Allow launching Custom GPTs inside Codex / Work with preserved instructions and knowledge

What variant of Codex are you using?

Codex App

What feature would you like to see?

Add the ability to launch an existing Custom GPT directly inside Codex and/or ChatGPT Work while preserving its configured instructions, knowledge files, and intended behavior.
Today, specialized Custom GPTs are used mainly in ChatGPT, while Codex and Work provide the tools needed for real project work: repository access, terminal/shell, tests, and agent execution.
As a result, users who already created a highly specialized Custom GPT have to manually recreate its behavior using AGENTS.md, Project Instructions, and additional prompt files. This is duplicative, error-prone, and makes it difficult to keep the behavior consistent.
Ideally, Codex and Work should allow selecting an existing Custom GPT as the active agent/profile for a session, while carrying over its custom instructions, knowledge files, and behavior where compatible.

Additional information

Example use case: a custom application-security / pentesting GPT could retain its security-review methodology while gaining access to the full repository, terminal, tests, git history, and local tooling. This would turn existing Custom GPTs into reusable agent profiles instead of chat-only configurations.

Start by reviewing how Codex App and ChatGPT Work currently load AGENTS.md, Project Instructions, and additional prompt files, then trace the session or agent-profile entry point. Compare that flow with existing Custom GPT instructions and knowledge files. Done means an existing Custom GPT can be selected for a session and its compatible instructions, knowledge, and behavior are preserved alongside repository, terminal, test, and git access.