eval, review

eval

kcap eval [<session>] [--model haiku|sonnet|opus] [--chain] [--questions <csv>] [--skip <csv>] [--threshold <bytes>] [--baseline-out <path>]
kcap eval --list-questions

Runs a local judge through your own Claude Code sign-in and stores the result. The question catalog comes from the server; categories are safety, plan_adherence, quality, efficiency. --questions and --skip are mutually exclusive; given without a value they stop the run with exit 2. --baseline-out also writes one JSON record per question with usage, cost and timing

review

kcap review <pr-url | owner/repo#N>

Opens a local Claude Code review with a tool server scoped to the pull or merge request. No other options. Reports No review context found rather than starting a context-free review