Allow custom prompt frontmatter to set additional parameters like model, model_reasoning_effort, ...

What feature would you like to see?

• Claude Code’s slash commands support richer YAML frontmatter (e.g. per-command model).

This makes prompts reproducible and shareable without relying on global config - Claude Slash Commands Doc

• In Codex CLI, custom prompts currently have very limited frontmatter (e.g. description/argument hint), so I can’t make a prompt self-contained with the right model + effort + output style.
• Request: extend Codex custom prompt frontmatter to support settings such as:
• model (e.g. gpt-5.2)
• model_reasoning_effort (e.g. medium|high|xhigh)
• model_verbosity (e.g. low|medium|high)
• tool constraints (e.g. allowed_tools, or an equivalent)
• optionally: a way to prevent model switching from inside the prompt

(disable_model_invocation or similar)

• Why: I maintain a prompt library (shared across machines/WSL) and want each prompt to be deterministic about model/effort/verbosity and safety constraints, without editing config.toml or passing flags every run.