AI Legal Assistant: Full Contract Review
One command launches five parallel AI agents to review a contract, producing a scored report with clause-by-clause risk analysis and negotiation priorities in under 60 seconds.
The skill performs no external write operations, includes a legal disclaimer in the report, and has a clear stop path on unreadable contracts; however, there is no user confirmation for fetching contracts via WebFetch or shipping sensitive contract text to 5 subagent prompts, incomplete data-flow disclosure, no rollback/data-retention notes, and unverified publisher identity — mid score only.
Instructions are self-consistent with good failure feedback on unreadable input (report error, request alternative, do not proceed); but the weighted-score formula (how agent weights become a 0-100 score) is undefined, subagent output formats depend on agents/*.md not shown here, and there are no tests or edge cases — deducted below the static ceiling of 10.
Scenario is clear (9-type contract classification table, explicit /legal review trigger), but capability boundaries and non-fit ranges (jurisdictional limits, unsuitable contract types) are undeclared; English-only, and core function depends entirely on Claude Code/Anthropic API, which is unreachable from mainland-China networks — deducted.
Clean structure, sensible progressive orchestration, concrete report template; but license metadata unknown, no versioning/changelog, no FAQ or known-limitations disclosure, no clear maintenance ownership or update path, and a curl|bash installer without matching dependency documentation.
Output design (scored CONTRACT-REVIEW.md with clause-level analysis) is complete and directly usable in principle, with plausible marginal value; but no real output samples or verification exist, 'review in 60 seconds' marketing is unsubstantiated, and legal-analysis quality is model-dependent — low band under static review.
Only author claims and unsourced README marketing figures (82%, $10,000+, etc.); no CI, no tests, no third-party execution evidence; the weighted scoring method cannot be independently recomputed; facts and inference are not separated — below the static ceiling.
- Full contract text is sent to 5 parallel subagents and may be fetched via WebFetch from URLs — contracts containing sensitive commercial terms face a data-egress surface; assess before use.
- The weighted formula behind the Contract Safety Score is undefined; the score should not drive signing decisions.
- Output is AI-generated reference analysis only; the report itself carries a disclaimer and a licensed attorney must review before signing.
- License is unknown; confirm authorization before enterprise procurement.
- Core function depends on Claude Code/Anthropic API, likely unreachable from mainland-China networks; English-only.
- Multiple README statistics and cost figures are unsourced marketing claims.
What does this skill do, and when should you use it?
This is the flagship skill in the ai-legal-claude repository, powering the `/legal review <file>` command. It ingests a contract from a file, pasted text, or URL, classifies its type and extracts metadata, then launches five parallel subagents covering clause analysis, risk assessment, compliance, obligations mapping, and recommendations. Results are aggregated with 20/25/20/15/20 weights into a 0–100 Contract Safety Score and a structured CONTRACT-REVIEW Markdown report. The skill is pure prompt orchestration — it contains no legal database and provides no legal advice.
Reads contracts via the Read tool, accepts pasted text, or fetches URLs via WebFetch; classifies the contract into one of nine types (service, employment, NDA, SaaS, freelancer, partnership, lease, sales, investment/SAFE); extracts parties, dates, term, governing law, and value; runs five subagents (legal-clauses, legal-risks, legal-compliance, legal-terms, legal-recommendations) in parallel via the Agent tool; computes a weighted safety score graded A+ through F; and produces a Markdown report with a risk dashboard, clause-by-clause analysis, missing protections, obligations and deadlines table, compliance flags, ranked negotiation priorities, and a next-steps checklist — then points the user to /legal negotiate and /legal report-pdf.
- A freelancer reviewing a client contract before signing, wanting to spot non-competes or IP-assignment traps
- A small business owner checking vendor or SaaS subscription agreements for auto-renewal and liability-cap clauses
- An AI automation agency offering contract first-review as a paid service ($500–$1,500 per review)
- A founder reviewing a SAFE or investment agreement for dilution and liquidation-preference issues
- Anyone preparing for negotiations who needs a ranked list of changes with specific replacement language
What are this skill's strengths and limitations?
- Five parallel agents cover clauses, risk, compliance, obligations, and recommendations for comprehensive output
- Weighted safety score with A+ to F grading quantifies overall risk
- Produces ranked negotiation priorities with concrete suggested language
- Auto-detects nine contract types and calibrates analysis accordingly
- Report embeds a legal disclaimer, keeping the positioning honest
- Deeply relies on Claude Code's Agent tool and parallel subagent orchestration, so it does not port as-is to other platforms
- Educational only — not legal advice; a licensed attorney must review before signing
- README cost-saving figures (e.g., $300–$500/hour attorney rates) are marketing stats with no cited sources
- License is unknown; confirm before commercial use
- Score weights are hardcoded heuristics with no test suite or benchmark validating accuracy
How do you install this skill?
The repository ships a one-line installer that installs all 14 skills, 5 agents, and the PDF scripts: curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/ai-legal-claude/main/install.sh | bash. This specific skill lives at skills/legal-review/SKILL.md. How the skill is registered as a slash command after install is not documented.
How do you use this skill?
In Claude Code, run /legal review <file path>, or paste the contract text directly, or supply a URL. After the review, run /legal negotiate for replacement language for risky clauses, or /legal report-pdf for a PDF version. PDF generation requires Python 3.8+ and reportlab.
How does this skill compare with similar options?
Sibling skills in the same repo — /legal risks (deep risk analysis only) and /legal freelancer (freelancer-perspective review) — are lighter alternatives, but /legal review is the only command aggregating all five agents into a single scored report.