Feature request: add path globs to control when skills are considered

What variant of Codex are you using?

CLI

What feature would you like to see?

Add a path-matching field to Codex skill metadata so a skill can be automatically considered only when the current work involves matching files.

For example, a skill could declare path globs like:

---
name: iconify
description: Search Iconify and turn results into code-ready icon imports or usage.
paths:
  - "apps/web-*/**/*"
  - "packages/ui/**/*"
---

Requested behavior:

• Codex should only automatically surface or activate the skill when work involves files matching those globs.
• The skill should not be considered for unrelated files.
• Manual invocation could still be allowed, depending on the existing skill invocation model.
• Path matching should work for monorepo package boundaries and common glob patterns.

This is different from broad semantic skill matching. Some skills are only correct in specific parts of a repository even if their descriptions sound generally relevant.

Examples:

• A React icon skill should apply under apps/web-* and packages/ui, but not backend services.
• A database migration skill should apply under packages/api/migrations, but not frontend code.
• A generated-client workflow should apply under packages/sdk, but not the app that consumes it.
• A design-system component checklist should apply under packages/ui, but not arbitrary React apps.

Additional information

Claude Code documents a paths field for skills:

https://code.claude.com/docs/en/skills#automatic-discovery-from-nested-directories:~:text=for%20configuration%20format.-,paths,-No

<img width="1107" height="192" alt="Image" src="https://github.com/user-attachments/assets/64dc1705-129a-4a36-b7fc-ffe636eb8b94" />

Its documented behavior is that paths contains glob patterns limiting when a skill is activated: Claude loads the skill automatically only when working with files matching the patterns.

This matters for both correctness and context hygiene:

• Correctness: the model should not apply package-specific workflows to the wrong part of a repo.
• Context hygiene: path scoping avoids surfacing irrelevant skill metadata during unrelated work.
• DX: users can encode “this workflow belongs here” once, instead of repeatedly telling the agent.

This pairs especially well with nested skill discovery, but it is useful independently: even repo-level skills often need file/path-specific applicability.