Reference Files

Audit Orchestration

• workflow-patterns.md - Multi-auditor invocation patterns and decision matrix
• report-compilation.md - Unified report structure and priority reconciliation

Evaluation Standards and Troubleshooting

• evaluation-criteria.md - Comprehensive standards for each component type
• common-issues.md - Frequent problems and specific fixes with examples
• anti-patterns.md - Common mistakes to avoid when building customizations

Shared References (Used by All Authoring Skills)

• naming-conventions.md - Patterns for agents, commands, skills, hooks, and output-styles
• frontmatter-requirements.md - Complete YAML specification for each component type
• when-to-use-what.md - Decision guide for choosing agents vs skills vs commands vs output-styles
• file-organization.md - Directory structure and layout best practices
• hook-events.md - Hook event types and timing reference
• customization-examples.md - Real-world examples across all component types

---

Audit Coordinator

Orchestrates comprehensive audits by coordinating multiple specialized auditors and compiling their findings into unified reports.

Available Auditors

The audit ecosystem includes:

evaluator (Agent)

Purpose: General correctness, clarity, and effectiveness validation Scope: All customization types Focus: YAML validation, required fields, structure, naming conventions, context economy Invocation: Via Task tool with subagent_type='evaluator'

audit-skill (Skill)

Purpose: Skill discoverability and triggering effectiveness Scope: Skills only Focus: Description quality, trigger phrase coverage, progressive disclosure, discovery score Invocation: Via Skill tool or auto-triggers on skill-related queries

audit-hook (Skill)

Purpose: Hook safety, correctness, and performance Scope: Hooks only Focus: JSON handling, exit codes, error handling, performance, settings.json registration Invocation: Via Skill tool or auto-triggers on hook-related queries

test-runner (Agent)

Purpose: Functional testing and execution validation Scope: All customization types Focus: Test generation, execution, edge cases, integration testing Invocation: Via Task tool with subagent_type='test-runner'

audit-agent (Skill)

Purpose: Agent-specific validation for model selection, tool restrictions, and focus areas Scope: Agents only Focus: Model appropriateness (Sonnet/Haiku/Opus), tool permissions, focus area quality, approach completeness Invocation: Via Skill tool or auto-triggers on agent-related queries

audit-command (Skill)

Purpose: Command delegation and simplicity validation Scope: Commands only Focus: Delegation clarity, simplicity enforcement (6-80 lines), argument handling, documentation proportionality Invocation: Via Skill tool or auto-triggers on command-related queries

audit-output-style (Skill)

Purpose: Output-style persona and behavior validation Scope: Output-styles only Focus: Persona definition clarity, behavior specification concreteness, keep-coding-instructions decision, scope alignment Invocation: Via Skill tool or auto-triggers on output-style-related queries

Orchestration Workflow

Step 0: Determine Audit Scope

FIRST, determine the scope of the audit by analyzing the user's request for scope indicators:

Local/Project Scope (.claude/ in current directory):

• Keywords: "local", "project", ".claude", "current directory"
• Search Path: .claude/ only
• Use Case: Audit project-specific customizations before committing
• Example: "Audit my local setup", "Check the project's .claude directory"

Personal/User Scope (~/.claude/ global configuration):

• Keywords: "personal", "user", "~/.claude", "global"
• Search Path: ~/.claude/ only
• Use Case: Validate personal global configuration independently
• Example: "Audit my personal setup", "Check my user-level configuration"

Full Scope (both .claude/ and ~/.claude/):

• Keywords: "full", "both", "everything", "complete", "all", or no scope specified
• Search Paths: Both .claude/ and ~/.claude/
• Use Case: Comprehensive audit including conflict detection
• Example: "Audit my entire setup", "Check everything", or just "/audit-setup"
• Default: If no scope indicators found, use Full scope

Scope Detection Logic:

1. Parse user request for scope keywords
2. If "local" or "project" found → Local scope
3. If "personal" or "user" or "global" found → Personal scope
4. If "f

...