plan-ceo-review

Invoke the plan-ceo-review skill on the plan file. It walks through four scope modes (expansion, selective expansion, hold, reduction), then runs an 11-section deep review where every finding is presented as an individual decision question before any change is made.

The review covers architecture, error and rescue mapping, security threat modeling, data flow edge cases, code quality, tests, performance, observability, deployment and rollout, long-term trajectory, and design/UX. No section may be skipped; sections with zero findings are explicitly marked as such.

Yes. After the section review, it runs an outside voice check using the Codex CLI when installed and authenticated, falling back to a Claude subagent otherwise. Cross-model disagreements are surfaced to the user, who decides each one explicitly.

No. Every finding, including ones with obvious fixes, requires explicit user approval through a structured decision question. Outside voice recommendations are informational only and are never applied without user consent.

Use CEO-mode review when questioning scope, ambition, or strategy of a plan, such as when asked to think bigger or rethink an approach. Use an engineering plan review for architecture and implementation rigor once scope is settled.