Nothing worked as a drop-in skill. So I built one grounded in the CUAD dataset (41 legal risk categories from 510 real contracts), tested it against actual agreements, and iterated until the output was useful for real negotiations.

The result: position-aware review with market benchmarks, document-type checklists, and actual redline language — not just a list of issues.

What It Does

Analyzes legal contracts and outputs:

• Risk assessment with severity ratings (🔴 Critical / 🟡 Important / 🟢 Acceptable)
• Red flags quick scan — instant danger sign detection
• Key terms table with section references
• Market standard benchmarks — how terms compare to industry norms
• Negotiability ratings — what's realistic to change given power dynamics
• Specific redlines — actual replacement language, not just "negotiate this"
• Missing provisions with suggested language to add
• Internal consistency checks (broken cross-references, undefined terms)

Generate Deliverables with legal-redline-tools

This skill outputs structured JSON redlines. To produce the tracked-changes Word docs and redline PDFs that lawyers actually send, pair with legal-redline-tools:

pip install git+https://github.com/evolsb/legal-redline-tools.git

# After the skill generates redlines.json:
legal-redline apply contract.docx redlined.docx \
    --from-json redlines.json \
    --pdf redline.pdf \
    --memo-pdf internal-memo.pdf

Features

Position-Aware Review

Tell it which party you are (customer, vendor, buyer, seller, receiving party) — the skill adjusts what it flags as risky.

Document Type Checklists

Specialized checklists for each contract type:

• NDA — confidentiality term, non-solicitation, standstill, destruction certification
• SaaS/MSA — SLA, data export, suspension rights, price caps
• Payment/Merchant — reserves, chargebacks, network rules, auto-debit
• M&A — earnouts, escrow, rep survival, sandbagging
• Finder/Broker — fee tails, covered buyer definitions, joint representation

Market Standard Benchmarks

Compares terms to industry norms with clear thresholds:

| Provision | Standard | Yellow | Red | |-----------|----------|--------|-----| | Liability cap | 12 months | 6-11 mo | <6 mo | | Auto-renewal notice | 90+ days | 60-89 | <60 | | Non-compete | 1-2 years | 3-4 years | 5+ | | Rep survival (M&A) | 12-18 mo | 24-30 mo | 36+ mo |

Negotiability Ratings

Tells you what's actually changeable:

• High — Mutual termination, cure periods, data export
• Medium — Liability cap increases, price caps
• Low — Network rules, regulatory requirements

Red Flags Quick Scan

Instant detection of danger signs:

• Liability cap < 6 months
• Uncapped indemnification
• Unilateral amendment rights
• Perpetual obligations
• Offshore jurisdiction (BVI, Cayman)

Jurisdiction Awareness

Flags when governing law affects enforceability:

• Non-competes void in CA/ND/OK/MN
• Delaware vs NY vs CA implications
• Offshore jurisdiction cost/enforcement concerns

M&A Support

Special handling for acquisition agreements:

• Earnout mechanics and measurement
• Rep & warranty survival periods
• Working capital adjustments
• Escrow/holdback provisions
• Employment comp in deal value calculations

Installation

This skill follows the open Agent Skills standard and works with any compatible tool.

# Claude Code
git clone https://github.com/evolsb/claude-legal-skill ~/.claude/skills/contract-review

# OpenAI Codex
git clone https://github.com/evolsb/claude-legal-skill ~/.codex/skills/contract-review

# Cursor, Copilot, Gemini CLI, etc.
# Clone to your tool's skills directory

Development (Symlink)

git clone https://github.com/evolsb/claude-legal-skill ~/Developer/claude-legal-skill
ln -s ~/Developer/claude-legal-skill ~/.claude/skills/contract-review

Usage Examples

Review this NDA for red flags - I'm the receiving party

Analyze the indemnification in this MSA - I'm the vendor

What are the termination provisions? I'm the customer.

Review this acquisition agreement - I'm the seller

Check this merchant agreement - what's my chargeback exposure?

See examples/ for full sample outputs.

Limitations

• Not legal advice — always have material terms reviewed by qualified counsel
• US law focus — analysis defaults to US; provisions vary by jurisdiction
• Context window — very long contracts may need section-by-section review

Accuracy

Based on ContractEval benchmarks, Claude achieves F1 ~0.62 on clause extraction. Best for first-pass review and issue flagging — not a replacement for attorney review on material deals.

Credits

• CUAD Dataset — Atticus Project (NeurIPS 2021)
• LegalBench — Stanford HAI
• ContractEval — Contract understanding benchmarks

Next Steps

• Need deliverables? Use legal-redline-tools to generate tracked-changes .docx, redline PDFs, and negotiation memos from the skill's output
• Want examples? See examples/ for full sample reviews
• Found an issue? Open a GitHub issue

Contact

Questions or feedback? Open an issue or email chris@ctsheehan.com.

License

MIT — see LICENSE