Codex_audit

Audit current or saved Codex tasks for token burn, cached-context amplification, model and effort choices, compactions, tool churn, and native subagent usage; then recommend codebase-specific routing and lifecycle experiments. Use when a user asks why five-hour or weekly Codex usage disappeared quickly, whether Sol/Terra/Luna or subagents are efficient, how to reduce task usage, or how to benchmark an orchestrator. Keep all actions read-only and recommendation-only unless the user separately authorizes changes.

Choose and operate a center-out Codex agent topology for spec-gated repository work while preserving one stable control task. Use automatically for broad implementation, audit, research, or QA work with clear contracts, checklists, blockerboards, or measurable acceptance criteria; route the control plane through Luna-xhigh, use Terra-high for judgment/integration, and escalate to Sol only for material architecture, security, release, or conflicting-evidence decisions.

Continuously monitor a live Codex orchestration tree and refresh compact agent telemetry plus human Markdown and HTML dashboards. Use for long-running multi-agent tasks, live token burn, route calibration, rework comparisons, stuck-agent detection, or when a Sol orchestrator needs inexpensive routing signals at decision boundaries.