contract-reviewer

Overview

Contract review is pattern recognition over dense text: find the clauses that matter (liability, termination, IP, payment), flag deviations from your standard positions, and compare versions. This skill covers a systematic review workflow — what to extract, what to flag, and how to stay in the proper lane between assistance and legal advice.

When to use

• Reviewing vendor agreements, MSAs, NDAs, or employment contracts
• Extracting key terms (dates, amounts, obligations) into structured summaries
• Comparing contract versions (redlines) clause by clause
• Building a clause library / standard positions for your organization
• Triaging a stack of contracts by risk level

Core concepts

Review against a playbook, not from scratch. Define your standard positions once (acceptable liability caps, payment terms, termination rights, governing law preferences) — then every review is "compare to playbook, flag deviations". Without a playbook, reviews are inconsistent and slow.

The high-risk clause checklist. Liability/limitation of liability, indemnification, termination (for convenience? notice period?), IP ownership and licensing, confidentiality scope and duration, payment terms and late fees, warranties and disclaimers, governing law and dispute resolution, auto-renewal, assignment/change-of-control, data protection/security obligations. Every contract gets checked against this list — missing clauses are as important as bad ones.

Extract, then analyze. First pass: pull structured facts (parties, effective date, term, renewal, key amounts, notice addresses). Second pass: clause-by-clause risk assessment. Structured extraction makes the analysis comparable across contracts and auditable later.

Redlines need three-way awareness. Comparing v1 → v2 shows what changed; but you also need v2 → playbook (does the change fix or worsen the deviation?). Track both deltas, and never review a redline without the clean current draft for context.

Assistance, not advice. Automated review accelerates issue-spotting; it doesn't replace legal judgment. Flag uncertainty explicitly, route high-risk deviations to counsel, and never present machine output as a legal conclusion.

Practical workflow

1. Normalize the input: convert to clean text/Markdown preserving clause numbering and structure (see pdf-to-markdown, docx-pro); confirm nothing was lost in conversion.
2. Extract key terms into a structured summary: parties, dates, term & renewal, financials, governing law, notice provisions.
3. Walk the clause checklist section by section; for each, record: present/ absent, matches playbook / deviates (how), risk level.
4. Flag the critical few: unlimited liability, one-sided indemnification, IP assignment of your work product, auto-renewal without notice, termination without cause denied to you — these get escalated, not just noted.
5. For redlines: produce a clause-level diff summary (what changed, in whose favor, playbook impact) — not just a raw diff dump.
6. Output a review memo: summary of terms, issues by severity with clause references, suggested fallback language from the playbook, and open questions for counsel.

Common pitfalls

• Reviewing without the playbook — every review becomes a fresh opinion; build the standard positions first.
• Missing what's absent — no limitation of liability, no termination clause, no data processing terms: absence is a finding, not a neutral.
• Trusting conversion blindly — clause numbers shifted by bad PDF extraction make references wrong; verify structure before analysis.
• Over-flagging trivia — a 40-item "issues" list where 35 are stylistic trains people to ignore the review; severity-rank ruthlessly.
• Presenting analysis as legal advice — label outputs as review assistance, keep counsel in the loop for decisions, know your jurisdiction's rules about unauthorized practice of law, and never present machine output as a legal conclusion.
• No version discipline — reviewing "the contract" without confirming which draft/version; always anchor to a specific document version.