Official plugin skills can trigger description length validation errors

What happened?

Codex repeatedly reports a length validation error while loading or validating skills. This is surprising because the skills involved appear to be official/plugin-provided skills, not only user-authored skills.

On my machine, the official Codex plugin cache contains many SKILL.md frontmatter descriptions that are substantially longer than a short description field. Examples from the local plugin cache:

918  .codex/plugins/cache/openai-curated/build-web-data-visualization/.../visualization-strategy-and-critique/SKILL.md
730  .codex/plugins/cache/openai-curated/build-web-data-visualization/.../uml-and-software-architecture-visualization/SKILL.md
722  .codex/plugins/cache/openai-curated/hyperframes/.../hyperframes/SKILL.md
550  .codex/plugins/cache/openai-curated/openai-developers/.../openai-platform-api-key/SKILL.md
501  .codex/plugins/cache/openai-curated-remote/data-analytics/.../index/SKILL.md
570  .codex/skills/.system/imagegen/SKILL.md

User-level skills were also checked separately: 73 skill directories / 73 SKILL.md files, frontmatter present, and no malformed frontmatter detected. This makes the error feel like the CLI is enforcing a schema/length constraint against official curated/bundled skill metadata that the official packages themselves do not satisfy.

Expected behavior

Official/bundled/curated skills should not produce user-facing length validation errors in normal Codex startup or skill loading.

If a length limit is intentional, Codex should either:

• ensure all official skills satisfy that limit before shipping,
• exempt trusted official plugin cache metadata from user-skill validation,
• truncate or normalize official descriptions before feeding them into the schema-constrained path, or
• show an actionable error that names the specific skill file and field that failed validation.

Actual behavior

Codex reports a generic/repeated length error, which is hard to act on. From the user perspective, this looks unreasonable because the affected skills appear to be official Codex plugin/runtime skills.

Environment

Codex Doctor v0.136.0
OS: macOS 26.5.1, arm64
Install method: npm
Package: @openai/codex
Runtime version: 0.136.0
Auth: ChatGPT auth configured
codex doctor: 16 ok, 1 warn, 0 fail degraded

codex doctor itself did not fail on skills, so this may be specific to the runtime path that loads/exposes skills/plugins rather than the base doctor checks.

Additional context

This came up after updating Codex to 0.136.0 and refreshing a shared local skill library. The local user-skill library validates cleanly, while the official plugin cache contains many long official descriptions. The key request is to make the validator and official skill metadata consistent, and to make any remaining error identify the exact skill and field instead of only saying the length is wrong.