Feature request: Codex Instruction and Skill Compatibility Check

What variant of Codex are you using?

ChatGPT desktop app and Codex CLI.

What feature would you like to see?

Scope

This should be a Codex-only capability, available across Codex surfaces. It would evaluate active AGENTS.md guidance and skills against the current Codex runtime: supported models, tools, surfaces, configuration, and conventions. It is not intended to validate generic or cross-ecosystem agent skills.

Problem

AGENTS.md files and skills can drift as Codex evolves. They may load correctly while still referring to unavailable models, tools, roles, commands, paths, or assumptions copied from another agent ecosystem.

Example

An AGENTS.md rule requires haiku-swarm-tracer with the haiku model. In an active Codex runtime, that tracer may not be a directly usable skill and the required model may be unavailable. The check should flag the rule as incompatible, explain why, and recommend the closest supported explorer or tracer.

Proposal

Provide an official, OpenAI-maintained compatibility check, exposed as a built-in Codex skill and backed by runtime-aware diagnostics.

The check should identify:

• unavailable or stale runtime references;
• conflicting or non-Codex guidance;
• scope, precedence, and instruction-size concerns;
• Codex-native replacement suggestions;
• a severity and confidence level for each finding.

It should be read-only by default and offer a reviewable patch only after explicit user approval.

Additional information

This capability must evolve with Codex and understand the active runtime. A generic community skill would itself become stale and cannot reliably validate Codex-specific capabilities.

It would also help users keep their own skills current: when Codex adds or changes a model, tool, surface, configuration option, or capability, the check could identify affected instructions and skills and recommend the appropriate Codex-native update.