A Claude Code kit with two roles that grade opposite halves of the same round. The MENTOR coaches how each request was framed before the work runs, then grades the ask — framing, leverage, learnings applied. The CRITIC judges, in fresh context and its own words, whether the thing you ASKED for actually happened, then grades the delivery — did it happen, was there evidence, did it stay in scope. Both write to one append-only ledger per repo, and the SCORECARD renders it as a published board where the human prompter and the model are scored side by side.
Claude Code3 Skills