What problem does it solve? Coordinating non-trivial software changes across multiple AI agents risks unverified model routing, unbounded delegation, and unreviewed worker output. This Skill enforces a verified orchestration workflow where a root agent plans and reviews while bounded workers implement and explore. ## Core Features & Use Cases - Verified routing: Fail-closed preflight checks confirm the root runs gpt-5.6-sol at medium effort and workers run gpt-5.6-luna at max effort by parsing persisted rollout metadata. - Bounded delegation: Implementation and exploration workers receive explicit file ownership, constraints, and verification commands, with conflict detection for overlapping edits. - Root-owned review and publication: The root independently inspects diffs, runs tests, and is the only agent allowed to commit, push, or publish. - Use Case: When implementing a multi-file feature in Codex, use this Skill to verify routing, delegate bounded code changes to a Luna worker, and review the actual workspace diff before committing. ## Quick Start Ask Codex to use the codex-maestro skill to plan and implement your feature, then run the routing preflight before delegating any work.
