Agent-visible runtime identity says GPT-5 even when using GPT-5.5

What is the problem?

Codex can be configured to use gpt-5.5 with a high/xhigh reasoning profile, but the agent-visible injected instructions still identify Codex as being based on GPT-5.

In the current session, the CLI version is:

codex-cli 0.135.0

The user-selected runtime/model is gpt-5.5 xhigh, but the injected agent instructions still include static wording like:

You are Codex, a coding agent based on GPT-5.

This led the assistant to reject or second-guess accurate user-provided provenance text such as:

Prepared with Codex (GPT-5.5)

Expected behavior

The model-visible runtime context should either:

• expose the effective model identity and reasoning profile, for example gpt-5.5 xhigh, or
• avoid making a static model identity claim in the injected prompt.

The assistant should not contradict the model selected in the CLI or reported by Codex runtime metadata.

Actual behavior

The injected prompt still says the agent is based on GPT-5, even when the active model is GPT-5.5. As a result, the assistant treats accurate model provenance requested by the user as potentially false.

Why this matters

This creates friction for:

• provenance footers in PR descriptions, commits, or generated artifacts
• debugging model-specific behavior
• reporting issues against the correct model/runtime
• user trust when the CLI-selected model and agent-visible identity disagree

Related issues

This looks related to model identity/reporting work, but I did not find an exact duplicate:

• #22944
• #13676
• #23594
• #16548
• #14113
• #6023