Agent Skills: Audit Coordinator

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 "full" or "both" or "everything" found → Full scope
5. Otherwise → Full scope (default)

Report Scope Section: Include scope information in report header:

# Comprehensive Audit Report

**Scope**: Local Project | Personal/User | Full Setup
**Project Path**: .claude/ {or "N/A - not in scope"}
**Personal Path**: ~/.claude/ {or "N/A - not in scope"}
**Date**: {timestamp}

{If full scope and both exist, add:}
**Conflicts**: {count} name(s) exist in both scopes (project takes precedence)

Step 1: Identify Target Type

Determine what needs auditing within the selected scope:

Single File:

• Agent file (*.md in agents/)
• Skill (SKILL.md in skills/*/SKILL.md)
• Hook (.sh or.py in hooks/)
• Command (*.md in commands/)
• Output-style (*.md in output-styles/)

Multiple Files:

• All skills
• All hooks
• All agents
• Entire setup

Context Clues:

• File path mentioned
• Type specified ("audit my hook", "check this skill")
• General request ("audit my setup", "review everything")

Path Filtering by Scope:

• Local scope: Only search in .claude/ directory
• Personal scope: Only search in ~/.claude/ directory
• Full scope: Search both, noting which files come from which scope

Step 2: Determine Appropriate Auditors

Use decision matrix based on target type:

Agent:

• Primary: audit-agent (model, tools, focus areas, approach)
• Secondary: evaluator (structure)
• Optional: test-runner (if testing requested)

Skill:

• Primary: audit-skill (discoverability)
• Secondary: evaluator (structure)
• Optional: test-runner (functionality)

Hook:

• Primary: audit-hook (safety and correctness)
• Secondary: evaluator (structure)

Command:

• Primary: audit-command (delegation, simplicity, arguments)
• Secondary: evaluator (structure)

Output-Style:

• Primary: audit-output-style (persona, behaviors, coding-instructions)
• Secondary: evaluator (structure)
• Optional: test-runner (effectiveness)

Setup (All):

• audit-agent (all agents)
• audit-skill (all skills)
• audit-hook (all hooks)
• audit-command (all commands)
• audit-output-style (all output-styles)
• evaluator (comprehensive)
• Can run in parallel

Step 3: Invoke Auditors

Execute auditors in appropriate sequence, passing scope information:

Sequential (when results depend on each other):

audit-skill → evaluator → test-runner

Parallel (when independent):

audit-skill (all skills) || audit-hook (all hooks) || evaluator (agents/commands)

Single (when only one needed):

audit-hook → done

Scope Filtering During Invocation:

• When invoking auditors, only pass files from the selected scope
• For local scope: Only pass paths starting with .claude/
• For personal scope: Only pass paths starting with ~/.claude/ or /Users/.../.claude/
• For full scope: Pass all paths, but note scope origin in reports

Step 4: Compile Reports

Collect findings from all auditors and create unified report with scope information.

Scope-Aware Report Compilation:

• Group findings by scope (local vs personal) when in full scope mode
• Note which scope each component belongs to
• Identify conflicts (same name in both scopes)
• Remember: project-local files take precedence over personal files

Step 5: Generate Unified Summary

Consolidate recommendations by priority and provide next steps.

Scope-Specific Guidance:

• Local scope: Focus on project-specific best practices, pre-commit validation
• Personal scope: Focus on global configuration consistency, tool compatibility
• Full scope: Include conflict resolution recommendations

Target-Specific Patterns

Pattern: Single Skill Audit

User Query: "Audit my audit-bash skill"

Workflow:

1. Invoke audit-skill for discoverability analysis
2. Invoke evaluator for structure validation
3. Compile reports
4. Generate unified recommendations

Output:

• Discovery score
• Structure assessment
• Progressive disclosure status
• Consolidated recommendations

Pattern: Single Hook Audit

User Query: "Check my validate-config.py hook"

Workflow:

1. Invoke audit-hook for safety and correctness
2. Optionally invoke evaluator for structure
3. Compile reports
4. Generate unified recommendations

Output:

• Safety compliance status
• Exit code correctness
• Error handling assessment
• Performance analysis
• Consolidated recommendations

Pattern: Setup-Wide Audit

User Query: "Audit my entire Claude Code setup"

Workflow:

1. Invoke evaluator for comprehensive setup analysis
2. Invoke audit-skill for all skills
3. Invoke audit-hook for all hooks
4. Run in parallel when possible
5. Compile all reports
6. Generate prioritized recommendations

Output:

• Setup summary (counts, sizes, context usage)
• Component-specific findings
• Cross-cutting issues
• Prioritized action items

Pattern: Multiple Component Types

User Query: "Audit all my skills and hooks"

Workflow:

1. Invoke audit-skill for all skills (can run in parallel)
2. Invoke audit-hook for all hooks (can run in parallel)
3. Compile reports
4. Generate unified summary

Output:

• Skills: Discovery scores, structure
• Hooks: Safety compliance, performance
• Consolidated recommendations

Report Compilation

When multiple auditors run, compile findings:

Consolidation Strategy

1. Collect all findings from each auditor
2. Group by severity: Critical → Important → Nice-to-Have
3. Deduplicate similar issues across auditors
4. Reconcile priorities when auditors disagree
5. Generate unified recommendations

Priority Reconciliation

When different auditors assign different priorities:

Rule 1: Critical from any auditor → Critical overall Rule 2: Important + Important → Critical Rule 3: Important + Nice-to-Have → Important Rule 4: Nice-to-Have + Nice-to-Have → Nice-to-Have

Unified Report Structure

# Comprehensive Audit Report

**Target**: {what was audited}
**Date**: {YYYY-MM-DD HH:MM}
**Auditors**: {list of auditors invoked}

## Executive Summary

{1-2 sentence overview of findings}

## Overall Status

**Health Score**: {composite score}

- {Auditor 1}: {status}
- {Auditor 2}: {status}
- {Auditor 3}: {status}

## Critical Issues

{Must-fix issues from any auditor}

## Important Issues

{Should-fix issues}

## Nice-to-Have Improvements

{Polish items}

## Detailed Findings by Component

### {Component 1}

{Findings from relevant auditors}

### {Component 2}

{Findings from relevant auditors}

## Prioritized Action Items

1. **Critical**: {consolidated must-fix items}
2. **Important**: {consolidated should-fix items}
3. **Nice-to-Have**: {consolidated polish items}

## Next Steps

{Specific, actionable next steps}

Quick Usage Examples

Audit a skill:

User: "Audit my audit-bash skill"
Assistant: [Invokes audit-skill, evaluator; compiles report]

Audit a hook:

User: "Check my validate-config.py hook"
Assistant: [Invokes audit-hook; generates report]

Audit entire setup (full scope - default):

User: "Audit my complete Claude Code setup"
Assistant: [Invokes evaluator, audit-skill, audit-hook in parallel; compiles comprehensive report]

Audit local project setup:

User: "Audit my local .claude configuration"
User: "/audit-setup local"
Assistant: [Determines scope: local; audits only .claude/ directory; generates project-specific report]

Audit personal global setup:

User: "Audit my personal Claude setup"
User: "/audit-setup personal"
Assistant: [Determines scope: personal; audits only ~/.claude/ directory; generates user-level report]

Audit with conflict detection:

User: "Audit everything and show me any conflicts"
User: "/audit-setup full"
Assistant: [Determines scope: full; audits both scopes; identifies conflicting names; generates comprehensive report]

Audit multiple skills:

User: "Check all my skills for discoverability"
Assistant: [Invokes audit-skill for each skill; generates consolidated report]

Integration with Other Auditors

With audit-skill

When to use together:

• Comprehensive skill analysis
• Combining discoverability + structure validation

Sequence: audit-skill → evaluator Output: Discovery score + structure assessment

With audit-hook

When to use together:

• Complete hook validation
• Safety + structure analysis

Sequence: audit-hook → evaluator (optional) Output: Safety compliance + structure validation

With evaluator

When to use together:

• Always, for structural validation
• Complements specialized auditors

Sequence: Specialized auditor first, then evaluator Output: Specialized analysis + general validation

With test-runner

When to use together:

• Functional validation requested
• After structure/discovery validation

Sequence: Other auditors → test-runner Output: Design validation + functional testing

Decision Matrix

Quick reference for which auditors to invoke:

| Target | Primary Auditor | Secondary | Optional | Sequence | | ------------ | ------------------ | --------- | ----------- | ---------- | | Skill | audit-skill | evaluator | test-runner | Sequential | | Hook | audit-hook | evaluator | - | Sequential | | Agent | audit-agent | evaluator | test-runner | Sequential | | Command | audit-command | evaluator | - | Sequential | | Output-Style | audit-output-style | evaluator | test-runner | Sequential | | All Skills | audit-skill | evaluator | - | Parallel | | All Hooks | audit-hook | evaluator | - | Parallel | | All Agents | audit-agent | evaluator | - | Parallel | | All Commands | audit-command | evaluator | - | Parallel | | Setup | all specialized | evaluator | test-runner | Parallel |

Summary

Audit Coordinator Benefits:

1. Automatic auditor selection - Chooses right auditors for target
2. Parallel execution - Runs independent audits concurrently
3. Unified reporting - Compiles findings from multiple sources
4. Priority reconciliation - Consolidates conflicting priorities
5. Comprehensive coverage - Ensures all relevant checks performed

Best for: Multi-component audits, setup-wide analysis, coordinated validation

For detailed orchestration patterns, see workflow-patterns.md. For report compilation guidance, see report-compilation.md.

Guidance Workflows

Beyond orchestrating audits, this skill provides guidance on Claude Code customization standards and best practices.

Pattern: Naming Guidance

User Query: "What should I name my new agent that reviews security?"

Workflow:

1. Identify component type (agent, skill, command, hook, output-style)
2. Reference naming conventions (see shared reference: naming-conventions.md)
3. Provide specific name suggestions with rationale
4. Offer examples of similar components
5. Explain naming pattern (e.g., {domain}-{role} for agents)

Output: Concrete name suggestions + pattern explanation

Pattern: Organization Guidance

User Query: "How should I organize my skill's reference files?"

Workflow:

1. Assess current structure
2. Reference file organization standards (see shared reference: file-organization.md)
3. Apply progressive disclosure principles
4. Recommend structure improvements
5. Provide migration guidance if restructuring needed

Output: Recommended directory structure + migration steps

Pattern: Pre-Deployment Validation

User Query: "Does this new skill look good?" (while editing SKILL.md)

Workflow:

1. Read target file
2. Invoke appropriate auditor (audit-skill, audit-agent, etc.)
3. Check against evaluation criteria (see references/evaluation-criteria.md)
4. Flag blocking issues (missing fields, invalid YAML, critical violations)
5. Provide quick validation summary

Output: Go/No-Go decision + critical fixes needed

Pattern: Troubleshooting Guidance

User Query: "Why isn't my skill being discovered?"

Workflow:

1. Identify symptom (not discovered, not triggering, errors, etc.)
2. Reference common issues guide (see references/common-issues.md)
3. Reference anti-patterns (see references/anti-patterns.md)
4. Provide specific diagnosis and fix
5. Offer to run diagnostic audit if needed

Output: Diagnosis + specific fix + optional follow-up audit

Pattern: Best Practices Consultation

User Query: "What are best practices for agents?"

Workflow:

1. Identify component type
2. Reference evaluation criteria (see references/evaluation-criteria.md)
3. Provide component-specific best practices
4. Give concrete examples of good patterns
5. Point to anti-patterns to avoid

Output: Best practices summary + examples + anti-patterns