model-optimizer-litelisted

# Model Optimizer Lite Use this skill in Claude or Codex; select the profile for the actual host. Keep one capable owner when the current session can finish the work. A recommendation does not switch models or authorize a delegated run. Preserve explicit model pins, scope, data restrictions, and the user's existing execution permissions. Read [shared routing policy](references/shared-policy.md) for routing or escalation. Load only the relevant host profile: [Claude](references/claude-profile.md) or [Codex and Astra](references/codex-profile.md). Load the other profile only for a cross-provider handoff. ## Choose the next action - Keep the current model when its context and capability fit the task. - For a new task or a justified escalation, select among models available in the actual client. State the task bottleneck and required checks. - Use the host's native model selector or delegation controls when supported. Confirm the resulting model if the host exposes authoritative metadata. - If the current desktop/chat environment cannot control models or run local commands, give the recommendation and a compact handoff. Describe the required manual step; do not claim an automatic switch. - A provider error, timeout, refusal, or missing capability is evidence to inspect. Do not silently change providers, relax permissions, or repeat the same run. ## Optional deterministic helpers Python 3.9+ is needed only for these helpers. Skill guidance and installation do not