What problem does it solve? Caveman optimization reports describe aggregate usage shapes but do not prove that a code change is safe or worthwhile. This Skill turns a report-only observation into an operator-approved candidate change validated by a paired baseline evaluation, preventing premature edits and unsupported savings claims. ## Core Features & Use Cases - Observation Reading: Runs caveman opportunities list and reads only the report_only_observations array, preserving server-provided titles and observations verbatim while rejecting retired profile ids. - Operator-Gated Workflow: Requires an explicit operator choice before inspecting callsites, and approval of the candidate and eval design before any edit. - Paired Baseline Evaluation: Designs and runs baseline and candidate arms on identical fixed inputs, recording quality checks, token or byte cost measures, fixtures, and confounders. - Use Case: A Caveman report flags a tool-output-size-profile observation. Use this Skill to present the observation to the operator, design a minimal candidate change at the evidenced callsite, run a paired eval, and report keep, reject, or inconclusive without claiming dollar savings. ## Quick Start Ask the AI to inspect the current Caveman optimization report and walk you through evaluating one observation with a paired baseline test.
