Validation Rules

Required checks

1. Frontmatter exists and includes only required routing fields:
   • name
   • description
2. Required files exist:
   • SKILL.md
   • agents/openai.yaml
   • references/usage-patterns.md
   • scripts/validate.sh
3. Link-first pattern is explicit:
   • command -v <link_name>
   • uxc link <link_name> <host>
4. Help-first pattern is explicit:
   • <link_name> -h
   • <link_name> <operation> -h
5. Execution input style is explicit:
   • at least one key=value example
   • at least one bare JSON positional example
6. Link naming convention is explicit:
   • use <provider>-<protocol>-cli
   • keep the name fixed for the skill
7. Discovery workflow is explicit:
   • start from user-provided host
   • include search + uxc <endpoint> -h probe before finalizing protocol/path
8. Auth detection workflow is explicit:
   • include how auth requirement is determined from probe/first call errors
   • include binding verification when OAuth/binding is used

Banned defaults

Reject these in default examples and workflow text:

• list/describe/call old command framing
• removed input flags or deprecated invocation forms
• raw JSON passed through --args
• dynamic runtime command renaming for link conflicts
• protocol/path/auth assumptions made without verification

OAuth and error handling boundary

For wrapper skills, keep provider-specific notes minimal and reuse skills/uxc for canonical OAuth and error playbooks.