Model provider

Model access is intentionally separate from engineering methodology. The plugin consumes the provider-neutral Harness agent/LLM interfaces and does not import, select, or call a provider adapter.

For GPT-5.6 Max through an OpenAI-compatible CLI proxy, configure the deployment's existing provider settings with the proxy's provider id, endpoint/command, credential binding, and model name. The effective agent route should be:

provider: <OpenAI-compatible CLI proxy adapter>
model: GPT-5.6 Max

The exact provider id is deployment-specific and must be read from the Harness model settings/adapter catalog. Never hard-code secrets, endpoint assumptions, or a provider package into fullstack-expert. This keeps the same methodology usable with another provider or model.