Add a Codex/OpenAI model

Each model's facts and policy live on its CODEX_MODELS entry in packages/core/src/codex/models.ts. Background routing tiers, review preferences, prompt profiles, and the Codex-aligned condense default are all derived from it. Docs and pinned tests stay manual.

1. Get the facts first

• Read the official model page (developers.openai.com/api/docs/models/<id>) and the "Using GPT-N" guide: context window, max output, input cap, reasoning efforts and default, Fast/Ultrafast support.
• Check whether it's on ChatGPT/Codex sign-in or API key only, and whether Codex's catalog marks it use_responses_lite or gives it different sign-in limits.
• Don't guess. If something isn't published, say so in the final summary.

2. packages/core/src/codex/models.ts

• CODEX_MODELS: add the entry. Required: routingTier (cheap Luna-class, balanced Sol/Terra-class, deep_reasoning Astra-class) and promptProfile (reasoning for Sol/Terra/Astra-class, compatibility for Luna/Spark-class). Set codexAlignedCondense: true if API prompts above 272K cost more. Then limits, efforts, defaultTextVerbosity, serviceTiers, and apiAvailable: false if sign-in only.
• Array position matters: within a tier, earlier entries win for background routing and reviews. Keep stronger models first (Sols before Terras).
• CODEX_PICKER_MODEL_IDS: picker position. Update the "last verified" comment on CODEX_CHATGPT_BACKEND_MODEL_IDS.
• Only if needed: CODEX_OAUTH_MODEL_OVERRIDES (different sign-in limits or a larger condenseTargetInputTokens), CODEX_OAUTH_RESPONSES_LITE_MODELS, and for replacements SUPERSEDED_MODEL_MIGRATIONS and getCodexUnavailableModelFallback.

3. Tests

• The "official N models" picker lists in packages/core/src/codex/models.test.ts, src/agent/providers/codex/CodexProvider.test.ts, and the count in apps/cli/src/config.test.ts. Add a capability test for the new model.
• packages/core/src/codex/models.test.ts "Codex routing tiers" if the strongest model in a tier changes.

4. Docs

• resources/builtin-skills/documentation/references/: capabilities.md (picker list, availability, Speed tiers), settings.md (limits, condense defaults, text verbosity), providers.md (prompt profiles), troubleshooting.md (rollout and Speed lists).
• CHANGELOG.md under ## Unreleased, then npm run docs:generate.

5. Validate

• npx oxfmt <touched files>, then npm run lint.
• Focused tests: npx vitest run packages/core/src/codex packages/core/src/promptProfilePolicy.test.ts src/agent/providers/codex src/agent/backgroundModelRouter.test.ts src/agent/modelCondenseThresholds.test.ts apps/desktop/src/desktopModelCatalog.test.ts apps/cli/src/config.test.ts.