two-claude-review

To get a skeptical staff-engineer review of your implementation plan, this skill uses a two-Claude pattern where Claude A generates the plan and Claude B critiques it as a reviewer subagent. It challenges architectural decisions and proposes safer alternatives to uncover gaps early.

The two-Claude pattern for architectural planning is a process where one Claude instance writes a complete plan with requirements and constraints, while a second reviewer subagent applies a staff-engineer prompt to critically evaluate edge-cases and risky assumptions before implementation.

You find gaps and risky assumptions by generating a plan with Claude A and requesting a reviewer assessment using the staff-engineer prompt provided by the skill. This iterative refinement continues with revisions until the architectural plan is solid and ready for implementation.

Yes, you need git and tmux to use this Claude review workflow, as the skill dependencies include git, tmux, and claude to manage the planning worktree and execute the structured plan creation and reviewer subagent phases.

You should use a reviewer subagent for multi-phase initiatives when early risk identification and rigorous critique are essential. It is specifically designed for complex features and architectural decisions where unverified assumptions carry high implementation risk.