This is a structured code review agent that prioritizes security and correctness over style. It automatically scopes to your session changes or uncommitted diff, then runs a severity-ranked check list plus domain profiles for things like TypeScript, Python, shell, or smart contracts. The output is concrete: file locations, impact statements, evidence from the actual diff, and suggested fixes. You can pass `--fix` to have it apply changes in severity order and rerun verification. The severity model is pragmatic: critical means exploitable or data loss, high means broken core behavior, medium is maintainability debt. It won't fabricate line numbers or give you vague advice, and it stops when intent is too ambiguous to classify safely.
npx skills add https://github.com/paulrberg/agent-skills --skill code-review