[FEATURE]: Request GitHub Copilot auto model routing API access + chat.model plugin hook

[fastdrumr]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Request GitHub to grant opencode access to the Copilot auto model routing API (/models/session).

VS Code's Copilot extension has a server-side auto model selection feature that uses ML to classify prompt complexity and route to the best model. The API endpoints are:

• POST /models/session — creates an auto-mode session, returns available models and a session token
• POST /models/session/intent — per-turn routing, returns ranked candidate models based on prompt analysis

This API is currently restricted to first-party OAuth clients (VS Code's client ID). When opencode's client ID (Ov23li8tweQw6odWQebz) calls these endpoints, it gets a 404.

Why this matters:

• The feat: compact and other improvements #1 user request for Copilot in opencode is auto model selection ([FEATURE]: Expose GitHub Copilot "Auto" option in model selector #10093, multiple upvotes)
• VS Code auto mode includes a 10% billing discount via the session token
• Server-side routing uses ML classification that outperforms any client-side heuristic
• PR feat(provider): add GitHub Copilot auto model selection #20136 implements a client-side workaround using GPT-5-mini as a classifier, but the server-side API would be better (faster, more accurate, includes the discount)

What's needed:
The opencode team already has an official partnership with GitHub for Copilot support. Requesting access to the /models/session and /models/session/intent endpoints for opencode's OAuth client ID would enable native auto model selection matching VS Code's experience.

Additionally, a chat.model plugin hook would allow the community to implement model routing as external plugins regardless of API access. Proposed signature:

"chat.model": async (incoming: {
  sessionID: string
  agent: string
  model: Provider.Model
  provider: Provider.Info
  message: MessageV2.User
}, output: {
  model: Provider.Model
}) => void

This would enable auto model selection, cost-based routing, A/B testing, and fallback chains as external plugins.

[pljones]

I'd expect this to be beneficial to the provider, too -- reducing unnecessary load on expensive models, so long as the "Auto" pre-filter load isn't overly expensive and the ML review is accurate.

Ideally, this would be available to free tier users, too, for the same reason.