Caveman Evidence Review skill: what it does and how to install it

What it does. Reads current Caveman Cloud data in the selected project, using overview, cost, score, workflow and savings reports before ranked headroom and bounded trace searches. Inspects representative metadata and compares cohorts without fetching payloads, keeping measured cost, inferred headroom, verified savings and evidence cost separate.

When to use it. Use it to investigate where spend, latency, errors or routing behaviour comes from. It does not start, approve, cancel or roll back experiments, and stops when login, project context or supporting evidence is missing.