. - Use `/eval-dataset` when the user wants to create, review, accept, reject, or improve eval cases under `.galileo/eval-dataset/`. - Use `/eval-measure` when the question is whether m