Skills for Audit

A comprehensive learning path for applying Anthropic's Claude Skills to financial audit workflows, designed for accounting professors, audit regulators, and practitioners.

🎓 Educational Purpose: This repository provides teaching materials and proof-of-concept demonstrations. It is not intended to replace professional judgment in actual audit engagements.

⭐ Quick Links

• [Quick Start Guide](QUICK_START.md) - Get running in 15 minutes
• [Learning Roadmap](AUDIT_SKILLS_ROADMAP.md) - 10-week curriculum
• [Contributing](CONTRIBUTING.md) - How to contribute
• [Discussions](https://github.com/GenAICPA/SkillsforAudit/discussions) - Ask questions and share ideas

Project Goals

Build progressive understanding of Claude Skills by creating an end-to-end audit assistant that:

• Manages audit checklists with structured data (JSON)
• Generates professional audit workpapers (Excel, PowerPoint, PDF)
• Parses and validates Inline XBRL financial statements
• Cross-references audit procedures with XBRL data points
• Demonstrates AI-assisted audit workflows for educational and regulatory audiences

Architecture Overview

graph TD
    A[Inline XBRL Financial Statement] --> B[Phase 3: Parse & Validate]
    C[Audit Checklist JSON] --> D[Phase 2: Structure & Validate]
    B --> E[Phase 4: Cross-Reference Engine]
    D --> E
    E --> F[Coverage Analysis]
    F --> G[Phase 1: Generate Reports]
    G --> H[Excel Workpapers]
    G --> I[PowerPoint Presentations]
    G --> J[PDF Audit Reports]

    style A fill:#e1f5ff
    style C fill:#e1f5ff
    style E fill:#fff4e1
    style G fill:#e8f5e9

Repository Structure

skills/
├── phase1_foundations/          # Basic Skills API usage
│   ├── 01_basic_pptx.py        # PowerPoint generation
│   ├── 02_basic_xlsx.py        # Excel workpaper templates
│   └── 03_code_execution_pattern.py  # Code + Skills integration
│
├── phase2_structured_data/      # JSON/XML audit data models
│   ├── 04_audit_checklist_model.py   # Define data structures
│   ├── 05_json_validation.py         # Validate audit data
│   └── 06_checklist_to_excel.py      # Generate workpapers from data
│
├── phase3_xbrl/                 # XBRL parsing and validation
│   ├── 07_parse_ixbrl.py       # Extract XBRL data
│   ├── 08_validate_ixbrl.py    # Validation rules
│   └── 09_ixbrl_analytics.py   # Analytical procedures
│
├── phase4_cross_reference/      # Link procedures to data
│   ├── 10_cross_reference_engine.py  # Core linking logic
│   ├── 11_coverage_analysis.py       # Audit gap identification
│   └── 12_traceability_report.py     # Audit trail documentation
│
├── phase5_prototype/            # Integrated demonstration
│   └── prototype_audit_assistant.py  # Complete workflow
│
├── sample_data/                 # Example audit files
│   ├── sample_ixbrl/           # Inline XBRL filings
│   ├── audit_checklists/       # JSON audit programs
│   └── evidence/               # Sample audit evidence
│
└── AUDIT_SKILLS_ROADMAP.md     # Detailed learning plan

Quick Start

Prerequisites

# Install required packages
pip install anthropic

# Set up your API key
export ANTHROPIC_API_KEY='your-key-here'  # Linux/Mac
# or
set ANTHROPIC_API_KEY=your-key-here       # Windows

Phase 1: Basic Skills (Start Here)

cd phase1_foundations

# Generate an audit presentation
python 01_basic_pptx.py

# Create an Excel bank reconciliation workpaper
python 02_basic_xlsx.py

# Use code execution with Skills for sampling calculations
python 03_code_execution_pattern.py

Phase 2: Structured Data

cd phase2_structured_data

# Create sample audit checklist (JSON)
python 04_audit_checklist_model.py

# Validate the checklist structure
python 05_json_validation.py

# Generate Excel workpaper from checklist
python 06_checklist_to_excel.py

Phase 3-5: Advanced Topics (Coming Soon)

XBRL integration, cross-referencing, and complete prototype demonstrations.

Key Concepts

Audit Checklist Data Model

Structured JSON representing:

• Audit procedures: What needs to be tested
• Status tracking: not_started, in_progress, completed
• Evidence references: Links to supporting documentation
• XBRL references: Which financial statement elements are tested
• Sign-offs: Auditor, date, notes, exceptions

Example structure:

{
  "audit_engagement": {
    "client": "Example Corp",
    "period_end": "2024-12-31",
    "sections": [
      {
        "section_id": "AS.2201",
        "name": "Revenue Recognition",
        "risk_level": "high",
        "procedures": [
          {
            "procedure_id": "REV-001",
            "description": "Test sales cutoff",
            "status": "completed",
            "xbrl_refs": ["us-gaap:RevenueFromContractWithCustomerExcludingAssessedTax"],
            "evidence_refs": ["WP-A1"],
            "sign_off": {
              "auditor": "Emily Rodriguez, CPA",
              "date": "2025-02-15"
            }
          }
        ]
      }
    ]
  }
}

Skills Integration Pattern

All examples follow this pattern:

import anthropic

client = anthropic.Anthropic()

response = client.beta.messages.create(
    model="claude-haiku-4-5-20251001",
    max_tokens=4096,
    betas=["code-execution-2025-08-25", "skills-2025-10-02", "files-api-2025-04-14"],
    container={
        "skills": [
            {
                "type": "anthropic",
                "skill_id": "xlsx",  # or "pptx", "pdf"
                "version": "latest"
            }
        ]
    },
    tools=[
        {
            "type": "code_execution_20250825",
            "name": "code_execution"
        }
    ],
    messages=[{"role": "user", "content": "Your prompt here"}]
)

# Extract file_id and download
for block in response.content:
    if hasattr(block, 'file_id') and block.file_id:
        file_content = client.beta.files.retrieve_content(block.file_id)
        with open('output.xlsx', 'wb') as f:
            f.write(file_content)

Learning Path

For Students/Professors

1. Week 1-2: Master basic Skills API (Phase 1)

• Understand document generation workflows
• Learn file handling patterns
• Practice with audit-specific examples

1. Week 3-4: Work with structured audit data (Phase 2)

• Define JSON schemas for audit programs
• Validate data quality
• Automate workpaper generation

1. Week 5-6: XBRL financial statement analysis (Phase 3)

• Parse Inline XBRL documents
• Extract and validate financial data
• Perform analytical procedures

1. Week 7-8: Build cross-reference systems (Phase 4)

• Link procedures to evidence to XBRL facts
• Analyze audit coverage
• Generate traceability reports

1. Week 9-10: Integrated demonstration (Phase 5)

• Complete end-to-end audit workflow
• Present to stakeholders
• Discuss regulatory implications

For Regulators

Focus areas:

• Audit trail transparency: How AI-assisted procedures are documented
• Data validation: Quality controls and validation logic
• Compliance: Alignment with PCAOB/AICPA standards
• Explainability: Understanding AI decision-making in audit context

See regulatory_documentation.md (forthcoming) for detailed compliance mapping.

Use Cases

1. Audit Program Management

• Generate customizable audit programs from templates
• Track procedure completion in real-time
• Automate status reporting

2. XBRL Financial Analysis

• Extract data from SEC filings automatically
• Validate mathematical relationships
• Identify unusual trends or outliers

3. Audit Documentation

• Generate workpapers from structured data
• Create presentation materials for audit committees
• Produce final audit reports (PDF)

4. Cross-Reference & Coverage

• Map which XBRL elements are tested by procedures
• Identify untested financial statement items
• Ensure comprehensive audit coverage

🚀 Development Status

🤝 Contributing

We welcome contributions from educators, practitioners, and developers! See CONTRIBUTING.md for guidelines.

Ways to contribute:

• 📚 Share classroom materials or teaching experiences
• 🔍 Add audit procedure examples (anonymized)
• 💻 Implement Phase 4-5 features
• 📖 Improve documentation
• 🐛 Report bugs or suggest enhancements

📚 Resources

Anthropic & AI

• Anthropic Skills API Documentation
• Claude Code

Audit Standards

• PCAOB Auditing Standards
• AICPA Audit Standards
• IAASB International Standards on Auditing

XBRL & Financial Reporting

• XBRL US GAAP Taxonomy
• SEC EDGAR Database
• XBRL International

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Educational Disclaimer: This software is for educational and research purposes only. Not intended to replace professional judgment in audit engagements. Users must comply with applicable professional standards.

📧 Contact & Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Questions: Open a discussion or review existing documentation

🌟 Acknowledgments

Built with Claude by GenAICPA for the advancement of audit education and practice.

📖 Citation

If you use this project in academic work, please cite:

@software{skills_for_audit_2025,
  author = {GenAICPA},
  title = {Skills for Audit: AI-Assisted Financial Audit Workflows},
  year = {2025},
  url = {https://github.com/GenAICPA/SkillsforAudit},
  note = {Educational toolkit for Claude Skills in audit applications}
}

⚠️ Important: Never commit real client data or API keys. See CONTRIBUTING.md for data privacy guidelines.