agentHost: fix context picker for ChatGPT Codex models

[Giuseppe Cianci (Giuspepe)]

Summary

• keep Codex app-server model/list authoritative for the ChatGPT models and reasoning levels available to the signed-in account
• read default and maximum context limits from the same bundled Codex binary via its supported codex debug models JSON view, using the app-server environment and configuration
• publish context-size choices for ChatGPT subscription models without requiring GitHub authentication, a Copilot subscription, or a matching Copilot catalog entry
• refresh the selected model snapshot when an existing identifier is republished with enriched metadata
• route the selected context window and thinking level to Codex for both Copilot and ChatGPT model providers

Testing

• npm run compile
• npm run hygiene
• 199 targeted unit tests across Codex model refresh, prewarm/session routing, and chat model selection
• git diff --check
• live launch + DAP verification with GitHub authentication disabled and a ChatGPT Plus account:
  • zero Copilot models were present
  • GPT-5.6-Sol showed Low through Ultra thinking levels and 272K / 872K context choices
  • selecting 872K and High produced model_context_window: 872000 and thinkingLevel: high in the actual Codex app-server requests
• live Copilot-provider verification remained intact for its independently sourced context and thinking options

[Copilot]

Copilot review overview

🟢 Approval recommended

The implementation is focused and well tested; the only finding is a non-blocking comment-concision cleanup.

Review tier: Balanced
Findings: 1

New issues introduced by this change (1)

Severity	Finding
	src/​vs/​workbench/​contrib/​chat/​browser/​widget/​input/​chatInputModelSelectionController.ts — This multi-line inline comment narrates the immediately following call. Condense it to a single…

What changed in this PR

Adds context-size selection for ChatGPT Codex models by enriching them with matching Copilot catalog metadata.

Changes:

• Inherits context-size schemas for matching ChatGPT models.
• Refreshes selected-model metadata after catalog enrichment.
• Tests both Copilot and ChatGPT routing and configuration payloads.

File	Description
codexAgent.ts	Enriches ChatGPT models with Copilot context-size metadata.
codexModelRefresh.test.ts	Tests context-size schema publication and removal.
codexPrewarmEviction.test.ts	Tests provider-specific thread configuration.
chatInputModelSelectionController.ts	Refreshes republished selected-model metadata.
chatInputModelSelectionController.test.ts	Verifies metadata refresh without reapplying selection.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.