Make Agent Profiles Portable

What variant of Codex are you using?

CLI

What feature would you like to see?

I understand why Codex 0.149.1 introduced bounded configuration overrides:
https://github.com/openai/codex/pull/39299

TL;DR: Installing an agent profile should not silently install arbitrary executable code.

However, as an agent publisher, this change breaks a core capability I was excited about:
functionally portable agent profiles.

Before, a role could package both behavior and enforcement, including its instructions, model configuration, start logic, identity binding, and completion validation. Now, executable guarantees a required to be in a separately trusted project or plugin layer, so the profile alone no longer describes the complete agent.

Removing role-local hooks makes a role less like a program and more like a prompt.

This makes sophisticated agents harder to share, install, version, test, and reproduce reliably across environments. The security change is defensible, but it should have shipped with a first-class agent package that can declare its required lifecycle capabilities, request scoped user approval, and let developers distribute the complete system as one inspectable, portable unit.

Additional information

No response

Start by reading the bounded configuration overrides change in PR #39299 and tracing how agent profiles currently represent lifecycle capabilities. A complete implementation would need an inspectable portable package model, scoped user approval for requested capabilities, and a way to preserve profile behavior across environments; no files or tests are named in the issue.