AI Legal Contract Review

Problem

Chinese contract review often gets stuck in scattered risks, weak citations, and missing follow-through. Contracts arrive as PDF, Word, scanned images, or plain text; risk levels, edit suggestions, and legal bases are easy to mix together; historical contracts, payment milestones, and recurring counterparty risks are hard to track. This skill turns the review workflow into repeatable command-line steps instead of manual clause hunting.

How It Works

The pipeline is: text extraction -> structure parsing -> rule engine + LLM semantic review -> report generation. The rule engine handles deterministic checks; the LLM handles semantic risk; use --scope, --role, and --industry to switch between full/key/quick review, Party A/Party B perspective, and industry packs. Outputs include risk grading, edit suggestions, compliance basis, and structured reports such as report.md.

• Clause library and revision draft: matches 275 standard clauses; --revise can generate a redline draft with red strikethrough, green replacement text, and gray comments; falls back to Markdown comparison if python-docx is unavailable.
• Legal and case support: --legal-base links Civil Code contract clauses, Company Law, and judicial interpretations; --guiding-cases and --case-law provide Supreme People's Court guiding-case points and tendency summaries.
• Ledger and search: reviewed contracts can be stored in a SQLite ledger with contract number, counterparty, amount, payment milestones, and expiry date; --archive-search supports inverted-index search, BM25 recall, and multi-dimension filters.
• Multi-contract analysis: --risk-trend aggregates risk-frequency trends, high-risk clause ratios, and recurring counterparty risks; --negotiate compares multiple versions to flag must-defend and negotiable terms; --align supports Chinese-English, Chinese-Japanese, and Chinese-Korean paragraph alignment and version mismatch detection.

Boundaries

This is useful for first-pass review, clause comparison, review drafts, and milestone tracking, but it does not replace licensed legal advice. OCR on scanned documents may misread amounts, dates, or entity names; cross-border contracts involving multiple jurisdictions should be reviewed with local counsel; using --text or external LLM APIs may send sensitive contract text outside the local environment.