feat: dynamic model instructions and gpt-5.3-codex support

[iam-brain]

Summary

• Implemented dynamic account selection and rate-limit tracking keyed by normalized model slugs, removing reliance on a static ModelFamily list.
• Added a server-first instruction resolution flow: /codex/models metadata is checked first, followed by a GitHub-based fallback cascade that dynamically maps gpt_*_prompt.md files to model slugs.
• Added first-class support for gpt-5.3-codex across code, configuration templates, and the testing suite.
• Updated documentation and personality logic to align with modern model variant handling and the removal of legacy substring-based normalization.

Test Plan

• Verified all 457 tests pass (npm test).
• Confirmed build succeeds (npm run build).
• Audited documentation for consistency across all guides.

[chatgpt-codex-connector]

💡 Codex Review

opencode-openai-codex-multi-auth/index.ts

Lines 283 to 289 in 4806546

	for (const modelId of Object.keys(models)) {
	const parsed = parseCodexMetadataModel(modelId);
	if (parsed?.effort) {
	delete models[modelId];
	continue;
	}
	if (!isAllowedMetadataModel(modelId)) delete models[modelId];

Preserve non-codex variant model entries

The cleanup loop in normalizeProviderModelMetadata deletes every model key that is not a codex slug or in the small legacy allow-list, which unintentionally drops the non-codex variant entries from the legacy config (e.g., gpt-5.2-high, gpt-5.2-low, etc.). Because the legacy config expresses variants as distinct model IDs, removing them here means those variants disappear from the provider metadata and users can no longer select them in OpenCode, even though they are still present in config/opencode-legacy.json. Consider extending the allow-list or normalizing general-purpose variants instead of deleting them.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[iam-brain]

@codex audit all updates in the PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: db0039678c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Allow base GPT-5.x slugs in dynamic model map

GPT_GENERAL_DYNAMIC_REGEX requires an effort suffix (e.g., -low) to recognize non‑codex GPT‑5.x models, so a base model slug like gpt-5.3 (no suffix) won’t be recognized by getNormalizedModel. Because transformRequestBody now throws UnknownModelError when getNormalizedModel returns undefined, any new base GPT‑5.x model coming from /codex/models (or user config) without a suffix will be rejected even though it should be valid. This breaks forward compatibility for new general GPT‑5.x releases unless you keep hard‑coding them in MODEL_MAP.

Useful? React with 👍 / 👎.