skills-eval

How to run the behavioural skills-eval — drive headless claude -p with prompts and assert the right Shipeasy skill fires, the right MCP tools get called, and the resulting server state is correct. Covers the self-contained eval:fresh run (wipes + seeds its own isolated DB per run), the manual-backend eval, the seed/test commands, every SHIPEASY_EVAL_* option, the report columns, the case format, and common recipes…