Legal Billing Auditor

Step 2: Check Dependencies

python3 "$SKILL_DIR/scripts/check_dependencies.py"

• Exit 0: all good. Exit 1: packages were installed (proceed). Exit 2: failed (report to user).

Step 3: Run Billing Audit

Determine the output directory:

• Single file: OUTPUT_DIR="{parent_dir}/{filename_without_ext}_audit"
• Directory: OUTPUT_DIR="{directory_path}/_audit_results"

mkdir -p "$OUTPUT_DIR"

python3 "$SKILL_DIR/scripts/audit_billing.py" \
  --input "<file_or_directory_path>" \
  --output-dir "$OUTPUT_DIR" \
  [--rate-caps "<rate_caps.json>"] \
  [--max-daily-hours 10]

The script prints a JSON summary to stdout. Capture and parse it.

Step 4: Present Audit Summary

Read the output files and present findings to the user:

1. Start with the summary: Read $OUTPUT_DIR/audit_summary.txt and present the key metrics:

   • Total entries reviewed
   • Total spend analyzed
   • Number of flags by severity (HIGH / MEDIUM / LOW)
   • Estimated potential savings

2. Highlight HIGH severity flags: These require immediate attention. For each HIGH flag, explain:

   • What was flagged and why
   • The specific billing entry details
   • Recommended action

3. Summarize MEDIUM and LOW flags: Group by rule type and give counts with examples.

Step 5: Direct to Detailed Outputs

Tell the user about the generated files:

• flagged_entries.xlsx - Full spreadsheet of all flagged entries for review
• spend_dashboard.html - Interactive charts showing spend patterns (open in browser)
• audit_report.json - Structured data for further processing

Step 6: Offer Formal Report

Ask: "Would you like me to generate a formal audit report as a Word document (.docx)?"

If yes, use the npm docx package to generate a professional audit report containing:

1. Title page: "Billing Audit Report", date, matter information
2. Executive Summary: Key findings and estimated savings
3. Methodology: Audit rules applied
4. Findings by Severity: HIGH, MEDIUM, LOW sections with tables
5. Spend Analysis: Key metrics and trends
6. Recommendations: Action items based on findings
7. Appendix: Full list of flagged entries

Accuracy and QA (Required)

Anti-hallucination rules (include in ALL subagent prompts):

• Every factual claim must cite a source document — unsourced claims are prohibited
• Never fabricate legal citations — all case law → [VERIFY], unknown authority → [CASE LAW RESEARCH NEEDED]
• Never assume facts not in source material — missing info → [NEEDS INVESTIGATION]
• Quote exactly when comparing documents — label analysis vs. facts distinctly

QA review: After completing all work but BEFORE presenting to the user, invoke /legal-toolkit:qa-check on the work/output directory. Do not skip this step.

Error Handling

• Path not found: Ask user to verify the path
• Unsupported format: Supported types are .txt, .ledes, .xlsx, .csv
• Empty directory: No supported files found -- ask user to check the folder
• Parse errors: Some files may have non-standard formats; report which files failed and continue with others
• No flags found: Report clean audit -- no issues detected
• Script not found: Verify the skill is installed (ls $SKILL_DIR/scripts/)