Plugin Validator

1. Frontmatter

   • Required: name, description, tools, model
   • Optional: skills, disallowedTools

2. Model Selection

   • opus - For critical work (security, architecture)
   • sonnet - For standard work (quality, docs)
   • haiku - For quick tasks

3. Skill References

   • All skills in skills: [...] exist in skills/ directory
   • Skills match agent's purpose

4. Tool Selection

   • Tools are valid Claude Code tools
   • Read-only agents use disallowedTools: Edit, Write

Phase 3: Skill Validation

Delegate to skill-validator skill for each skill, plus:

1. Agent Coverage

   • Which agents reference each skill?
   • Are any skills orphaned (no agent uses them)?

2. Category Balance

   • Domain skills (security, architect, testing, etc.)
   • Language skills (golang, typescript, python, etc.)
   • Cloud skills (aws, gcp, azure)

Phase 4: Command Validation

Check each file in commands/*.md:

1. Frontmatter

   • Required: name, description

2. Content

   • Clear usage instructions
   • Example output shown
   • Error handling documented

Phase 5: Cross-Reference Validation

Check relationships between components:

1. Skill-Agent Mapping

   • All skills referenced by agents exist
   • Identify orphaned skills
   • Identify agents with no skills

2. Consistency

   • Naming conventions followed
   • No duplicate names across components

Red Flags - STOP and Fix

Manifest Red Flags

// ❌ Invalid JSON
{ name: "bosun" }  // Missing quotes

// ❌ Missing version
{ "name": "bosun", "description": "..." }

// ❌ Invalid semver
{ "version": "1.0" }  // Should be "1.0.0"

Agent Red Flags

# ❌ Missing required field