Code review

One skill, clear modes. Model and orchestration policy lives in the shared AGENTS.md; runtime capability names live in ../RUNTIME_TOOLS.md.

Mode	When	Depth	Primary model
closeout (default)	Before commit/ship; "ready?"; autoreview; second opinion	P0/P1 blockers first	One reviewer selected by shared policy
axes	User wants Standards vs Spec; "review since main"; issue/PRD fidelity	Two independent reports	Runtime-native agents; distinct model only when requested and available
full	Large PR, multi-surface, security-critical, or `/workflows-review`	Multi-agent P1–P3	Load `workflows-review` skill

How this differs from `workflows-review`

They are different jobs, not duplicates: