## Problem

Proxy config files (`config.yaml` / `model_list` JSON) are only validated at proxy startup, and many errors surface as 500s on the first request rather than as clear preflight failures. Operators who want to confirm a config is well-formed before deploying have to either start the proxy (which needs DB credentials, provider API keys, etc.) or hand-parse the YAML and visually inspect it.

Concrete failure modes the new CLI will catch:

1. Typos in `model_name` that pass schema validation but cause "model not found" at runtime 2. Duplicate `model_name` entries that silently shadow each other in the Router 3. Unknown `router_settings.routing_strategy` values (typo of `usage-based-routing-v2` etc.) that get ignored 4. `litellm_params.api_key: os.environ/VAR_NAME` references that point to unset env vars, which currently 500 on first request 5. Malformed top-level keys (camelCase vs snake_case drift) that the proxy silently ignores 6. `litellm_params.model` set to a complex value (dict, list) instead of a string

## Motivation

`litellm doctor` (env), `litellm cost-estimate` (cost), and `litellm token-count` (size) already ship as offline CLIs. The natural fourth member is co…