Caveman Manage skill: what it does and how to install it

What it does. Reads project, experiment state and results through Caveman MCP or CLI, then reports samples, quality, guardrails, evidence cost and lifecycle status. It recommends one supported action such as start, approve, cancel or rollback, while blocking unsafe or incomplete changes.

When to use it. For reviewing or managing Caveman Cloud eval-gated experiments. It does not execute lifecycle mutations; after an external operator acts, it re-reads the server state and reports the observed result.