Asset Validation

Asset Validation

Validate Codex-native AI assets before considering the package complete.

Scope

Validate:

• Root or scoped AGENTS.md
• .agents/skills/*/SKILL.md
• Companion markdown resources for skills
• .codex/templates/*
• .codex/checklists/*
• explicit support scripts used by the Codex package

Do not use this skill for application linting, unit tests, or runtime deployment checks.

Validation Layers

1. Structure

Confirm:

• The asset is stored in the correct directory
• SKILL.md lives under .agents/skills/<skill-name>/
• Companion resources are colocated with their skill or under .codex/
• The package does not introduce unsupported runtime assumptions

2. Content Rules

Confirm:

• English only
• No secrets or credentials
• No broken references
• No Claude-only runtime instructions
• Instructions are concrete and scoped

3. Frontmatter

For SKILL.md:

• name present
• description present and specific
• Any optional fields are valid and necessary

For AGENTS.md, templates, and checklists:

• No YAML frontmatter unless there is a clear reason

4. Dependency Integrity

Check:

• References to AGENTS.md
• References to .agents/skills/<name>/
• References to .codex/templates/ and .codex/checklists/
• References to local scripts

Flag missing targets and orphaned assets that no workflow will discover.

5. Token Discipline

Check:

• SKILL.md files are concise enough for progressive disclosure
• AGENTS.md files do not bury critical policy in long prose
• Reference material is moved out of hot-path assets when appropriate

Report Format

Return:

• Summary
• Passed checks
• Findings
• Required follow-up

Use validation-checklist.md for the detailed checklist and script patterns.