Support disable-model-invocation in SKILL.md

What variant of Codex are you using?

App and CLI

What feature would you like to see?

The disable-model-invocation flag in SKILL.md seems to be slowly becoming more standardized, with Claude Code, Cursor, Pi, etc supporting it. Currently, Codex supports an agents/openai.yaml file with:

policy:
  allow_implicit_invocation: false

But Codex doesn't support the disable-model-invocation flag.

I think this is important because many public skills work under the assumption of the disable-model-invocation flag such as matt-pocock/skills, and having a consistent standard makes it easier to pass skills around.

Suggested fix/feature:
Allow disable-model-invocation flag to work but still allow agents/openai.yaml to take precedence.

Additional information

_No response_

policy:
  allow_implicit_invocation: false