audit-case-rag

What This Skill Does

The audit-case-rag skill provides a specialized, local-first RAG (Retrieval-Augmented Generation) engine specifically designed for commercial real estate auditing and forensic investigations. Unlike general-purpose document search tools, this skill uses a structural, event-driven folder architecture that maps directly to audit workflows—such as policy basis, procurement processes, contractual agreements, and settlement payments. By indexing documents within these stages, the agent can perform deep context-aware queries and provide verifiable, page-level citations using file:// links that open directly in your local PDF viewer or browser.

Installation

1. Ensure you have Python 3 installed on your machine.
2. Install the skill using the ClawKit CLI: clawhub install openclaw/skills/skills/jack4world/audit-case-rag.
3. Navigate to the skill directory and set up your virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate
   pip install -r scripts/requirements.txt
   

4. For full document support including Word/Excel conversion to PDF, ensure LibreOffice (soffice) is installed and accessible in your system PATH.

Use Cases

• Contractual Compliance Audits: Quickly cross-reference payment requests against signed contract terms and supplemental agreements.
• Procurement Investigation: Search for discrepancies in bidding processes by filtering queries by the 02_process stage.
• Payment Verification: Validate if settlement amounts align with project milestones and site acceptance reports.
• Forensic Evidence Collection: Rapidly synthesize data from emails, interview transcripts, and meeting minutes to identify timeline inconsistencies during an audit.

Example Prompts

1. "Analyze the project 2023-A01 in the payment stage and list all instances where invoice amounts exceeded the milestone budget."
2. "Search the interview transcripts in stage 06_interviews for any mentions of unauthorized vendor approvals regarding the landscaping contract."
3. "Does the current settlement in the 04_settlement_payment folder align with the original tender document in 02_process? Provide page-level citations."

Tips & Limitations

• Structure Matters: The accuracy of the retrieval depends heavily on using the predefined stage folders (01_policy_basis, 02_process, etc.). Always name your root folders correctly using the 项目问题编号__标题 format.
• Performance: For large audit cases, index the folder once and reuse the generated .joblib file. Avoid re-indexing unless documents are added or modified.
• Privacy: This is a strictly local tool. No data is sent to external cloud APIs, making it ideal for highly sensitive financial and internal audit documents.
• Hybrid Search: The system uses a combination of vector embeddings and TF-IDF reranking; for keyword-heavy queries, consider being specific about project terminology.