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

What it does. Reads caveman learn report --json, explains the Cave Score and ranked token sinks, then proposes fixes one at a time. With consent, it trims costly configuration, offloads repeated context to Cavemem, or runs controlled experiments for repeated procedures, measuring and reporting each result.

When to use it. Use it to lower agent token costs, review Caveman savings, trim heavy CLAUDE.md files, or move repeated context into memory. It does not alter load-bearing findings or turn behavioural observations into mandatory changes.